Analyzing the Evolving Landscape of Schedule 13D/13G Filings

2025年4月2日 Jerry NgLeave a comment

Deep Dives, Implications, and a Case Study

I. Introduction

Recent changes to the SEC’s beneficial ownership reporting rules—and the release of new Compliance and Disclosure Interpretations (C&DIs)—have rapidly shifted the ground under both companies and investors. In my discussions with governance professionals, one point resonates clearly: passive managers (often large index funds) face some of the biggest adjustments.

Historically, passive funds have been comfortable filing on Schedule 13G, reflecting a passive investment approach rather than an activist agenda. Now, shorter deadlines, heightened scrutiny, and broader definitions of “influencing control” compel these managers to think twice about routine engagements on governance matters (like declassifying boards or altering executive pay structures). As a student of corporate governance, I find it fascinating how these shifts are forcing advisors to reshape engagement protocols—especially when coordinating with passive shareholders who do not want to be seen as activists.

II. Schedule 13D vs. 13G: A Quick Refresher

Schedule 13D
- Trigger: More than 5% beneficial ownership and the purpose or effect of changing or influencing control.
- Requirements: Detailed disclosures on acquisition purpose, plans, and any governance or control-related proposals.
- Deadline: Must file within five business days after crossing the ownership threshold (under recent amendments).
Schedule 13G
- Trigger: Exceeding 5% ownership but remaining passive, or qualifying as a specific category (e.g., Qualified Institutional Investor).
- Requirements: Streamlined, high-level disclosures about share ownership.
- Deadline: Varies by filer type (some must file within five days, others have slightly longer windows), but all have been accelerated relative to past rules.

Key Update: Shorter filing windows and a broader interpretation of what constitutes “activism” raise the stakes for investors who wish to remain passive, especially those that begin accumulating shares below 5% and soon cross that threshold.

III. The 2025 “Passive Investor” Guidance and Its Impact

In February 2025, the SEC staff clarified through C&DIs (Questions 103.11 and 103.12) that:

Passive Investors Cannot Appear to “Exert Pressure”
- Merely voicing preferences on ESG or governance typically does not jeopardize a 13G filing.
- However, pressuring management—e.g., “We won’t support board nominees unless you declassify the board”—is now firmly in “activist” territory.
Passive Funds & Fear of Activist Label
- Index funds and other large passive investors often hold positions across hundreds or thousands of companies.
- Many are pausing certain engagements or “toning down” their governance demands to avoid being mislabeled as activists (and forced onto Schedule 13D).
Information and Timing Constraints
- Passive managers typically rely on universal voting guidelines, not company-specific “change in control” agendas.
- Tighter 13G deadlines create significant administrative burdens, with some funds crossing thresholds unexpectedly if a stock’s price shifts or if inflows change portfolio allocations. This dynamic intensifies the risk of missing a filing deadline or inadvertently acting like an activist.

IV. Why Passive Managers Are Uniquely Affected

“Materially Changing the Operation” vs. “Passive Approach”
- Index funds are designed to mirror benchmarks, not drive corporate operations. They historically engage on high-level governance topics—like board declassification—without seeking direct control.
- The new rules blur the line between “voicing best-practice expectations” and “influencing control,” making even routine feedback risky if delivered too forcefully.
High Volume of Holdings
- Passive giants can easily exceed 5% in numerous portfolio companies, often beginning under that threshold but creeping above through normal market movements or index weighting changes.
- This means multiple 13G filings can be triggered quickly—each bringing compliance scrutiny.
Desire to Avoid 13D
- Shifting from 13G to 13D is a significant escalation in disclosure obligations.
- Passive managers typically want to maintain a neutral stance, so being viewed as an “activist” can damage their relationships with issuers and complicate compliance processes.

V. Implications for Advisors Communicating with Passive Shareholders

Careful Engagement Language
- Why It Matters: Even an unintentional hint—such as “We may not support your entire slate if you ignore X governance demand”—can suggest an attempt to influence control.
- Advisor Strategy:
  - Emphasize broad principles rather than mandates.
  - Frame feedback as “preferences” or “views” rather than “requirements.”
  - Remind passive funds that the new rules make subtle threats (even implied ones) risky.
Scenario-Based Planning
- Why It Matters: Advisors need to anticipate how an index fund’s routine “best practice” communications could be construed as pressuring management.
- Advisor Strategy:
  - Develop scripts/examples showing the difference between “We believe declassifying boards improves governance” and “Declassify your board or lose our support.”
  - Train client teams to keep meeting notes, ensuring language used doesn’t border on activism.
Pause or Reshape Certain Engagements
- Why It Matters: Some passive managers may simply pause deeper dialogues to avoid activism charges, creating an informational vacuum for boards that rely on their feedback.
- Advisor Strategy:
  - Counsel passive investors on how to continue “safer” forms of engagement—like participating in earnings calls or broad stakeholder roundtables—without making company-specific demands.
  - Encourage them to clarify the universal nature of their governance guidelines, so they’re not seen as targeting any single issuer to effect control.
Early, Real-Time Filing Prep
- Why It Matters: Many passive funds approach or pass the 5% threshold unexpectedly as markets move.
- Advisor Strategy:
  - Establish robust tracking to catch ownership changes in near real-time, enabling timely 13G amendments.
  - Provide guidelines on communication protocols at each stage (e.g., under 5%, just over 5%, well over 5%) to minimize unintentional activism signals.
Coordination with Legal and Compliance
- Why It Matters: The line between “passive preference” and “activist demand” can be blurry.
- Advisor Strategy:
  - Work closely with securities counsel to evaluate if standard stewardship policies might be seen as pressuring.
  - Maintain compliance logs, disclaimers, or clarifying statements that reaffirm passive status in engagement settings (for instance, “We are not advocating a control change; we are merely sharing best practices we support across all portfolio companies.”)

VI. Hypothetical Case Study: XYZ Corporation & a Passive Index Fund

Situation:
- XYZ Corporation has a staggered board. A major index fund approaching 5% ownership typically champions “one-year” board terms across all its holdings.
- In routine engagements, the index fund mentions that it “encourages all companies to declassify boards.” Over time, it crosses 5% due to market cap changes.
Red Flag for the Index Fund:
- As soon as it files 13G, the fund worries that repeated statements like “We’ll strongly reconsider our votes for the nominating committee if you remain staggered” might cross into “influencing control.”
- It consults an advisor to revise its stewardship language.
Advisor’s Intervention:
- Engagement Reboot: The advisor helps the fund shift to more generalized language—emphasizing a portfolio-wide governance philosophy, not an ultimatum.
- Filing Vigilance: The advisor sets up real-time alerts for position changes.
- Outcome: The index fund remains a 13G filer, reaffirming its passive stance while still conveying governance preferences in a low-pressure manner. XYZ Corporation, meanwhile, continues to hear the fund’s views without feeling threatened by an activist campaign.

VII. Key Takeaways & Next Steps

Recognize Passive Funds’ Growing Sensitivity
- Advisors must acknowledge that large index funds are among the most impacted by the new 13D/13G rules.
- These funds often operate globally, have large volumes of holdings, and can’t afford to be labeled activist.
Craft Nuanced Engagement Strategies
- Boards: Avoid reading normal “best-practice” suggestions from passive holders as guaranteed activism, but stay alert to any shift in tone.
- Advisors: Emphasize language that clarifies broad, portfolio-wide governance philosophies, rather than perceived threats or ultimatums.
Refine Compliance Protocols
- Tracking share accumulations in real time is crucial for both companies and investors under accelerated filing deadlines.
- Advisors can create checklists or gating questions (e.g., “Is this statement or demand pushing beyond passive guidelines?”) to use before engaging with companies.
Stay in Sync with Legal Experts
- The definition of “control” is intentionally broad. Frequent counsel check-ins help ensure a passive investor’s approach doesn’t become inadvertently activist.
- Subtle shifts—like referencing upcoming director elections—can trigger closer scrutiny from regulators.
Preserve Constructive Dialogue
- Ultimately, good governance discussions can still thrive if approached thoughtfully. Board declassification or ESG improvement can be championed without threats.
- Advisors should help all parties maintain an open exchange of views—reinforcing that passivity doesn’t mean disengagement, just cautious communication.

VIII. Additional Considerations for Active Managers

While passive investors have garnered much attention due to their broad portfolios and passive strategies, active managers—such as T. Rowe Price, Fidelity, and other large, actively managed mutual funds—also face new complexities under the revised 13D/13G framework:

Frequent Portfolio Adjustments
- Active managers regularly buy and sell positions in response to market conditions, company fundamentals, or performance metrics.
- Crossing the 5% threshold unexpectedly may force them into shorter reporting windows, prompting the need for more robust real-time monitoring.
Selective Engagement & The Activist Perception
- Because active managers often engage more deeply on specific governance or strategic issues, they risk appearing as though they are “influencing control.”
- Key Concern: An active strategy that frequently involves pushing for management changes (e.g., capital allocation, board refreshment) may trigger scrutiny under the broader definitions of activism.
Balancing Engagement with Reporting Obligations
- Active managers might try to shape corporate policies to enhance long-term returns, yet remain wary of the higher disclosure and potential stigma of filing Schedule 13D.
- In practice, this could lead them to dial back certain discussions or to structure engagements in a way that emphasizes portfolio-wide philosophies rather than single-company ultimatums.
Advisors’ Role in Guiding Active Strategies
- For advisors working with T. Rowe Price–style investors, a key task is clarifying the threshold for “influencing control.”
- This might include training portfolio managers on how to approach board discussions, ensuring language remains consistent with a “long-term investment thesis” rather than a push for direct control measures.
Potential for Collaboration with Management
- Despite heightened caution, many active managers continue to serve as constructive partners, offering insights on operational or strategic improvements.
- If approached carefully—framed as collaborative rather than confrontational—such input can still avoid tipping into 13D territory.

By recognizing these nuances, active managers can continue pursuing alpha-driven engagements without triggering activist classifications. Advisors play a pivotal role in striking the right balance: effective communication that advances shareholder value while staying firmly within the passive or “non-control-influencing” realm.

IX. Conclusion

The SEC’s updated 13D/13G framework places passive funds—particularly large index investors—under the microscope like never before. Yet active managers also face critical decisions about how, when, and how forcefully to engage with portfolio companies. For shareholder advisors, this shifting landscape means revisiting counsel for both passive and active clients, ensuring any calls for governance or strategic changes are framed as advisory rather than prescriptive.

Meanwhile, companies must adapt by refining communication protocols and paying closer attention to smaller investors and proxy advisors—especially if big asset managers pare back direct demands to avoid activist scrutiny. Ultimately, open dialogue and meticulous compliance can preserve a productive relationship between investors, issuers, and the broader governance ecosystem—even in a world of tighter filing deadlines and heightened scrutiny.

Jerry’s 2025 Market Outlook: Trump 2.0, Secondaries, Fintech, Technology

2025年1月18日 Jerry NgLeave a comment

Jerry here! ;p

1. Executive Summary

Overview of 2024 and Key Themes for 2025
Heading into 2025, global markets stand at an inflection point shaped by moderating inflation, an unusual global monetary easing cycle, and a second Trump administration in the United States. The prior 12 months have shown how quickly market sentiment can pivot from tightening fears toward optimism over potential rate cuts and pro-business fiscal policies. That transition has rekindled dealmaking activity—particularly in private markets—while also exposing new geopolitical flashpoints that could disrupt commodity flows and trade relationships.

One of the headline stories this year has been the relative resilience of the U.S. economy, which ended 2024 with GDP growth around 2.1–2.2%, according to estimates by Goldman Sachs¹ and J.P. Morgan². At the same time, eurozone growth dipped below 1%, reflecting structural challenges and weaker consumer sentiment, while China’s economy registered around 4.0–4.4% expansion (IMF projections³). All of these developments have refocused the investment community on the interplay between fiscal stimulus, global trade (including potential new tariffs), and the divergent outlook for inflation in major economies.

Looking ahead, private equity and private credit managers foresee improved transaction flows in 2025 as the spread between buyers’ and sellers’ price expectations tightens and financing costs begin to ease from peak levels. Enthusiasm for AI-related investments continues to run high, fueling robust valuations in TMT, while industrials and old-economy companies see fresh demand for automation, onshoring/nearshoring, and supply-chain resilience. Below, we examine the macro and policy environment, the dynamics in rates and government bonds, equities, and then proceed to private markets and thorough analyses of secondaries, fintech, FIG, TMT, and industrials.

2. Macro & Policy Environment

2.1 Geopolitical Flashpoints & Commodity Price Implications

Conflicts in Ukraine and the Middle East
The ongoing conflict in Ukraine—which has, among other impacts, restricted Black Sea grain exports—and unrest in parts of the Middle East have significantly influenced global commodity markets, particularly for energy and agricultural products. For instance, data from S&P Global⁴ indicates that approximately 20–25% of Europe’s wheat imports originated through Black Sea routes before Russia’s invasion of Ukraine. When these supply lines are disrupted, it leads to commodity price spikes, stoking inflationary pressure. A similar phenomenon occurs with energy prices each time there is a threat to major oil or gas transit routes in the Middle East.

As we enter 2025, tensions in both regions remain unresolved, creating an overhang of risk for investors. Higher volatility in commodities could translate into unpredictable input costs for manufacturers, shipping lines, or utilities—resulting in profit margin uncertainty. For businesses in industrials (e.g., automotive, aerospace) or TMT (e.g., semiconductor supply chains requiring rare metals), sudden commodity price swings can jeopardize earnings forecasts. On the positive side, private equity can seize opportunities through carve-outs of non-core divisions from large conglomerates that seek to hedge commodity risk or restructure their supply chains. Investors with a strong macro-hedging strategy could position themselves in more stable-cost regions (like North America) or invest in companies that have diversified procurement channels.

2.2 Trump 2.0: Potential Fiscal Stimulus, Tariffs & Deregulation

Following Donald Trump’s re-election, many analysts, including Goldman Sachs⁵ and Rothschild & Co⁶, are carefully assessing how the administration’s pro-business stance might play out. In his prior term, Trump championed tax cuts (notably the Tax Cuts and Jobs Act of 2017), deregulation (particularly in financial services and environmental rules), and a confrontational approach to trade (tariffs on Chinese and other foreign goods).

In 2025, the potential areas of impact include:

Tax Breaks and Deregulation: Sectors like energy (both fossil fuels and renewables), industrial manufacturing, and financial institutions might see a rollback of environmental or consumer protection regulations that had been introduced under Biden. This relaxation could reduce compliance costs and accelerate expansions or capital projects that were previously on hold. However, it also raises reputational or ESG-related concerns among some investors and could trigger retaliatory legislation at the state or international levels.
Infrastructure and Defense Spending: Trump has signaled a willingness to boost fiscal outlays for highways, 5G infrastructure, AI R&D, and defense modernization. This injection of spending presents an opportunity for private equity firms investing in infrastructure—for instance, consortia that build toll roads or data centers—and for private debt providers who can finance these government-backed projects. According to a Morgan Stanley estimate⁷, each 1% boost in federal infrastructure spending could raise GDP growth by 0.3% over a 12-month horizon, although deficits might also swell if tariffs fail to cover the additional outlays.
Trade & Tariffs: Plans to impose or increase tariffs on Mexico, China, and possibly Europe introduce uncertainty into corporate supply chains. If enacted, tariffs could lead to higher input costs for U.S. manufacturers and higher consumer prices. Nonetheless, certain segments (e.g., domestic steel producers, electronics assembly plants in the U.S.) might flourish with less foreign competition, creating attractive targets for buyout or growth equity investments. Conversely, tariff wars are a risk factor: if other nations retaliate, American exporters of consumer goods, technology, or agricultural products could face new barriers, dampening earnings.

As a result, the opportunity for investors in 2025 might revolve around nearshoring or onshoring strategies (e.g., manufacturing in Mexico to serve the U.S. market more cost-effectively than Asia), or capturing concessions in sectors that Trump deems “critical,” such as semiconductors or defense. The risk is that abrupt policy changes or uncertain geopolitical relationships compress margins and valuations, especially if inflation resurfaces or the USD remains strong.

3. Rates & Government Bonds: Turning Point or Transitory Easing?

With the Federal Reserve and European Central Bank (ECB) expected to cut rates—some forecasts see a 100 basis point reduction in the Fed’s policy rate by end-2025—the direction of 10-year Treasuries and Bunds will likely hinge on two competing forces: residual inflation risk and subdued global growth. Rothschild & Co’s Consensus Outlook 2025⁸ places U.S. 10-year yields around 3.55–4.25%, and German Bunds in the 1.65–2.05% range, reflecting some skepticism about whether policy rates are truly at “restrictive” levels.

Implications of Sticky Inflation vs. Rate Cuts

If inflation were to pick up because of higher tariffs or commodity supply shocks, central banks might slow or pause their planned cuts, reintroducing rate volatility. This would directly affect the cost of leverage in private equity deals, with impact on sponsor negotiations, and yield curves might flatten if recession worries resurface.
On the flipside, if inflation continues moderating (with U.S. core PCE near or below 3%), the “looser monetary policy” environment could reinvigorate growth stocks and push more capital into longer-duration assets, e.g., technology or infrastructure.

Market Strategies
Asset allocators are increasingly adopting barbell approaches, mixing short-term Treasuries for yield and longer-dated inflation-linked bonds as a potential hedge if growth or inflation surprise to the upside. For private debt managers, higher front-end rates still provide strong deal economics, but if the Fed’s pace of cuts accelerates, floating-rate structures could see returns compress slightly.

4. Equities: A Broadening of Leadership & Regional Nuances

In 2024, the Magnificent 7 “AI-driven” tech giants led the S&P 500 to new highs, but by Q4, there was visible rotation into dividend payers, cyclicals (industrials, consumer discretionary), and mid-cap names. Many buyside analysts expect this sector rotation to persist, especially if rate cuts become visible and a “Trump stimulus” effect underpins the U.S. economy.

4.1 U.S. Equities—Striking the Balance

While forward multiples seem lofty—some top tech names trade at ~30x 2025 earnings—underlying profitability remains solid. The average S&P 500 operating margin is close to 12%, surpassing the 10-year average of 10.5% (FactSet data⁹). This margin expansion is partly due to cost discipline carried over from the pandemic era. As the Fed cuts rates, a slow but steady expansion in consumer demand could keep earnings growth at mid-single digits. Investors, however, remain wary of potential political showdowns (spending, tariffs) that might cause short-term volatility.

4.2 European Equities—Value Trap or Opportunity?

Europe trades at a ~40% valuation discount to the U.S. on a forward P/E basis, according to Bloomberg¹⁰. While that discount partly reflects slower growth, some see a “reform impetus” brewing. A widely discussed report by Mario Draghi for the European Commission urges deregulation and capital markets union to jumpstart competitiveness. If reforms make progress, coupled with any resolution on the Russia–Ukraine front, we could see a re-rating in key industrial, healthcare, or financial European names. If, however, the Trump administration hits Europe with new tariffs on autos or agricultural goods, the region may see further headwinds.

4.3 Emerging Markets—Divergent Paths

China’s policy easing and infrastructure push (estimated at ~10% of GDP, though below the 2009 stimulus scale) aim to revitalize domestic demand¹¹. If successful, it could spill over to suppliers in Southeast Asia and resource-rich Latin American countries. Meanwhile, India’s growth (projected ~6.5% in 2025 by the IMF¹²) remains a bright spot, especially in technology services and consumer markets. However, a strong U.S. dollar and any intensification in Sino-U.S. tensions are critical to watch for EM equity valuations.

5. Private Markets: Reaccelerating from a Low Base

5.1 Dealmaking & Fundraising Trends

Over 2023–2024, we saw private equity investment activity slow significantly from the 2021 peak, particularly in large-cap buyouts. Pitchbook data¹³ shows that global buyout deal value through Q3 2024 reached about $1.3 trillion, up 30% on the same period in 2023, but still well below the $1.7 trillion annual record of 2021. The fundamental reason for the tepid pace has been high financing costs and a persistent valuation gap between buyers and sellers.

However, in the final months of 2024, buyers and sellers appear closer on price, aided by the prospect of lower interest rates. Morgan Stanley’s credit analysts¹⁴ also note that private credit funds are stepping in to fill financing gaps where syndicated loans remain expensive, leading to more deals clearing. If 2025 continues in this direction, we could see an uptick in exits, which in turn frees up LP capital and rejuvenates fundraising momentum.

5.2 Secondaries

Drivers of Booming Volume
In H1 2024 alone, secondaries deal volume reached $70 billion, putting the market on track for ~$140 billion by year-end, according to Evercore¹⁵. The fundamental driver is the mismatch between LPs’ desire for liquidity—particularly as distributions slowed—and the robust dry powder among secondary buyers. Simultaneously, GPs increasingly use GP-led secondaries (continuation funds, single-asset deals) to hold onto stellar assets longer while offering liquidity options to their LP base.

Why 2025 Could Remain Strong
Even if exit activity recovers, secondaries are now a mainstream portfolio management tool. New sellers, including first-time smaller pension funds and endowments, discovered in 2023–2024 that secondaries could help them reduce overexposure or rebalance sector tilt without waiting for a typical 5–7 year fund life. With an estimated $253 billion of dry powder in secondaries, based on data from Secondaries Investor¹⁶, the market can absorb a wave of LP-led transactions. As for GP-led, sponsor appetite to crystallize returns or extend top-performing assets (particularly in growth themes like TMT or healthcare) remains unabated.

Implications for Investors
For limited partners in secondaries funds, the attraction is acquiring seasoned portfolios at potentially discounted valuations. This can mitigate the J-curve and accelerate distributions compared to primary commitments. However, buyers must be selective; in an environment where some GPs have concentrated 2021–2022 vintage assets at high valuations, thorough underwriting is essential.

5.3 Fintech

From Boom to More Measured Growth
The fintech sector experienced a slowdown in 2023, triggered by rising interest rates that made growth capital pricier and forced many startups to pivot toward profitability. According to PwC’s Fintech Deals Monitor¹⁷, total global fintech deal value dropped by 30% in 2023. Yet in 2024, there was a modest recovery—up about 12% year-over-year through Q3 2024. Part of this rebound stems from financial sponsors noticing that valuations had become more reasonable and from the fundamental demand for digital financial services.

What Drove the 2024 Fintech Uptick?

Payment Platforms & Embedded Finance: Consumers resumed travel and e-commerce spending, especially in North America and parts of Asia. This lifted transaction volumes for payment processors and B2B payments solutions. At the same time, large incumbents realized they could embed new payment or lending solutions into existing ecosystems—like buy-now-pay-later integration with major shopping apps.
Shift to Profitability: As venture capital for early-stage fintech became scarce, mid- to late-stage startups focused on cost discipline, “flat” or “down” rounds, and reaching break-even quicker. For example, certain digital banks in Europe renegotiated their user-acquisition strategies to slash marketing costs by up to 40%. This fosters more measured valuations that private equity growth investors find attractive.
Regulatory Clarifications: The U.S. clarified some stablecoin rules, while Europe finalized elements of the Markets in Crypto-Assets (MiCA) framework, reducing legal uncertainties for fintech infrastructure providers.

Looking Ahead: 2025 Opportunities
Payment solutions bridging e-commerce and brick-and-mortar are poised for continued tailwinds. Adoption of digital wallets is surging not only in the U.S. but especially across Asia (China, India) and Latin America (Brazil, Mexico), where smartphone penetration fosters leapfrogging from cash to mobile. By 2025, research by Goldman Sachs¹⁸ suggests that digital wallet transactions could represent 25% of global consumer payments, up from ~15% in 2023. This trend opens a door for:

Strategic buyers: Big banks and legacy payment networks might seek acquisitions to expand mobile offerings.
Private equity sponsors: Financing roll-ups of fintech companies that unify back-end payment processing or deliver cross-border functionalities.
Emerging Mark Opportunities: High unbanked population in certain regions means strong potential for digital lending, micro-insurance, and cross-border remittance solutions.

Nonetheless, the risk of new tariffs or cross-border data restrictions could complicate expansions for fintechs reliant on global cloud infrastructure or cross-border transfers. Investors must weigh the resilience of each business model against potential regulatory or political shocks.

5.4 Financial Institutions

Why FIG is in Focus
Financial institutions faced cyclical headwinds from 2022’s higher rate environment, particularly regional banks in the U.S. that struggled with deposit outflows to money market funds. As central banks pivot, the cost of deposits may stabilize, and a wave of consolidation could emerge. S&P Capital IQ data¹⁹ shows that in 2024, mid-tier bank M&A in the U.S. was up 28% year-over-year, driven by a search for scale and cost synergies.

Potential Deregulatory Effects
Trump’s return to the White House might prompt partial rollback of Dodd-Frank rules or thresholds, potentially reducing compliance burdens for smaller or regional banks. If so, it could enhance profitability for these banks and encourage them to expand via acquisitions. A similar dynamic occurred post-2017, when over 20 notable regional bank tie-ups closed within two years. However, if a more polarized Congress complicates major legislative changes, only incremental deregulation may occur.

Opportunities

Carve-outs of Non-Core Assets: Large banks might divest wealth management units or specialized lending units to streamline, presenting buyout opportunities.
Insurance Consolidation: As rates stabilize, run-off blocks of life insurance or annuities could attract private equity interest for yield strategies.
Fintech-FIG Convergence: Traditional insurers or banks might seek AI-driven underwriting platforms or digital-broker solutions. Partnerships or acquisitions here can unlock new client segments.

Risks revolve around changing capital requirements or the reemergence of bank-run anxieties if deposit rates remain competitive, undermining smaller institutions’ deposit bases.

5.5 TMT

AI ‘Hype’ vs. Reality
In 2024, Big Tech’s annual capital expenditures soared 90%, primarily for data center build-outs and advanced GPUs to train and deploy generative AI models, according to Bloomberg Intelligence²⁰. Hyperscalers such as Alphabet, Microsoft, and Amazon collectively spent over $280 billion in capex, up from $150 billion just two years prior. The impetus: capturing the enormous potential of AI in enterprise software, cloud services, and cybersecurity.

Yet these outlays raise questions about overinvestment and ROI timelines. Some brokers note that an “AI winter” could materialize if actual enterprise adoption lags behind current bullish projections. Nonetheless, any near-term correction in AI-related stock valuations could be short-lived given the genuine productivity gains that advanced analytics and automation promise. Indeed, Morgan Stanley²¹ estimates the total addressable AI software market to exceed $600 billion by 2027.

M&A Outlook
Private equity interest in TMT remains robust, with a particular tilt toward:

Vertical Software: Solutions for healthcare, real estate, legal or financial services. By acquiring niche players, sponsors can integrate advanced data analytics, scale distribution, and cross-sell to new markets.
Telecom Infrastructure: Ongoing 5G rollouts, fiber expansions, and edge computing facilities. Some mid-sized telecoms look to spin off tower portfolios or data-center arms to monetize stable cash flows.
Media & Content: While streaming competition compresses margins, certain production assets or specialized content libraries remain attractive as consolidation avenues.

5.6 Industrials

Global Manufacturing & Nearshoring
Industrials have contended with supply chain reconfigurations since the pandemic. Now, with potential new tariffs looming, manufacturers are reevaluating reliance on Asia and considering expansions or relocations within North America and Europe. Mexico, for instance, saw a 40% YoY uptick in foreign direct investment in 2024, according to Economist Intelligence Unit²², reflecting nearshoring demand from U.S. companies. This environment fosters:

Roll-up strategies in specialized machine shops or automotive component suppliers that embed robotics or AI-driven production lines.
Green energy manufacturing, e.g., battery gigafactories or hydrogen-related equipment, benefiting from large government incentives in Europe (RePowerEU) or the U.S. (Inflation Reduction Act).

Digital Industrial Revolution
According to S&P Global²³, industrial automation investments soared by over 30% year-on-year through Q3 2024. Private equity sees a chance to use advanced analytics, IoT sensors, and AI-based scheduling software to drastically cut costs and improve throughput in under-optimized manufacturing plants. KKR’s acquisition of Fuji Soft exemplifies a private equity push into automation—merging software capabilities with hardware solutions.

Risks & Potential Headwinds

Industrial businesses remain highly sensitive to commodity price fluctuations. If global tensions create short-term energy spikes, margins can be squeezed.
The success of nearshoring also hinges on labor availability and wage competitiveness. A shortage of skilled labor could hamper expansions or push up operating costs.

6. Conclusion: Balancing Opportunities & Risks in 2025

Heightened potential for pro-growth policy in the U.S. and a gradual pivot to rate cuts in most advanced economies sets the stage for an acceleration in private market deals and equity performance next year. Simultaneously, geopolitical tensions and a still-frail global trade environment serve as reminders that macro risk remains elevated. In this climate, investors can consider:

Moderate Overweights in equities, with a focus on cyclical (industrials, certain consumer names) and dividend stocks as leadership broadens.
Selective Private Equity Strategies in secondaries (where discounted entry points and liquidity solutions thrive), growth segments of fintech, AI-driven TMT, and nearshoring-led industrial deals.
Credit exposures that harness private lending’s higher yields, ensuring diligence on borrower quality amid potential policy volatility.

Overall, 2025 offers promise of a renewed business cycle marked by advanced technology adoption, macro tailwinds from falling interest rates, and carefully timed fiscal expansions. Yet success in this environment will demand deeper operational value creation, astute management of geopolitical risks, and flexible capital structures that can adapt to rapid shifts in policy or commodity markets.

Footnotes / Data References

Goldman Sachs U.S. Economic Research Note, October 2024. ↩
J.P. Morgan Global Data Watch, November 2024. ↩
IMF World Economic Outlook Update, October 2024. ↩
S&P Global Commodity Insights, 2024 Black Sea Grain Corridor Report. ↩
Goldman Sachs U.S. Policy Outlook, “Trump 2.0: Potential Market Impacts,” November 2024. ↩
Rothschild & Co. 2025 Global Investment Perspectives, December 2024. ↩
Morgan Stanley Macro Research, “Fiscal Multiplier Analysis,” September 2024. ↩
Rothschild & Co. “Consensus Outlook 2025,” Market Survey Data, October 2024. ↩
FactSet Earnings Insight, Q4 2024. ↩
Bloomberg Terminal data, Global Equity Valuations Dashboard, November 2024. ↩
People’s Bank of China and Ministry of Finance announcements, aggregated by CEIC, August 2024. ↩
IMF World Economic Outlook Database, October 2024. ↩
Pitchbook Global PE & VC Report, Q3 2024. ↩
Morgan Stanley Credit Research, “Private Credit Surge Amid Syndicated Loan Pullback,” October 2024. ↩
Evercore Secondary Market Survey, H1 2024. ↩
Secondaries Investor data platform, “Global Secondary Dry Powder,” August 2024. ↩
PwC Fintech Deals Monitor, September 2024. ↩
Goldman Sachs, “Digital Wallets & The Future of Payments,” July 2024. ↩
S&P Capital IQ FIG M&A Tracker, October 2024. ↩
Bloomberg Intelligence, “AI CapEx Soars 90% in 2024,” November 2024. ↩
Morgan Stanley Equity Research, “The AI Software Opportunity,” May 2024. ↩
Economist Intelligence Unit (EIU), “Mexico FDI Trends 2024,” October 2024. ↩
S&P Global Industrial Automation Overview, Q3 2024. ↩

Reddit (RDDT): Fundamentals

2025年1月2日2025年1月18日 Jerry NgLeave a comment

Question #1: Is This a Good Business?

Subjective vs. Objective

Subjective Indicators
- Massive Network of Communities (aka “Subreddits”): Over 100,000+ active subreddits, covering a vast array of user interests. This “community of communities” fosters deep engagement and user stickiness.
- Highly Engaged User Base: Reddit’s U.S. daily active uniques (DAUs) are spending an estimated 25–30 minutes/day on the platform, surpassing many social platforms in “time spent per session.”
- Unique Demographic: ~50% of U.S. Reddit users are male, heavily skewed 18–44—an audience often considered harder to reach on other mainstream social networks.
Objective Indicators (via Cash Flow Potential)
- Revenue Growth Trajectory: Management and street estimates see 40%+ yoy ad revenue expansion in 2024. Many forecast a 25–30%+ CAGR over the next several years.
- Profit Margin Outlook: Despite a negative GAAP operating margin in the past, high incremental margins (given software-like cost structure) suggest potential to reach 35–45% adjusted EBITDA margins over time.
- Capital Intensity: Server and AI-related investments remain a key near-term expense. However, Reddit’s “capital” is mostly intangible (software engineering + content moderation). Physical capex is relatively modest compared to hardware-intensive peers.
- Terminal Value Perception: If Reddit successfully monetizes (via improved advertising, data licensing, and possibly search ads), the market could apply a premium akin to high-growth ad-tech peers. If user growth decelerates or brand safety becomes an outsized issue, that premium compresses.

Value Creation “Focus Five”

Organic Growth: Currently mid-to-high double-digit yoy (ads).
Margin Trajectory: Negative GAAP margins historically, but Street expects a path to 35%+ EBITDA margins by 2026–2027.
Capital Intensity: Primarily R&D + AI/ML infrastructure, leading to moderate intangible intensity. Physical capex is lower than average.
Capital Deployment: Net proceeds typically reinvested in platform improvements, AI (for relevance/ranking), moderation tools, and incremental acquisitions (e.g., Spiketrap).
Terminal Value Perception: A “community-driven” moat with high user loyalty and potential for multiple monetization streams.

Verdict: Reddit’s “community + conversation + search synergy” can yield a profitable, high-margin business at scale. Its intangible “flywheel” (community growth → more content → more engagement → more ad inventory) suggests a positive long-term outlook if it continues delivering on user acquisition and advertiser ROI.

Question #2: How Does This Business Make Money?

Revenue Model
- Advertising (~90%+ of revenue)
  - Primarily performance-driven ads (cost-per-click, cost-per-install) plus brand-focused “takeover” ads.
  - Auction-based system for standard placements; some reserved inventory for 24-hour takeovers.
- Other (~10% of revenue)
  - Data Licensing: Partnerships with major tech players (e.g., Google, OpenAI), leveraging Reddit’s user-generated data for AI training or search indexing.
  - Premium Subscription: Ad-free experience + exclusive features at $5.99/month. Estimated 100–150k paying subs.
  - Avatars & Virtual Goods: Niche but growing commerce line.
Pricing Model
- Ad units sold on a CPM or CPC basis. Some takeover options sold at fixed daily rates.
- Data licensing deals typically multi-year with usage-based or subscription-based fees.
Customer Concentration & Contracts
- Top 10 advertisers contribute ~25–30% of ad revenue. Diversifying advertiser base is a key objective.
- Reddit’s data-licensing deals (e.g., with Google) often run 2–3 years with locked-in fees.
Recurring Revenue
- Ads are somewhat recurring if advertisers see stable ROI. But budgets can shift quickly.
- Data licensing is more contractually predictable, though still a minority of total revenue.
- Subscription revenue is small but stable, given monthly renewals.

Summary: Advertising remains the dominant engine, with large upside if ad-tech improvements continue (improving targeting, conversions, automated bidding). Data licensing and subscriptions provide incremental, less cyclical revenue streams.

Question #3: What Is the Nature of the Cost Structure?

Economics at the Granular Level
- Fixed Costs: R&D (AI engineering, platform reliability), a portion of G&A (compliance, overhead), safety/moderation infrastructure.
- Variable Costs: Hosting (cloud infrastructure, content delivery), sales and marketing commissions, moderation support for scaling user content.
Operating Leverage
- High portion of expenses in R&D and overhead → once the platform’s ad-tech, ML models, and community frameworks are in place, incremental ad revenue flows at strong contribution margins.
- As user traffic and engagement expand, hosting and moderation costs rise but typically at a slower rate than revenue (once scale is reached).
Incremental Margins
- Because Reddit’s user-generated model doesn’t require content spend (like Netflix), each additional monetized impression can generate high incremental margins.
- On new ad products (e.g., search-based ads), early estimates point to significant margin expansion if successful.
Cost “Line Items”
- R&D: Possibly 20–25% of sales near term, heavily invested in ML for ad relevance and new ad formats.
- S&M: ~15–20% of sales as Reddit expands direct sales teams and mid-market/self-serve channels.
- G&A: ~5–10% of sales, covering overhead, legal compliance, content policy enforcement.
- Infrastructure/Hosting: ~10–15% of sales, though subject to scaling nuances with AI workloads.

Conclusion: Once Reddit’s overhead and R&D costs are absorbed, incremental revenue from ad load or improved ad pricing flows disproportionately to EBITDA, implying robust leverage potential.

Question #4: Key Drivers of the Business

User Growth & Engagement
- DAU: Currently ~90M+ daily U.S. uniques, with total global DAU near 180–200M by some estimates. Street targets a mid-teens CAGR.
- Time Spent: ~25–30 minutes/day for U.S. logged-in users. Even modest improvements in session length compound revenue potential via ad impressions.
Ad Product Innovation & Performance
- Better Targeting & Measurement: Tools like CAPI (conversion API), brand safety solutions, new ad formats (e.g., in-comment ads) drive improved ROI → higher ad spend.
- Search Ads: Big potential driver if fully launched, given >1B monthly on-platform searches.
- End-to-End Automation: Auto-bidding and dynamic creative optimization can unlock more advertiser demand.
International Expansion
- Currently ~50% of Reddit’s DAUs are in the U.S. Large under-penetration in EMEA, LATAM, APAC.
- Rolling out localized UIs, machine translation, region-specific subreddits to accelerate adoption.
Partnership & Ecosystem
- Data Licensing: Partnerships with Google, OpenAI, and potential new AI players.
- Commerce/Shopping: Partnerships with e-commerce platforms or enabling transactions on subreddits.
- Developer Tools: Encouraging third-party tooling can expand the ecosystem and user use-cases.
Brand Safety & Moderation
- Effective content moderation fosters advertiser confidence. Negative PR around “toxic” content could hamper brand ad spend.

Model Sensitivities

A 10% upside in DAU or user time can drive 15–20% revenue upside given fixed overhead.
A 1–2% improvement in average CPM from better ad performance can yield outsized incremental EBITDA margins (sometimes 50%+ on incremental).

Debates

Competition: Big players (Meta, Google) continuously refine ad solutions—can Reddit truly differentiate?
Regulatory Risk: UGC (user-generated content) compliance, privacy laws, content liability concerns.
Search Monetization: Unclear how quickly Reddit can build an at-scale ad product in on-platform search.

Question #5: Business Momentum

Recent Accelerations/Decelerations
- Ad Revenue Growth Re-Acceleration: After a slower 2022–early 2023 environment, Reddit saw 50%+ yoy ad revenue growth in 3Q23, partly from new ad placements and rising user volumes.
- New Partnerships: Data licensing commitments exceeding $200M total contract value over a couple of years with Google.
2- to 3-Year Stack
- Up significantly in daily usage vs. 2021, though still small share (<1%) of global social ad spend ($180B market).
- Product pipeline (conversational ads, auto-bidding, first-party measurement) is rolling out from 2023–2025.
Forecast: Bull/Base/Bear Cases
- Bull: Rapid success of search ads, 20% yoy DAU growth, 35%+ EBIT margins by 2026. Potential stock upside from 2–3x EV expansion.
- Base: ~15% yoy user growth, mid-30s incremental margin, consistent product execution. Ad revenue +30% yoy near term.
- Bear: Monetization stalls if brand safety or slow ad-tech adoption hits. Single-digit user growth, teens-level margin, leading to more modest top-line expansion.
Expectations Gap
- The market may be pricing in “just” 20–25% ad CAGR. If Reddit hits 30–35% or more, a valuation re-rating is likely.
Momentum Indicators
- Advertiser surveys indicating ROI improvement.
- Growth in daily search queries (1B+ monthly) suggests readiness for search-based monetization in 2025–2026.

Conclusion: Reddit’s near-term momentum is driven by ad-tech enhancements and robust user traction. Any step-change in search ads or global expansions could further accelerate revenue. Investors watching sustained double-digit growth and margin ramp may revalue the stock in line with higher-growth social peers.

Final Takeaway

Reddit combines a deeply “community-first” platform with potential for high-margin ad expansion (via advanced targeting, new ad formats, and global scale). Although brand safety, content moderation, and intense ad-tech competition remain risks, the fundamental unit economics—large engaged user base, user-generated content, and software-like incremental margins—position Reddit for meaningful long-term value creation if it executes on product innovation and monetization levers.

Generative AI Value Chain: From Data to Deployment

2025年1月2日 Jerry NgLeave a comment

1. Introduction

In one of my old reading notes from Running Money—written in the late ‘90s—there was a line predicting that “today’s intangible knowledge is tomorrow’s capital.” Back then, it sounded bold, but if we look at Generative AI now, it’s an exact fulfillment of that prediction. The huge swaths of human-generated text, images, code, and speech act as “capital” that big labs, startups, and traditional enterprises alike are racing to harness.

In this updated overview, I’ll map out the entire generative AI pipeline end to end. That means we’ll look at not only how data is created and how models are trained, but also how post-training processes—such as inference infrastructure, product integration, monetization, and ongoing model updates—flesh out a practical ecosystem. I want to maintain the detail level that we gave to content providers and training for the post-training portion as well, including which players do what and how they’re establishing moats or encountering key challenges.

Let’s begin with a broad backdrop of what got us here and the overall value chain.

2. Why GenAI? The Forces Behind Its Rapid Rise

2.1 Moving Beyond Traditional ML

For decades, “machine learning” mostly meant classification: spam detection, image recognition, or forecasting. These tasks were specialized but rarely generated brand-new content. Then came the transformer architecture, enabling AI to handle sequences—like paragraphs of text or lines of code—and produce original output. Systems like GPT, Claude, or Bard exemplify how the AI community pivoted from “interpreting data” to “generating content,” achieving a sense of creativity that enthralls both the general public and specialized domains (e.g., coding, finance, law).

2.2 Key Catalysts

Massive Data Availability: User-generated content—Reddit discussions, Tweets, Q&A sites—plus institutional text from newspapers, journals, and books.
Accelerated Compute: Nvidia’s GPU dominance and cloud providers’ HPC setups let labs train on increasingly giant corpuses.
Commercial Momentum: Microsoft invests billions in OpenAI; other tech giants respond with their own generative initiatives, fueling competition (and hype).

2.3 Stakeholder Groups

Labs and Startups: OpenAI, Anthropic, Stability AI, etc.
Incumbent Tech Giants: Google, Microsoft, Meta, Amazon, Apple.
Content & Data Owners: Media, publishers, social platforms, academic resources.
Enterprises: Banks, pharma, retailers who want domain-specific AI.

All these players form an intertwined ecosystem, with content fueling the training, specialized clouds facilitating compute, and end-users driving monetization.

3. The Full AI Value Chain

To structure our discussion, let’s parse the entire chain into stages:

Content Generation & Curation: Platforms and providers who create or hold raw text, images, code, etc.
Data Preparation & Labeling: Specialized tasks that clean, structure, or annotate data for training.
Model Architecture & Training: Designing networks (transformers, chain-of-thought) and harnessing HPC for large-scale training runs.
Inference Infrastructure: Serving model outputs in real time, typically using GPU clusters or dedicated hardware.
Utilization & Monetization: Integrating AI into apps, enterprise workflows, or consumer-facing products.
Continual Learning & Model Updates: Ongoing refinements, user feedback loops, fine-tuning for specific clients, and model iteration over time.

We’ll discuss each with an equal level of detail, including examples of major players, moats, and controversies.

4. Content Generation & Curation

4.1 Data Sources

Social Media & Community Platforms
- Reddit: Detailed, topic-specific discussions in subreddits. Essential for capturing casual, in-depth user viewpoints.
- Stack Overflow/Kaggle: Programming Q&A and data science competitions provide a valuable problem-solution format.
- Quora, X/Twitter: Short-form queries, broad user base.
Institutional Media & Specialized Repositories
- Newspapers, Magazines: NYT, WSJ, Wired, trade journals with carefully edited, high-quality text.
- Academic Papers: ArXiv, PubMed, or Elsevier for peer-reviewed knowledge.
- Patents: Provide thorough technical descriptions, especially in engineering or biotech contexts.
Code Hosting Platforms
- GitHub: Rich, though varied, code repositories from open-source communities.

4.2 Curation & Cleaning

After raw data collection, labs or specialized data-engineering firms invest in:

Deduplication: Avoiding repeated text or near-duplicates.
Filtering for Quality: Eliminating profanity, spam, or poorly formatted content.
Domain Segmentation: Tagging content by domain or style to help “bucketing” during training.

4.3 Content Provider Moats

Exclusive Archives: A big newspaper with historical archives can charge a premium for licensing because labs crave comprehensive domain coverage.
Community Ecosystems: Platforms like Reddit are distinct because their user base is both the content generator and a potential groundswell for backlash if they disagree with licensing deals.
Data Quality and Trust: The New York Times has editorial rigor, raising perceived reliability for training. Meanwhile, code from professional GitHub repos might impart better coding style than random personal projects.

4.4 Why It Matters

A model that’s never seen certain forms of text (like specialized science data) will underperform in that area. Or if it’s only fed “casual chat” from social media, it might struggle with formal business writing. The breadth, quality, and representativeness of content define the baseline intelligence a model can achieve.

5. Data Preparation & Labeling

Beyond collection, the next step is transforming raw text/images into workable training sets.

5.1 Human and Automated Labeling

Human-Led Annotation: Workers read content, categorize it, or highlight key elements. For images, bounding boxes or segmentation. For text, sentiment or correctness tags.
Automated Pre-labeling: Some labs use heuristics or smaller pre-trained models to expedite the labeling process, with humans verifying edge cases.

5.2 Major Labeling Players

A specialized sub-industry addresses the labeling pipeline:

Scale AI
- Known for providing large-scale annotation services, with a global workforce.
- Partnerships with key OEMs in autonomous driving (object detection in images) and also with text labeling for NLP tasks.
Appen
- Australian firm focused on data annotation at scale for speech, text, and more.
- Historically strong in search engine relevance tasks, now pivoting to GenAI labeling.
Sama
- Known for ethical sourcing of annotators from emerging markets.
- Provides everything from bounding-box labels to complex textual QA tasks.
Cloud Vendors’ In-House
- AWS offers “Ground Truth” labelers; Google has internal data teams from acquisitions like Kaggle.

5.3 The Labeling Moat

Workforce Scale: Large agencies or platforms can ramp up thousands of annotators, handling big jobs quickly.
Expertise in Domain: Some labeling shops specialize in medical or legal text, providing higher-quality annotation for those use cases.
Tooling Ecosystem: Advanced annotation UIs, automated QC checks, or “human-in-the-loop” pipelines differentiate top-tier providers.

5.4 Pitfalls & Controversies

Labor Practices: Ethical concerns around low pay or psychological toll (e.g., content moderators seeing extreme material).
Accuracy vs. Speed: Rushed labeling can pollute the dataset with errors.
Cost: High-quality labeling for huge corpuses can be very expensive, creating a barrier to entry for smaller labs.

6. Model Architecture & Training

6.1 Primary Stakeholders

AI Lab Startups (OpenAI, Anthropic, Stability AI):
- Typically rent HPC from Azure, AWS, or GCP.
- Innovate with new transformer tweaks or chain-of-thought approaches.
- Data deals with publishers or labelers feed them exclusive or high-grade inputs.
Tech Giants’ Internal Labs (Google DeepMind, Meta, Microsoft, Amazon):
- Large in-house research staff and specialized hardware.
- Often have unique internal data from their own consumer platforms, e.g., search logs (Google), social media (Meta).
Enterprise-Specific Teams (Big banks, pharma, auto):
- Either do partial fine-tuning or smaller proprietary training runs.
- Emphasize compliance, domain specificity (e.g., financial language, medical notes).

6.2 Compute & Hardware

Nvidia: The undisputed leader with GPUs like A100, H100.
Google: Custom TPUs for internal training, also available on GCP.
AWS: Trainium and Inferentia custom chips, though GPU clusters remain dominant.
Meta: R&D into custom accelerators, but reliant on GPUs for now.

6.3 Methodological Approaches

Scale-Up (Massive Models)
- GPT-4 style, rumored hundreds of billions of parameters.
- Pros: Broad coverage, emergent capabilities.
- Cons: Ridiculously expensive training and high inference costs, diminishing returns if not curated well.
Mid-Sized & Specialized
- 7B–20B parameter models tuned to a domain (finance, law, coding).
- Pros: Cheaper training, often better within that domain.
- Cons: Less flexible outside the domain, can lose broader context.
Chain-of-Thought / Iterative
- E.g., “o1” models that do multi-step reasoning internally before finalizing an answer.
- Pros: More accurate logic, solves complex queries or mathematical tasks better.
- Cons: Demands more inference compute/time (the model is effectively “thinking aloud” internally), raising cost per query.

6.4 Risks and Tensions

GPU Shortages: Everyone clamoring for HPC capacity drives up costs, restricting smaller labs.
Legal Uncertainty: Ongoing lawsuits about copyrighted material used in training.
Return on Investment: As models grow, capital needed inflates, but revenue from final products may lag behind.

7. Inference Infrastructure

Even the best-trained model is moot if you can’t serve user queries quickly and affordably. Let’s dig into inference’s cost structure and major players.

7.1 Real-Time vs. Batch Inference

Real-Time: ChatGPT or Google Bard must respond in seconds, requiring top-tier GPUs or specialized hardware and robust orchestration to handle concurrency.
Batch: Some enterprise tasks can be processed offline, cutting costs with lower-tier hardware or time-scheduled usage.

7.2 Key Providers

Public Clouds (AWS, Azure, GCP):
- Offer managed AI inference clusters.
- HPC-like configurations with GPU pods or custom ASIC options.
- Solutions like Azure’s “OpenAI Service” seamlessly link training to deployment.
On-Premises HPC
- Enterprises like banks with strict data compliance might run inference on internal GPU racks.
- Nvidia sells DGX systems for high-performance local deployment.
Edge Inference
- Qualcomm or ARM-based chips for mobile or edge devices.
- Typically only feasible for smaller or pruned models (think local voice recognition on a phone).

7.3 Performance vs. Cost Trade-Off

High Parameter Models: Potentially better answers but cost significantly more to run.
Token-Level Billing: Some solutions charge by input/output tokens. Overhead can spike with large prompts or chain-of-thought expansions.
Latency Minimization: Users expect near-instant chat responses, driving architectural design for caching, GPU load balancing, or model distillation.

7.4 Monetization Implications

Inference cost heavily influences final product pricing. Tools like ChatGPT Plus/Pro tier or domain-specific enterprise offerings reflect the tension between providing high-quality responses and controlling usage-based overhead.

8. Utilization & Monetization: Bringing AI to Products

Now that we have a trained model deployed on robust infrastructure, how do we integrate it into real use cases?

8.1 Productivity Suites and Tools

Microsoft 365 Copilot: Integrates GPT-based features into Word, Excel, Outlook. Automates draft generation, summarization.
Google Workspace: Duet AI for composing emails, summarizing documents.
Adobe Firefly: AI-driven creative assistance within Photoshop, Illustrator.

8.2 Vertical/Domain-Specific Applications

Finance

AI-driven portfolio optimization or risk modeling.
Natural-language question answering on investment products (private banks adopting GPT-fine-tuned systems).

Healthcare

Radiology image interpretations with large vision-language models.
Chat triage or doctor-patient summarization tools, mindful of HIPAA compliance.

Legal

Contract review, e-discovery.
Summaries of case law.
Sensitive to hallucinations (incorrect references) and must maintain confidentiality.

Emergency Services (like an AI “Ambulance” scenario)

Real-time triage chatbots in crisis lines.
Predictive models for resource allocation (where ambulances might be needed next).
Potential for misinformation if not carefully validated.

8.3 Monetization Models

Subscription Tiers
- E.g., ChatGPT Plus ($20/month), ChatGPT Pro ($200/month) for more advanced or unlimited queries.
- Enterprise deals negotiated for seat-based or usage-based fees.
API Pricing
- Developers pay per token or per request.
- Some labs also offer monthly packages for certain volumes.
Advertising or Referral
- Possibly integrated into queries (like Bing Chat embedding sponsored links).
- Lower-likelihood for heavy enterprise usage, more relevant for consumer-level queries.

8.4 The Role of Ecosystem Partnerships

Integrators & Consulting: Accenture, Deloitte, etc. building custom solutions with LLMs inside big enterprises.
Startups: Provide specialized front-ends, domain datasets, or agent-based shells on top of foundational models.
Open-Source: Models like Llama 2 let smaller players create their own specialized spin-offs without licensing big-lab IP.

9. Continual Learning & Model Updates

No model stays static—iterative improvements are key.

9.1 Approaches to Ongoing Refinement

Fine-Tuning on New Data: If an enterprise accumulates new domain data, they can further refine the model’s weights.
User Feedback Loops: Some products prompt the user: “Did this answer help?” to gather signals for reinforcement learning.
Active Learning: Model flags uncertain samples for human review, retraining only on these critical data points.

9.2 Player Ecosystem

OpenAI: Conducts routine “model refreshes,” collects anonymized user queries to refine system prompts and guardrails.
Self-Hosted: Companies that deploy local versions must decide how often to pull updated weights or do their own iterative training.
Consultancies: Provide ongoing support to calibrate new data or shift model biases if the real-world environment changes.

9.3 Risks & Benefits

Data Drift: If the world changes (new laws, new events), a stale model may give outdated or harmful advice.
Misalignment: Continual tweaks can produce unexpected side effects in emergent behavior if not tested thoroughly.
Competitive Edge: Faster iteration can lead to better performance and user retention.

10. Tensions and Debates Across the Chain

Data Licensing vs. Free Scraping
- Publishers want revenue; labs want more text. Courts are deciding what’s “fair use.”
Consolidated Compute Market
- Nvidia GPUs overshadow alternatives. Startups can’t always outbid major players for HPC capacity, limiting new entrants.
Quality vs. Cost in Labeling
- Over-labeled or poorly labeled data can hamper model accuracy. Good labeling is expensive.
Inference Pricing
- End users may balk if subscription tiers or per-query fees get too high—on the other hand, labs can’t sustain indefinite free usage with high GPU bills.
Model Hallucination & Liability
- For medical/legal, who’s responsible if the model’s suggestion causes harm?
- Regulators and industry associations might impose usage disclaimers or demand thoroughly tested processes.
Ethical & Societal Impact
- Risk that easy generation of misinformation outpaces factual content.
- Risk that generative AI exploits user content without giving enough credit or compensation.

11. Future Outlook: The Evolving GenAI Ecosystem

11.1 Continued Growth & Diversification

More Domain-Focused: Finance, law, biotech, etc. could each see specialized LLMs, licensed with curated data from those industries.
Increased Open-Source Competitiveness: Groups like Meta or Hugging Face pushing more advanced open models, accelerating innovation outside big labs.

11.2 Hardware Innovations

New HPC Paradigms: Cloud vendors refining custom chips or new GPU generations to slash inference costs.
Edge & On-Device: Distilled or quantized models for real-time local inference (especially for voice or AR applications).

11.3 Regulatory Frameworks

Copyright: Formal guidelines clarifying what labs can or can’t scrape.
Privacy & Security: Requirements that personal user data be stripped or anonymized in training sets.
Model Accountability: Third-party audits on large models, especially for high-stakes domains.

11.4 Societal and Content Ecosystem Rebalancing

As generative text proliferates, human-originated content might be overshadowed, prompting content creators to demand stricter licensing or unique monetization channels. Platforms like Reddit are likely to keep rethinking user policies to maintain strong communities while also monetizing data deals.

12. Conclusion

Generative AI’s journey is shaped by the interplay of content providers (and their licensing terms), labeling specialists, training labs, infrastructure providers, application builders, and end-users. Each link in this chain has its own set of moats, controversies, and strategic levers:

Content: The bedrock of AI’s knowledge, with data owners increasingly seeking compensation.
Preparation & Labeling: The crucial step to ensure the raw data is accurate, well-organized, and ethically sourced.
Training: The high-stakes, high-cost process of forging a model’s “intelligence.” The choice between bigger multi-domain vs. specialized smaller models remains an active debate.
Inference & Integration: Determines how effectively real-world products can deliver AI’s benefits—and at what price.
Continual Learning: Ensures the model stays relevant in a changing world.

We’ve also seen how final “utilization” covers everything from plugging the model into business productivity tools to specialized verticals like finance, law, and even AI-driven ambulance triage. Each segment will keep evolving, especially as labs refine technologies like chain-of-thought inference or specialized HPC solutions that reduce cost while improving performance.

Key takeaway: The GenAI industry is a dynamic web of interdependent players, each racing to protect their portion of the value chain—whether by controlling data assets, selling HPC capacity, innovating in labeling pipelines, or offering domain-specific usage. For those building in this space, understanding each node of the chain is not optional; it’s essential to see where your strategic advantage or synergy might lie.

With this broader pipeline in hand, we can better interpret why certain AI companies or partnerships form and foresee where friction or synergy will shape the next generation of breakthroughs. I’ll continue exploring these themes, especially how domain verticalization and real-time/edge inference create new categories of products—like AI financial modeling or legal co-pilots, among others.

Ultimately, as Running Money foretold, knowledge is the capital driving 21st-century technology. And in the GenAI era, that knowledge must be carefully sourced, curated, modeled, delivered, and continuously improved to stand out in an increasingly competitive field.

Reading Notes from Liar’s Poker by Michael Lewis

2024年12月24日 Jerry NgLeave a comment

1. Key Takeaways & How Lewis Addressed Them

Take #1: The Culture of Wall Street Excess

What It Means
In Liar’s Poker, Michael Lewis immerses us in the late-1980s Salomon Brothers—a place of aggressive risk-taking, brash traders, and a culture where making money is revered above all else. Lewis, a new recruit, portrays the environment as a microcosm of Wall Street’s competitive, often cutthroat ethos.

Real Examples in the Book

Solly’s Trading Floor: Lewis describes the trading floor as a “madhouse” brimming with shouting traders—bond salesmen who treat deals like a zero-sum gladiatorial contest.
The “Big Swinging Dicks”: A recurring phrase for the top traders who dominate and set the tone: bigger bonuses, bigger egos, bigger risks. This is how Salomon elevates ruthless personalities into role models.

How Lewis Addressed It

Personal Observations: Lewis joined as a naive young man with an art history background. By observing how “Big Swinging Dicks” behave, he reveals the internal tension he feels between wanting success vs. moral or ethical hesitation.
Humor & Cynicism: He uses witty, often sardonic storytelling to show how over-the-top the culture is—caricaturing each character’s vanity and greed.

Take #2: The Rise of Mortgage Bonds & Financial Innovation

What It Means
A major theme is how Salomon Brothers pioneered mortgage-backed securities (MBS), effectively bundling home loans into tradable bonds. This “financial innovation” unleashed enormous profits in the 1980s.

Real Examples in the Book

Lewis Ran in the Mortgage Department: He started in bond sales, then specialized in mortgage bonds, watching how a “quiet, dull corner of finance” became a multi-billion-dollar money machine.
Traders’ Risk Appetite: Mortgage traders kept inventing new derivatives, layers of complexity to repackage or hedge risk, which often led to ballooning volumes (and fortunes).

How Lewis Addressed It

Exposes the Hype: Through direct inside anecdotes (e.g., managers bragging about “no one can lose money on mortgages,” or pushing deals to unsuspecting clients), Lewis reveals the hidden fragility beneath those profits.
Simplifies Complexity: By writing with clarity and humor, he demystifies MBS and the origin of 1980s “junk-bond mania,” showing how ignorance or greed overshadowed caution.

Take #3: “Liar’s Poker” as a Metaphor for Deception & Bluffing

What It Means
“Liar’s Poker” is a game played with dollar bill serial numbers, a mind game of bluffing and intimidation—mirroring how traders at Salomon Brothers operate. The bigger your bluff, the more you can push around the market or your colleagues.

Real Examples in the Book

John Gutfreund’s Challenge: Lewis recounts an infamous scene where CEO John Gutfreund proposes a $1 million per-point Liar’s Poker match. The audacious bet is symbolic of the huge risks and ego-driven showdowns.
Daily “Bluffs” on the Trading Floor: Traders routinely mislead or “talk their book” to manipulate bond prices, test each other’s nerves, or scare timid new hires.

How Lewis Addressed It

Focus on Human Psychology: He highlights how personal pride and fear of looking “weak” drive decisions more than logic.
Self-Reflection: In one anecdote, Lewis declines to challenge Gutfreund, recognizing that ego fights overshadow rational investing. His outside perspective allows him to see the humor and danger in these “bluffs.”

Take #4: Rewards vs. Ethics in Trading

What It Means
High bonuses create moral hazards. Salesmen push questionable deals on unsuspecting clients. Inside Salomon Brothers, a “survival-of-the-fittest” ethic can overshadow ethical responsibility.

Real Examples in the Book

Selling Junk Mortgage Bonds: Some traders knowingly packaged suboptimal mortgage tranches for S&Ls (savings & loans). They reaped outsized commissions but left clients with toxic holdings.
Client vs. Firm Conflict: Lewis notes that the firm’s priority is maximizing short-term profit—ethics or client well-being is secondary.

How Lewis Addressed It

Personal Conflict: He recounts times he hesitated to push a product he found dubious. He does so anyway under group pressure but remains uneasy, which eventually leads him to quit.
Exposé Purpose: By writing Liar’s Poker, Lewis lifts the veil on these practices, hoping outsiders see how moral corners are cut when big bonuses are at stake.

Take #5: Luck, Timing & Wall Street Careers

What It Means
Lewis suggests Wall Street success hinges on being in the right place at the right time (e.g., the mortgage bond boom). He was hired almost by accident, not for deep qualifications.

Real Examples in the Book

Lewis’ Accidental Start: He meets a Salomon exec at a dinner party and, improbably, that leads to a job. Many top traders also stumbled into Salomon without a strong finance background—once they’re in, fortunes can be made fast.
1980s Boom: The timing of the roaring bond markets catapults novices into high-paying roles they might never replicate later.

How Lewis Addressed It

Reflections on Fate: He repeatedly calls himself “lucky,” highlighting that big money and big trades are not always about skill alone.
Irony & Self-Awareness: He eventually quits, concluding that the entire system is more luck-driven than most want to admit, an epiphany that shapes the book’s cautionary tone.

2. Key Opinion (Lewis’s Overall Standpoint)

“Wall Street is part bravado and part big money, with little accountability. The real game is about bluffing others into believing you know more than you do—and sometimes it works until it doesn’t.”

Reflection: Liar’s Poker pulls back the curtain to show an industry with high stakes and swagger, yet underpinned by illusions. Lewis believes that while some truly understand financial products, many profit from hype or fleeting edges.

3. Selected Quotes & Commentary

Quote #1: On the Trading Floor Environment

“If a trader could make his boss $100,000, that boss would turn a blind eye to just about anything.”

Relevance: This epitomizes the moral hazard—profits overshadow decency or caution.

Quote #2: On Mortgage Bonds

“The business was brand new. There was no history to study, no well-worn rules. We made them up as we went along.”

Relevance: Shows the “Wild West” aura: unregulated or ill-understood markets letting cunning players pocket huge gains.

Quote #3: On Luck vs. Skill

“The real wonder of Salomon wasn’t that we were good at what we did; it was that, for a brief shining moment, others believed we were.”

Relevance: This underscores the power of perception—clients and the market placed blind trust in Salomon’s “genius.”

4. Reflection: How Lewis Tackles Each Point

He Observes from the Inside: As a “middleman” in the mortgage desk, Lewis can see how trades are pitched and closed.
He Depicts Characters as Archetypes: He rarely sugarcoats. E.g., The “Big Swinging Dicks” symbolize hyper-masculine, money-obsessed success.
He Infuses Irony & Humor: By writing with comedic flair, Lewis invites readers to be both entertained and alarmed.

5. Building Goodwill & Connections

Salomon’s “Class System”: New hires must survive hazing. Senior traders or managers prefer loyalty to big deals. If you bring in money, you’re golden.
Client Relationships: Often borderline manipulative—selling them bonds they don’t fully understand. Still, forging strong personal ties matters for future deals.

6. Dealing with Authority

Internal Authority (CEOs, MDs): People like John Gutfreund or the mortgage-bond desk heads rule by intimidation. Their approval is everything.
Regulatory Oversight: In the 1980s, it was weak, so unscrupulous behavior soared until bigger crises (like the S&L collapse) forced partial clampdowns.

Background & Conclusion

Michael Lewis wrote Liar’s Poker as partly a memoir, partly an exposé, capturing the era of 1980s bond trading mania at Salomon Brothers. Its lasting resonance comes from these revelations:

Wall Street’s Bold, Edgy Culture sets short-term revenue above ethics.
Financial Innovations (like mortgage-backed securities) can be rocket fuel or a time bomb, depending on knowledge and caution.
Luck, Ego, and Bluffing shape fortunes just as much as skill.

Lewis ultimately leaves Wall Street, disillusioned by the greed and folly he witnessed, concluding that the “big game” is more fragile than it appears.

Final Thought: Liar’s Poker stands as a timeless cautionary tale about the illusions of unbridled capitalism: fortunes can be won quickly by cunning or luck, but deep cracks—ethical or economic—always lurk beneath. As Lewis implies:

“The biggest risk is believing your own hype. In a world of Liar’s Poker, the best bluff might fool everyone—even yourself.”

Intel’s Mobileye Spin-Off and IPO: Unlocking Value While Preserving Control

2024年12月22日 Jerry NgLeave a comment

Intel’s decision to spin off and partially IPO Mobileye was years in the making. Although Intel originally acquired Mobileye in 2017 for about $15 billion, the chip giant remained determined to reveal Mobileye’s intrinsic value as a leading autonomous driving company. By listing Mobileye on the public markets while still retaining a large majority stake, Intel aimed to raise capital, refocus on its core priorities, and let Mobileye shine under its own brand.

I. Background and Context

The Mobileye Acquisition

Original Deal: Intel purchased Mobileye in 2017 for $15 billion, seeking to gain a foothold in the Advanced Driver Assistance Systems (ADAS) and autonomous vehicle (AV) space.
Strategic Rationale: Mobileye’s EyeQ chips, software stack, and deep relationships with major automakers gave Intel an instant presence in automotive semiconductors—an area expected to grow substantially.

Why a Spin-Off?

Surfacing Value: Within Intel’s broad portfolio, Mobileye’s growth potential was overshadowed by Intel’s legacy x86 business. A separate public listing was seen as the best way to highlight Mobileye’s ADAS/AV leadership and achieve a higher multiple akin to pure tech or auto-tech peers.
Funding Intel’s Transformation: Intel has been undergoing a major overhaul—substantial CapEx needs for advanced node development and foundry expansions put pressure on Intel’s balance sheet. Partially monetizing Mobileye offered an infusion of cash.
Operational Independence: Spinning off Mobileye would grant it the agility to pursue partnerships, respond to automotive customers more directly, and potentially use external silicon sources if needed—while still drawing on Intel’s scale where beneficial.

II. Transaction Structure

IPO Details (Late 2022)

Partial Listing: In October 2022, Mobileye (MBLY) held its initial public offering on NASDAQ, with Intel selling only about 5–10% of the total shares.
Valuation: The offering was priced below early expectations, debuting at a valuation around $17–$20 billion—less than rumors of $50+ billion floated previously.
Outcome: Despite a challenging IPO market, Mobileye managed to raise roughly $800–$900 million, with Intel retaining around 94% ownership initially, then ~88% after secondary share sales.

Retained Majority

Ownership and Control: Intel emerged holding 88% of Mobileye’s equity. Though a separate public company, Mobileye is still subject to Intel’s voting power on large decisions.
Future Stake Sales: Intel can sell additional shares in the future, or maintain the stake for longer-term strategic benefits. Market participants watch closely for any sign Intel might reduce its position further.

III. Core Rationale and Strategic Motives

Value “Unbundling”
- Mobileye as a Premium Asset: Advanced Driver Assistance and potential L2+/L3 solutions typically garner premium valuation multiples.
- Tech-Like Multiples: By exiting Intel’s overshadow, Mobileye can be appraised as a high-growth, software-rich automotive tech company.
Raising Capital for Intel
- Funding Node Development: Intel’s drive to create leading-edge processes (e.g., Intel 3, 18A) is immensely capital-intensive.
- Supporting Foundry Goals: The partial sale of Mobileye helps partially finance Intel’s push to become a major foundry for external customers.
Maintaining Strategic Ties
- Shared Resources: Mobileye could still leverage Intel’s scale if/when beneficial—like advanced packaging or certain IP blocks.
- Control and Upside: By holding the majority stake, Intel remains poised to benefit from Mobileye’s future success.

IV. Post-Spin-Off Relationship

Operational Collaboration

Silicon Sourcing: Mobileye historically sources EyeQ chips from both Intel (for some lines) and TSMC (for others). The spin-off hasn’t disrupted these arrangements drastically.
Ongoing R&D Synergy: Mobileye can tap certain Intel R&D resources, like the pursuit of specialized packaging or advanced silicon processes, if cost-effective.

Governance and Independence

Separate Management: Mobileye has a distinct executive leadership, emphasizing agility in automotive partnerships and development.
Intel’s Oversight: With ~88% ownership, Intel wields voting power for strategic moves such as large acquisitions or share issuance. Daily decisions, though, are Mobileye’s to manage.

Potential Stake Sales or Buybacks

Overhang Concerns: The market keeps an eye on whether Intel might offload shares, which can weigh on MBLY’s stock price.
Recent Reassurances: In 2024, Intel stated it “isn’t currently planning” to divest the majority stake, temporarily easing investor worries.

V. Financial and Market Impact

Immediate IPO Results

Proceeds: Intel netted over $800 million initially, then sold an additional chunk in 2023–2024, raising about $1.5 billion total. The amounts are modest relative to Intel’s $25+ billion annual CapEx but still helpful.

Valuation Changes

Initial Pricing: Mobileye’s market cap started near $17–$20 billion. After some volatility, it has fluctuated around $10–$15 billion range, influenced by ADAS rollouts and macro headwinds.
Stock Overhang: With Intel in control, some institutional investors fear a potential large block sale, which can depress valuations.

Impact on Intel’s Strategy

Small but Symbolic Funding: The partial spin-off underscores Intel’s willingness to restructure and rationalize assets.
Focus on Core CPU & Foundry: Letting Mobileye’s management chart their own path frees Intel’s C-suite to concentrate on CPU competitiveness and new foundry services.

VI. Key Lessons and Takeaways

Partial Spin-Off Pros and Cons
- Pros: Surfaces hidden value, garners capital, offers operational autonomy for the subsidiary.
- Cons: The controlling shareholder’s large stake can remain an overhang, limiting stock liquidity and possibly dampening the share price.
Timing the IPO
- Market Conditions: Launching in a weaker IPO environment constrained Mobileye’s initial valuation.
- Intel’s Capital Needs: Despite suboptimal timing, Intel chose to proceed to meet near-term funding demands.
Balancing Synergy vs. Autonomy
- Synergy: Ongoing alliances in R&D and potential manufacturing can be valuable, especially for an ADAS company reliant on advanced SoC production.
- Autonomy: Mobileye’s brand and culture flourish when free of Intel’s overshadow, better matching the automotive sector’s ecosystem approach.

VII. Conclusion and Future Outlook

Spin-Off Recap

Intel’s decision to partially spin off Mobileye stands as a strategic bet that both clarifies Mobileye’s growth story and provides Intel with a capital infusion. By retaining majority ownership, Intel retains a seat at the table for the future of automotive ADAS and advanced driverless solutions.

Post-IPO Performance

Mobileye must show it can continue winning ADAS and L2++ design wins in a competitive environment. Its partnership strategy with global OEMs and SuperVision’s success will be pivotal.
Intel may remain open to further share sales. Although Intel reaffirmed in 2024 that it won’t sell a majority stake “currently,” the possibility still looms if capital demands intensify.

The Road Ahead

For Intel: Freed from day-to-day oversight of Mobileye, it can double down on CPU roadmap execution, advanced node developments, and foundry expansions. If it eventually sells more Mobileye shares at a higher valuation, that’s a win-win.
For Mobileye: The newly public entity must scale advanced ADAS while navigating competitive threats. If it can expand its presence in L3–L4 systems or robotaxis, Mobileye could command a higher multiple akin to pure-play autonomy leaders.

In sum, Intel’s Mobileye spin-off remains an ongoing story of value unbundling: an attempt to let ADAS/autonomous technology thrive in public markets while fueling Intel’s epic transformation. Yet as with any partial IPO, the outcome hinges on synergy vs. independence dynamics—and whether Intel’s majority stake proves a valuable anchor or an overhang on Mobileye’s public market journey.

Intel Corporation (INTC) Fundamentals – Large Legacy CPU & Foundry Ambitions; PT: $25 v. $24.00

2024年12月22日2024年12月23日 Jerry NgLeave a comment

Question #1: Is This A Good Business?

Subjective vs. Objective

Subjectively: Intel remains the historical leader in x86 CPUs for PCs and Servers, operating as an IDM (Integrated Device Manufacturer) with in-house fabrication. It has strategic importance to the semiconductor industry and a strong brand (especially in data center and PC CPU markets).
Objectively (via cash flow):
- Free Cash Flow (FCF): Currently under pressure due to heavy CapEx for next-gen process nodes and foundry ambitions.
- Revenue Growth: Had been declining in 2023–24 due to share loss in servers (to AMD, ARM) and a PC market slump. Potential re-acceleration possible if the product roadmap regains competitiveness and foundry strategy monetizes.
- Profit Margin: Historical mid-50%+ gross margin is down to ~40% range. Operating margin was ~30% historically but has dipped closer to 15–20% recently due to underutilized fabs and intense R&D.
- Capital Investment: Enormous outlays ($25–$30B/year range) to achieve advanced process leadership, plus the newly minted foundry business.
- Terminal Value: If Intel reestablishes CPU leadership or effectively pivots to a profitable foundry model, the market could maintain a strong multiple. But if competition outpaces them, Intel’s large manufacturing assets could remain underutilized.

Business Value Creation Distilled to Five Factors

Organic Growth: Currently negative yoy, but potential rebound if new CPU roadmaps deliver and foundry customers ramp.
Margin Trajectory: Margins depressed near term by ramp costs; can rebound to ~25%+ operating margin if node execution improves and volumes fill fabs.
Capital Intensity (ROIC)
- Formula: ROIC=NOPAT/Invested Capital
- 2023–24 ROIC: ~2–4% (very low given heavy CapEx).
- Potential 2026+ ROIC: If volume recovers and foundry scales, could approach high single digits or ~10%.
Capital Deployment: Currently, dividend was historically a priority, but it has been cut recently. Ongoing M&A (Tower Semi) and large CapEx overshadow share buybacks.
Terminal Value Perception: If Intel regains CPU or foundry leadership, the multiple could expand. If not, it’s an aging x86 franchise with underutilized fabs.

Question #2: How Does This Business Make Money?

Revenue Model

Client Computing Group (CCG)
- What & How: Sells x86 CPUs for notebooks (~~$120–$140 ASP historically) and desktops (~~$150–$200).
- Major Customers: PC OEMs (e.g., Dell, HP, Lenovo, Apple historically—but Apple is phasing out x86).
- Contract Duration: Typically 1–3 years per platform. High switching costs for OEMs, but competition from AMD and Apple’s ARM-based silicon has grown.
- Recurring Revenue? Mostly one-time CPU sales. Some incremental over-the-air updates are possible but minimal.
Datacenter & AI Group (DCAI)
- What & How: Sells Xeon CPUs for servers (ASP historically $600–$700) to enterprises, cloud providers, and HPC.
- Major Customers: Hyperscalers (AWS, Azure, GCP), enterprise OEMs (Dell EMC, HPE, Lenovo), government HPC.
- Revenue Concentration: Cloud ~50%, Enterprise ~30%, Comms ~20%.
- Recurring Revenue? Minimal. CPU is a one-time sale. Some enterprise multi-year contracts exist but are not subscription-based.
Network & Edge (NEX)
- What & How: Supplies silicon for network infrastructure, vRAN, and edge computing.
- Major Customers: Telecom OEMs, enterprise networking.
- Recurring Revenue? Largely upfront, but longer design cycles.
Intel Foundry Services (IFS)
- What & How: A separate business to manufacture wafers for external customers. Revenue is tied to wafer volumes and utilization.
- Pricing Model: Typically cost+ or market-based foundry rates.
- Contract Duration: 3–5+ years for large engagements. Still a small portion of total revenue (<5%).
Mobileye, Programmable Solutions Group (PSG), and Others
- Small segments that provide automotive ADAS chips (Mobileye, partially IPO’d) or FPGAs (PSG) to enterprise/communication.

Customer Concentration

Heavily reliant on a few major PC OEMs and cloud providers. No single end-customer >20% of total revenue, but the top 5 might represent ~40–50%.

Recurring Revenue

Very limited on the CPU side. If Intel Foundry Services grows with multi-year wafer supply agreements, that could bring more stable recurring revenue.

Question #3: Nature of the Cost Structure

A. Granular Breakdown of R&D

For 2024E, total R&D ≈ $15B or ~25–30% of revenue (excl. Mobileye). Example sub-allocation:
- Process Node R&D (~50%): e.g., Intel 3, 20A, 18A transistor development, advanced packaging.
- CPU/GPU SoC & Architecture (30%): new CPU microarchitectures, GPU expansions, AI accelerators.
- IFS & EDA Tools (10%): foundry enablement, internal EDA flows.
- Safety & Validation (5–10%): advanced reliability, security.

B. Fixed vs. Variable Costs

Fixed: R&D, a portion of G&A, overhead for fab depreciation.
Variable: Materials (silicon wafers, chemicals), direct labor, packaging, testing.
Operating Leverage: Mixed. Fabs have large fixed depreciation, so higher wafer volumes can significantly boost GM. In 2023–24, underutilization weighs on margins.

C. Incremental Margins

Historically, each +10% revenue can yield +15–20% EBIT growth if fabs are well utilized. Right now, idle capacity reduces incremental margin.
If demand returns to fill capacity, incremental GM can jump to 60%+ on extra units.

Fixed Costs

Depreciation: A major line item (tens of billions).
R&D: ~$15B.
G&A: ~$6–$7B.

Variable Costs

BOM for each CPU or chip, plus packaging costs.
Foundry external wafers (for some designs, e.g. GPU, certain chiplets).

Operating Leverage Proxy

Under normal conditions, a +10% revenue growth might drive +15–25% EBIT improvement. Currently, negative leverage while volumes remain below the fab’s cost-recovery threshold.

Comparison with Comps

TSMC: Pure-play foundry, runs at ~50%+ operating margin, more consistent utilization.
AMD: Fabless, with lower overhead, invests heavily in design.
NVIDIA: Primarily GPU-based, outsources manufacturing, ~70% GM, leaner overhead.

Question #4: Key Drivers of the Business

Driver #1: Product Competitiveness in Server CPUs

Narrative & Debate: Intel’s new server chips (Granite Rapids, Sierra Forest on Intel 3) must regain performance/watt leadership vs. AMD Genoa/Bergamo. If Intel’s roadmap slips again, share loss will continue.
Financial Impact: Each ~5% server share shift ~$2B in revenue swing. Potential margin pivot if volumes saturate or if it remains underutilized.

Driver #2: PC TAM and Market Share

Narrative & Debate: Post-COVID slowdown decimated PC shipments from 350M to ~250M. Can Intel maintain ~70% share vs. AMD, Apple, ARM entrants?
Impact: The PC segment is ~40% of Intel’s revenue. Even partial share changes can move EPS significantly.

Driver #3: Foundry (IFS) Ramp

Narrative & Debate: Can Intel attract external customers for advanced nodes by 2026–27? Potential synergy or conflict if Intel Product competes with foundry customers.
Impact: If IFS hits >$10B by 2027 with decent margin, it’s a new growth pillar. If adoption is slow, the large idle capacity drags overall margin.

Driver #4: AI Acceleration

Narrative & Debate: AI training/inference is shifting to GPUs/accelerators. Intel invests in Gaudi (Habana) and GPU, but needs traction. CPU-based AI PC is also a new theme.
Impact: Gains in AI adjacency can add billions in the data center or client sales, or it could remain overshadowed by Nvidia’s GPU dominance.

Driver #5: Cost Cuts & Capital Efficiency

Narrative & Debate: Management commits to $8–$10B in cost savings by 2025. Also exploring Smart Capital (shell-first, government grants, Brookfield JV).
Impact: If successful, Intel can preserve FCF and dividends. If underutilization persists, negative GM leverage continues to hamper EPS.

Model Sensitivities & Debate

Where is the Debate? Execution on 18A node, the timeline for regaining server leadership, foundry traction, and the magnitude of cost discipline vs. needed R&D.
Shiny Ball Syndrome? Robotaxi, AI GPUs, and foundry expansions all overshadow the core x86 CPU cyclical reality. The real driver remains CPU competitiveness.

Question #5: Business Momentum

Recent Accelerations / Decelerations

Revenue: 2023 ~$52–$54B (-13–15% yoy). Down from $63B in 2022.
2–3 Year Stack: Down ~25% vs. 2021 peak, reflecting share loss + cyclical slump.
Potential reacceleration in 2025 if server roadmap improves and PC normalizes.

Trajectory of Key Drivers

Bull Case: Intel successfully launches Granite Rapids on schedule, stabilizes server share at 70% by 2025, and foundry wins a handful of external customers. EBIT margin climbs to high-20s, revenue recovers to $65B+ by 2026.
Base Case: Slight server share declines continue, foundry adoption is modest, cost cuts partially offset idle capacity. Revenue stagnates near $55–$60B range, with ~20% operating margin.
Bear Case: Repeated roadmap delays, foundry flops, major HPC, and enterprise shift to AMD/ARM. Revenue <$50B by 2026, margins stuck ~15%.

Expectations Gap

The market is discounting a prolonged share loss offset by partial cost improvements. The stock trades near 1.2x book value, reflecting subdued growth prospects.
Some see an upside if Intel’s new leadership revives CPU competitiveness or effectively monetizes foundry JV strategies. Others remain skeptical given repeated misexecution.

Putting It All Together
Intel’s fate hinges on server CPU competitiveness, foundry scaling, and disciplined cost management. Underutilized fabs weigh on near-term margins, but if future node and product roadmaps succeed, a margin expansion to ~25–30% is feasible. Investors are cautious, reflecting historical delays, but any positive surprise on execution or foundry traction can lead to re-rating.

Thesis: “Intel – Potential Turnaround if CPU Roadmaps Deliver, But Big If”

Price-Embedded Key-Drivers

The stock (~$24) seems to reflect partial CPU share erosion and high capital burdens.
If Intel demonstrates on-time CPU product launches with improved performance vs. AMD, we see potential multiple expansion.
Foundry success remains uncertain, but even moderate uptake helps fab utilization.

Base Case View on Key Drivers

PC TAM normalizes around 250–270M by 2025; Intel holds ~65–70% share.
Server share stabilizes near 60–65% by 2025 if new leadership swiftly addresses product execution.
Foundry revenue remains single digits ($3–$5B) by 2026.

Why the Mispricing or Misperception Exists

Past node delays degrade confidence. Some investors assume perpetual server share decline.
Street not fully crediting the cost cuts or potential synergy from government subsidies.

When the Mispricing Might Close

Could close by late 2025 if Intel hits major CPU milestones (e.g., on-time next-gen server CPU) or announces meaningful external foundry customers.
Alternatively, a spin-off or partial separation of IFS could crystallize value.

Valuation

Under Our Bull Scenario

We assume revenue recovers to $65B by 2026, operating margin rebounds to ~28%.
EPS surpasses $3.00, at ~13x multiple = mid-$40 stock.
Base Case
Stays near $55–$60B rev, mid-teens to 20% margins, EPS $2.00 by 2025–26, at ~12–13x = $24–$26 stock.
Bear Case
Fabs remain underutilized, share loss to AMD/ARM, EPS near $1.00 or less, stock ~$15–$18.

Final Thought
If Intel can revitalize its CPU roadmap and keep capital discipline, the stock could re-rate from its depressed multiples. But skepticism runs high given repeated misses. We remain neutral (Equal-Weight) with a price target of $25.

Mobileye Global Inc. (MBLY) Fundamentals – Tremendous TAM with High Operating Leverage; PT: $50 v. $17.52

2024年12月22日2024年12月23日 Jerry NgLeave a comment

Question #1: Is This A Good Business?

Subjective vs. Objective

Subjectively: Mobileye remains a global leader in Advanced Driver Assistance Systems (ADAS) and aims to extend that dominance into autonomous driving (AV). Premium brand perception and deep engineering partnerships with OEMs highlight its strategic value.
Objectively (via cash flow):
1. Free Cash Flow (FCF): Base ADAS historically supports positive FCF, though heavy R&D (~30–35% of sales) constrains free cash flow near-term.
2. Revenue Growth: Projected ~12% yoy in 2023 to $2.1B. Slower vs. prior +35% yoy in 2022, but potential re-acceleration from advanced ADAS (SuperVision) post-2025.
3. Profit Margin: ~31% EBIT margin in 2023 (non-GAAP), temporarily weighed by R&D. Margins can expand toward mid/high-30% if SuperVision/Chauffeur volumes scale.
4. Capital Investment: Physical capex $120–$150M/year. The real “capital intensity” is intangible R&D.
5. Terminal Value: If Mobileye cements L2++/L3 leadership, the market may apply a “tech-like” premium. If OEM in-sourcing or new rivals undercut them, the long-term moat could narrow.

Organic Growth, Margins, Capital Intensity, Capital Deployment, Terminal Value Perception

Organic Growth: Currently ~12% yoy (2023), with a path to 20%+ if advanced ADAS design wins accumulate.
Margin Trajectory: Could dip in the near term due to R&D needs, but potentially re-approach ~35–37% EBIT margins by 2025–2027 if volumes ramp.
Capital Intensity & ROIC:
- ROIC formula: ROIC=NOPAT/Invested Capital
- 2022–2023: ~4.5–5% ROIC, depressed by heavy up-front R&D.
- By 2025–26E: If advanced ADAS scales, ROIC could top 8–10%.
Capital Deployment: No dividends; all free cash goes to R&D, product dev, and sensor partnerships.
Terminal Value Perception: If Mobileye’s EyeQ + REM ecosystem becomes mission-critical to OEMs that cannot replicate it in-house, its terminal multiple could remain high.

Question #2: How Does This Business Make Money?

Revenue Model

Base ADAS (Level 0–2)
- What it Provides & How it Works:
  - Primarily sells EyeQ chips ($45–$50 ASP) to Tier-1 suppliers who embed them in forward-facing camera modules.
  - The EyeQ chip and Mobileye’s computer vision algorithms power features like lane-keeping, automatic emergency braking, and adaptive cruise control.
- Major Customers: Global OEMs such as GM, Ford, Nissan, Volkswagen, Toyota, Stellantis, and Hyundai/Kia via Tier-1s (e.g., Magna, ZF, Aptiv).
- Recurring Revenue?
  - Mostly per-vehicle sales. Some over-the-air software updates exist (e.g., incremental ADAS features), but not a large subscription portion.
- % of Overall Revenue: Historically ~80–85% of Mobileye’s total revenue (though shrinking as SuperVision ramps).
- Impact on Growth: Provides stable, long-duration contracts (5+ years per platform), modest yoy unit growth, and robust switching costs.
SuperVision (L2++)
- What it Provides & How it Works:
  - An integrated solution (EyeQ SoC + ECU hardware + software stack + REM map + sensor suite) enabling near “hands-off” driving on highways and certain urban settings.
- Major Customers: Zeekr, Polestar (with Volvo/Geely), future expansions to Porsche (VW Group) and potentially other global OEMs.
- Revenue & Recurrence:
  - ASP $1,200–$1,500, capturing both hardware and software fees.
  - Potential for partial recurring streams if OEMs adopt subscription-based feature unlocks or monthly driver-assist upcharges.
- % of Overall Revenue: ~10% in 2023, could climb to 25–30%+ by mid-decade if design wins materialize.
Chauffeur (L3–L4) & Drive (Robotaxis)
- What it Provides:
  - Chauffeur: ~L3/L4 advanced consumer offering with “eyes-off” capability in geofenced or highway environments.
  - Drive: Full driverless solutions (with teleoperations) for robotaxis or shuttles, e.g., test fleets in Munich, Tel Aviv, and Austin.
- Revenue & Recurrence:
  - Consumer L4 ASP could be $3–$6k.
  - Robotaxi (Drive) might have a per-mile fee model, but currently in pilot stage.
- % of Overall Revenue: Negligible now (<1%). Could be meaningful post-2025 if L4 adoption accelerates.

Recurring Revenue?

At Present:
- Road Experience Management (REM)-based cloud services and over-the-air updates represent a small slice (<5% of total revenue).
- Estimated Annual Recurring Revenue (ARR) from these mapping/cloud services is $30–$40M (~2% of total).
Future Outlook:
- As more OEMs adopt cloud-enhanced ADAS with continuous map updates, recurring revenue could scale.
- This subscription usage might eventually reach 5–10% of total revenue by 2026–2027.

Question #3: Nature of the Cost Structure

A. Granular Breakdown of R&D

For 2024E, total R&D ≈$850M(~34–35% of revenue). Estimated sub-allocation:
1. EyeQ Chip Development (30%) ≈$255M
  - EyeQ6 & EyeQ7 architecture co-design, firmware, and power optimization.
2. Software Stack & AI Models (25%) ≈$210M
  - CAIS (Compound AI System), sensor fusion, neural networks, driving policy (RSS).
3. Sensing & Mapping (20%) ≈$170M
  - Imaging radar in-house, LiDAR alliances (Innoviz), REM map labeling.
4. Validation & Regulatory (15%) ≈$127.5M
  - ISO 26262, SOTIF 21448, local safety/reg frameworks, “RSS proofs.”
5. Test Fleet & Simulation (10%) ≈$85M
  - Physical demos (e.g., VW ID.Buzz for robotaxis) plus large-scale simulation frameworks.

B. Fixed vs. Variable Costs

Fixed: R&D (~30–35% of sales), a portion of G&A (3–5%), overhead, chip design.
Variable: Bill of materials (BOM) for each ADAS/AV solution—camera modules, sensors, radar, LiDAR, ECUs.
Operating Leverage: High, because a significant chunk of R&D and overhead is fixed. Once advanced ADAS is in volume production, incremental margin is robust.

C. Incremental Margins

SuperVision vs. base ADAS:
- SuperVision gross margin: ~50% of $1,500 ASP = $750 gross profit.
- Base ADAS: $45–$50 ASP with ~80% GM = $36–$40 gross profit.
- Doubling advanced ADAS volumes can drop R&D % of sales from mid-30s to sub-25%, potentially pushing EBIT margin above 37%.

Fixed Costs

R&D (30–35% of sales): Salaries for AI/vision engineers, chip design, integrated software.
G&A (3–5%): Corporate overhead, intangible support (legal, finance).
Chip Design & Overheads: Amortized over multiple EyeQ generations, multi-year ROI window.

Variable Costs

BOM for each solution: camera modules, ECUs, sensors.
For SuperVision, the BOM cost is higher, but absolute gross profit/unit is far above base ADAS.

Operating Leverage Proxy

Each +10% revenue might yield +15–20% EBIT growth.
For instance, if SuperVision volumes surprise to the upside by +10%, revenue could see +$200–$250M, while EBIT could surpass baseline by an incremental $30–$50M because fixed R&D overhead is already “covered.”

Comparison with Comps

NVIDIA (data center & automotive): Operating leverage in the 1.4–1.7x range. Heavy fixed R&D but high margins on advanced GPUs/SoCs.
Qualcomm (handsets & auto): Moderate leverage (~1.3–1.5x), handset cyclicality overshadowing smaller auto segment.
Tesla: Very different cost structure—capex-heavy for manufacturing. FSD software line has high incremental margin, but large overhead in auto production.

Conclusion: Mobileye likely exhibits higher operating leverage than typical auto suppliers or standard semis, given:

Significant fixed R&D.
High software-value-add.
Low additional overhead once EyeQ platforms are developed.

Question #4: Key Drivers of the Business

Driver #1: Major OEM Partnerships

Narrative & Debate: Some argue OEMs may try in-house ADAS (GM Ultra Cruise, VW Cariad fiasco). Others see OEMs, under time pressure, outsourcing to Mobileye for a turnkey L2++ solution.
Financial Impact: A single major OEM awarding an L2++/SuperVision program at ~200k vehicles/yr can add $300M+ revenue. Over multiple such awards, revenue can leapfrog consensus.

Driver #2: Consumer Adoption of L2++/L3

Narrative & Debate: Will drivers pay $2k–$5k for hands-free capabilities? If yes, OEMs push more vehicles with advanced ADAS—boosting MBLY’s content. If uptake disappoints, volumes remain niche.
Impact: Rapid consumer acceptance = 2–3x advanced ADAS volumes. If uncertain, slower adoption leaves MBLY more reliant on base ADAS.

Driver #3: Competition & In-House Solutions

Narrative & Debate: Tesla might license FSD to third parties, Chinese Tier-2 (Horizon Robotics) could undercut. Some say Mobileye’s data advantage (REM, RSS) remains decisive. Others fear OEM in-sourcing.
Impact: Potential share shift in China (already tough) or large OEMs returning to MBLY after internal stumbles (like VW) could swing revenues by hundreds of millions.

Driver #4: Regulatory Push for Safety

Narrative & Debate: Certain markets (EU, Japan) might soon mandate advanced safety features or partial L3. Could create a tailwind for standardized solutions (Mobileye?). On the other hand, a patchwork of global rules might slow a uniform rollout.
Impact: Accelerated or inconsistent timelines can move MBLY’s revenue growth +/- 5–10% over a multi-year window.

Driver #5: Macro Auto Production Cycles

Narrative & Debate: If global auto volumes stagnate or a recession hits, ADAS shipments also flatten. But long-term secular shift remains intact as OEMs rarely remove ADAS from new launches.
Impact: Could dampen near-term growth, but backlog from awarded programs often provides partial insulation.

Model Sensitivities & Debate

Where is the Debate?:
- Advanced ADAS ramp timing: Bulls expect 1M+ SuperVision units by 2027. Bears see OEM in-sourcing or slower consumer adoption.
- Tesla FSD licensing: Some believe it could crowd out Mobileye. Others argue OEMs want a neutral partner.
- Chinese Underpricing: Could hamper MBLY’s share in the fastest-growing EV market. But maybe less relevant outside China.
Shiny Ball Syndrome?:
- Robotaxis (L4) still 3–5+ years away from large revenue. The near-term pivot is focusing on L2++ expansion for mass-scale ADAS.

Question #5: Business Momentum

Recent Accelerations / Decelerations

Revenue: 2023 at $2.1B (+12% yoy), a slowdown from +35% yoy in 2022. Major factor: Chinese competition, Zeekr inventory resets.
2–3 Year Stack: Up ~80% vs. 2021, showing underlying growth in base ADAS.
SuperVision: Polestar 4 launched 2H23 in China, more robust volumes in 2024 for Europe. Potential new OEM wins in 2025 could accelerate.

Trajectory of Key Drivers

Bull Case: 3–4 major L2++ awards in 2024, consumer acceptance high, surpasses 1M advanced ADAS units by 2027, ~38% EBIT margin.
Base Case: 1–2 big OEM programs, advanced ADAS ~700k units in 2027, ~33–35% EBIT margin.
Bear Case: OEM in-house solutions plus competition in China erode share, <400k advanced units in 2027, ~30% margin.

Expectations Gap

What the Stock May Be Discounting: ~20–25% revenue CAGR through 2026, partial success in advanced ADAS.
Our View: The Street underestimates how quickly large OEMs might pivot to Mobileye if their internal software struggles. This pivot could drive a re-rating if design wins land.

Putting It All Together

Mobileye offers advanced ADAS solutions with strong software IP (RSS, REM) and co-designed EyeQ chips. R&D near 35% of revenue depresses near-term margins (~31% EBIT), but operating leverage is high: doubling advanced ADAS volumes can drop R&D ratio below 25% and lift EBIT margin to 37% or higher. Key success drivers are OEM adoption of L2++/L3 systems and the competition’s inability to out-engineer Mobileye’s integrated stack.

Short-Term:

Slower growth from base ADAS in China, heavy R&D outlay, and partial wins from Polestar/Zeekr.
The Street expects ~20–25% CAGR from 2024–26. If Mobileye secures 2–3 new major SuperVision awards by 2025, upside is likely.

Long-Term:

If advanced ADAS hits 1M+ units by 2027, revenue could exceed $4B, with EBIT margins in the upper 30s.
If OEMs maintain partial in-house ADAS or Tesla FSD captures third parties, Mobileye’s ramp could underwhelm, stabilizing near $2.5–$3B revenue.

Theses

Thesis

“Mobileye’s integrated EyeQ + REM + RSS solution is the leading turnkey ADAS platform. With OEM timelines compressed, we expect multiple new SuperVision awards in 2024–25, pushing revenue beyond $4B by 2027. EBIT margin could exceed 35%, fueling strong EPS growth and a re-rating of the stock. The Street undervalues Mobileye’s data+software moat and the high switching costs for OEMs.”

Valuation

Under our Bull/Long Thesis, we assume:

Multiple New SuperVision (L2++) Awards: 3–4 major OEM signings in 2024–2025.
Faster Consumer Adoption of advanced ADAS features, driving volumes of >1M SuperVision units by 2027.
Significant Operating Leverage from spreading fixed R&D across higher revenue, lifting EBIT margins into the mid-to-upper 30s (~38%) over the next 3–4 years.

Revenue & Margin Assumptions (Bull Case)

2027 Revenue: Potentially reaching $4.0–$4.5B (vs. ~$2.1B in 2023).
Gross Margin: ~65–68%, reflecting higher SuperVision BOM but large absolute profit per unit.
EBIT Margin: ~35–38%, up from ~31% in 2023, on improved scale and partial transition to subscription REM revenue.

Illustrative Valuation Framework

Estimate 2027 EBIT
- Revenue: $4.0–$4.5B
- EBIT Margin: 35–38%
- EBIT: $1.4–$1.7B
Apply an Earnings Multiple
- Given Mobileye’s high-margin, high-R&D, tech-centric profile, a 20–25x EV/EBIT multiple can be reasonable—bridging between traditional auto suppliers (8–12x) and premium semis/tech (25–30x).
- Alternatively, use a P/E approach. In a bull scenario, 25–30x forward P/E could be justified, reflecting robust growth and technology advantage.
Illustrative EV/EBIT Approach
- Midpoint EBIT assumption: $1.55B for 2027.
- At ~25x EV/EBIT, Enterprise Value: $39B.
- If net cash builds to $3–$5B by 2027 (on strong FCF), Equity Value: $42–$44B.
- Current Market Cap: ~$14B.
- Implied Upside: ~3x the current equity value.
Implied Stock Price
- Current Stock Price: $17.52.
- Bull-Case Upside: 2.0–3.0x implies a multi-year price range of $40–$50.
Discounting to Present / Annualized Return
- Over a 3-year horizon, discounting the potential $42–$44B equity value at 10–12% suggests $26–$30B in present-value terms—still ~2.0x the current market cap of $14B.
- Per-share, that could imply a $40+ target vs. $17.52 today.
- If realized in ~3 years, this equates to an annualized total return of about 32-44% per year.

Key Risks to Bull Valuation

In-House ADAS: If OEMs successfully develop internal systems, Mobileye’s projected volumes could underperform.
Licensing of Tesla FSD: Tesla offering Full Self-Driving to other automakers might reduce Mobileye’s share.
China Competition: Continued undercutting from domestic suppliers or further share losses.

Summary

If SuperVision proliferates quickly, R&D is leveraged across larger revenue streams, and more OEMs adopt Mobileye’s turnkey L2++ system, 2027 EBIT could surpass $1.5B. At a 25x multiple, EV approaches $39–$40B. Adding net cash potentially lifts equity value to $42–$44B—around 3x the current market cap. That implies a $40–$50 stock price range over the next few years. With the stock currently at $17.52, the bull thesis points to a potential 32-44% annualized return if these upside drivers materialize.

Reading Notes from Running Money by Andy Kessler – Finding Your Own Edge & Looking for “Steam Machines”

2024年12月22日 Jerry NgLeave a comment

1. Key Takeaways & How Kessler Addressed Them

Take #1: Find and Exploit Your “Edge”

What It Means
Kessler explains that simply saying “We’re based in Silicon Valley” is not enough of a competitive advantage. Hedge fund managers often believe location alone or superficial knowledge is an edge. Instead, Kessler finds that true “edge” comes from:

Immersing yourself in the workings of companies, verifying claims face-to-face.
Understanding second- and third-order effects (like cost collapses that create entirely new markets).
Knowing something others cannot—either through direct scuttlebutt, analyzing margin structure, or reading technological breakthroughs earlier than mainstream investors.

Real Examples in the Book

Jack Nash Meeting: Kessler pitches the fund’s “tech focus” to legendary hedge fund manager Jack Nash. Nash demands: “What do you know that no one else does?” Kessler initially stumbles, saying “We’re in Silicon Valley,” but realizes that is insufficient. He leaves the meeting determined to develop deeper insight—especially about cost structures, demand triggers, and direct company visits.
Company Site Visits: He invests days shuttling around from CFO to CFO (the “four-door office”), extracting firsthand knowledge that other big-fund managers, stuck on the East Coast, do not gather. This constant “on foot” approach becomes his real information advantage.

How He Addressed It

Meeting CFOs in Person: When confronted with contradictory or vague statements, Kessler visits the competitor or calls their major customers. He might notice if a CFO literally closes the door (as with Red Brick) and treat that as a red-flag sign.
Watching Margins & Growth: Kessler keeps spreadsheets of each firm’s cost declines, new product cycles, R&D pipelines—deciding if those “technical edges” are real or hype.

Take #2: Technology “Waterfalls” & Cost Breakthroughs

What It Means
Kessler loves the metaphor of a “waterfall,” describing situations where a technology’s cost drops so steeply that entire new applications (and mass demand) open up. He compares it to how Watt’s steam engines in 1775 reduced horsepower costs, or how cheap DVD laser chips launched huge consumer markets.

Real Examples in the Book

Elantec’s Laser Diode Drivers: Kessler discovers Elantec’s low-cost laser drivers, which cost around $1–2 each, yet could be sold at a $5–10 premium. Once CD/DVD burning booms (sparked partly by Napster music swapping), the attach rate surges from near-zero to the majority of PCs. This triggers a “waterfall” in Elantec’s sales and stock. Kessler’s small stake turns into a 50x return.
Alteon’s Edge & Bandwidth Growth: Alteon built layer-4 to -7 switches for the Internet. As broadband expands (cost of bandwidth dropping each year), companies demand these advanced switches. Alteon’s “waterfall” soared so high that Nortel paid $7.8 billion for it.

How He Addressed It

Constantly Studying Cost Curves: Kessler calls it “second derivative thinking”—not just whether DVD drives are cheaper this year, but how fast they are cheapening and how that might explode usage (like Napster or home video editing).
Upside Patience: Once he identifies a “waterfall,” he invests for the big upside, tolerating day-to-day volatility.

Take #3: “Steam Engine” Moments & Historical Parallels

What It Means
Kessler devotes chunks of the book to the Industrial Revolution, analyzing Boulton & Watt’s steam engine breakthroughs. He does this to illustrate how a new engine (power source) cut costs drastically and changed entire global industries. By analogy, microprocessors (Intel), operating systems (Microsoft), and optical networks (Cisco, Alteon) do the same in the late 20th century.

Real Examples in the Book

Studying Boulton & Watt: Kessler reads about how John Wilkinson’s precise boring tool gave Watt’s steam engine 5 times more power. Boulton didn’t even sell the engine—he rented its horsepower, a new “business model.” Kessler sees parallels in licensing IP for chips or software, where margin is perpetually high if you have the “indispensable” part.
Margin Surplus vs. Trade Deficits: Kessler then extends the steam engine lesson to U.S. modern economies. Like steam power letting Britain run profitable factories, the U.S. uses chip/software IP to run massive profits, overshadowing the “on paper” trade deficits.

How He Addressed It

Applied the Lesson to Tech Stocks: He hunts for companies that, like Boulton & Watt, own the essential IP or the “monopoly vantage.” E.g., a crucial DVD chip or advanced network switch.
Focus on Profit Streams: Boulton’s model was collecting a fraction of a horse’s cost over decades. Similarly, Kessler invests in software or chip companies that sell repeated licenses or high-margin designs.

Take #4: Managing “Mania” & Forced Redemptions

What It Means
Kessler sees short-lived booms and crashes—from dot-com IPO frenzies to currency crises. When mania hits, prices become detached from fundamentals. He protects his fund by systematically sending back capital to investors each month (“forced redemptions”), ensuring they don’t get stuck if the market collapses.

Real Examples in the Book

Dot-Com Euphoria of 1999: Kessler recounts conferences where random B2B or e-commerce names soared on minimal revenue. People pitched “the next Netscape.” Once Kessler sees a room full of “peach-fuzz” junior analysts endorsing $100+ price targets, he recognizes mania.
Selling Out of MP3.com: He gets 10,000 IPO shares at $28, flips them for $60+ on day one, and never looks back. Later, the stock crashes near zero due to lawsuits. That’s how he uses mania to his advantage but keeps risk tight.

How He Addressed It

Proactive Partial Selling: Instead of waiting for a meltdown, he deliberately sells pieces of big winners, month by month. This “slow exit” technique protects him from the bubble bursting in 2000.
Closes the Fund on Schedule: He ends the fund after 5 years, even though performance is still strong. He believes returning capital ensures that mania won’t devour gains.

Take #5: Building One’s Own Research Process

What It Means
Instead of relying on sell-side analysts, Kessler conducts rigorous, first-person due diligence. He calls these face-to-face visits “soap operas,” because each company’s CFO or CEO drama reveals crucial details about product viability, sales traction, or hype.

Real Examples in the Book

Versant & Red Brick: Kessler invests in both software companies but quickly spots management red flags (closed-door CFO conversations or “over-rosy” sales pipelines). He learns the hard way to confirm real demand by talking to customers.
Fidelity to the “Four-Door Office”: Driving daily around Silicon Valley, meeting 6+ companies, collecting “pattern recognition” about CFO behaviors, product demos, and whether “the new version is shipping or delayed.”

How He Addressed It

Cross-Checking Competitors: If Company A says “Our competitor’s product is worthless,” Kessler visits the competitor to get the story. He forms an independent mosaic.
Evaluating CFO Body Language: “Door shutting,” “guarded talk,” or repeated deflections signal potential problems. If he senses trouble, he sells swiftly (like Red Brick at $26 before it plunges to $5).

2. Key Opinion

“Stay close to reality, not stock-ticker illusions. If you’re ‘long and wrong,’ find out fast. If you’re ‘long and strong,’ ride the wave but know mania can flip anytime.”

Kessler’s overarching viewpoint is that knowledge from real people—engineers, CFOs, customers—trumps fancy charts or macro rumor-mongering. His secret sauce is that he physically shows up in the CFO’s office or invests time calling customers, while many big funds skip that step.

3. Selected Quotes & Relevance

Quote #1: On Edge

“Just being in the Valley doesn’t cut it. I have to know which chip is on which board in whose product.”

Why It Matters: This is Kessler’s admission that “local advantage” must be turned into factual advantage. He invests hours in scouring product lines to see who truly wins the next generation.

Quote #2: On “Steam Engine” Parallels

“If you could invest in Boulton & Watt’s 25-year steam-engine patent in 1775, you’d ride an empire. We do the same with network routers and DSL chips.”

Why It Matters: Kessler sees the present day “monster markets” (Internet infrastructure) as analogous to 18th-century cheap power. The takeaway: intellectual property is the new horsepower.

Quote #3: On Forced Redemptions

“You can’t wait for the party to end. You must slip out early, or you’ll be picking up shards of broken glass.”

Why It Matters: His monthly forced redemption strategy is how he “slips out early,” securing big gains ahead of the 2000 crash.

4. How This Shaped His Story & Approach

He Overcame “Impostor Syndrome” by developing deeper knowledge than momentum chasers.
He Found Real Gems (Elantec, Alteon, Inktomi) by verifying the “waterfall effect” of cheap or pivotal IP.
He Learned from Failure (Versant, Red Brick) to walk away quickly when management red flags appear.
He Exited at the Right Time by forcing partial distributions before the market meltdown, saving his fund from major losses.

5. Brief Reflection on “Steam Machines” & Intellectual Property

Bullet Recap

Watt’s improvement in steam engine design parallels “shattering cost barriers” for Kessler’s DVD or networking picks.
Patents & IP (like Boulton & Watt’s 25-year exclusive) match high-margin chip and software designs: you can “rent” them repeatedly.
Vision Over Indecision: Kessler says you must foresee when a cost barrier is about to vanish so you’re in position to invest well before hype sets in.

6. Notes to Self

Do the Work: Like Kessler, talk to CFOs, customers—don’t rely on hype or quick tips.
Focus on Tectonic Shifts: If you catch a “steam engine moment” early, the upside can be enormous.
Manage Risk: Bubbles form fast—adopt or invent strategies (like forced redemption) to preserve gains.

7. Building Goodwill & Connections

Show Up, Earn Credibility: Kessler is blunt that big returns come from “knocking on doors,” not waiting for phone calls. Over time, CFOs trust him to do repeat visits.
Skeptical of “fast talkers”: The smarmy “placement agent” who demanded 50% ownership taught him that raising money can turn into losing your firm—Kessler always tries to keep personal control.

8. Dealing with Authority & Institutional Investors

Let Performance Speak: Kessler’s run-ins with suspicious institutional investors (like Jack Nash, or the big family offices) show that numbers & track record eventually matter most, not glitzy offices.
Light Touch with Regulators: He’s not a fan of heavy regulation but recognizes the meltdown in 2000 or the fiascos at LTCM, Enron taught him that “someone must watch the watchers.”

Background & Conclusion

Throughout Running Money, Andy Kessler chronicles his pivot from Wall Street analyst to co-founder of a Palo Alto tech-focused hedge fund in the mid-1990s. He systematically pursues:

Deep “soap opera” research on companies,
Spotting “steam engine” cost leaps that unleash massive new demand, and
Protecting gains when mania hits.

The overarching lesson is a combination of thoroughness, imagination about the future, and discipline in managing a portfolio. Just as steam once powered entire revolutions in manufacturing, cheap computing & bandwidth power modern wealth creation, if you position yourself ahead of that shift.

Final Thought: Kessler’s story affirms that the real “edge” is knowing exactly which disruptions and margin surpluses matter—and going in big before the crowd sees it. But also, as he repeats, “Don’t let mania keep you at the casino table.” Sometimes the best edge is stepping away with gains intact.

“Margin is everything. If you find that unstoppable IP, buy early and watch the ‘steam’ carry you—just don’t ride over the falls.”

The Self-Driving Landscape: Strategies, Data, and Future Prospects

2024年12月21日 Jerry NgLeave a comment

I. Introduction

For much of the last decade, self-driving cars were heralded as the next transformative leap in mobility. Investors poured billions into fledgling AV (autonomous vehicle) startups, while large automakers and tech giants made bold promises of widespread Level 4 or Level 5 deployment. However, the reality has proven more nuanced:

Regulatory barriers, especially local and state rules, hamper deployment speed.
Technical edge cases — from unpredictable pedestrian behavior to construction hazards — test the limits of AI.
Economics remain daunting, with sensor costs, safety drivers, and minimal near-term revenues leading to repeated cautionary tales.

Today’s self-driving efforts reveal an industry pivot: GM halted Cruise’s standalone service, Waymo and Tesla modified their go-to-market, and smaller players are seeking or deepening alliances. This article surveys major players, their technology edge, performance data, market share, and future prospects, culminating in a look at the total addressable market (TAM) for ride-hailing, consumer ADAS, and logistics.

II. Major AV Players and Their Strategies

Below is a side-by-side comparison of six leading self-driving players, highlighting their strategic focus, test fleets, financials, margins, growth outlook, technology approach, and approximate market share. All numerical references are cited or derived from estimates by industry analysts (see Sources at end).

Company	Strategic Focus	Approx. Vehicles in Test/Deployment	Financials	Margins	Growth Outlook	Tech Approach	Market Share
Waymo (Alphabet)	Licensing + select owned fleets Shifting from operating robotaxi services in a few cities to licensing the “Waymo Driver” to fleets	~700 total test & ride-hail vehicles in SF & Phoenix; plan to add ~300–500 across LA, Austin, Atlanta in 2025[^1]	Alphabet “Other Bets” lost $6B in 2023 total; Waymo invests $1B+ annually. [^2]	Negative operating margin on direct AV ops; licensing could yield ~10–15% software margins	Expanding from 3 to 5–10 US metros by 2026–2027; freight pilots could add $5B–$10B if widely adopted[^3]	High-end LiDAR + multi-sensor Robust geofencing + advanced simulation + the “Waymo Driver” autonomy stack	~1–2% of US ride-hail trips in Phoenix + SF (small overall share). Leads driverless safety miles in CA[^4].
Tesla	Consumer autonomy Selling cars w/ “Full Self-Driving” (L2) for owners, targeting a future “Tesla Network”	No separate driverless fleet; ~3M Teslas on roads globally, 400K–500K in FSD Beta in US at any given time[^5]	Auto revenue $80B+ in 2023, FSD subscription & add-ons recognized over time. [^6]	~18–25% auto gross margin (2023); FSD recognized slowly, deferring some revenue	Aims ~1M FSD Beta users by end-2025; L4 timeline uncertain, overshadowed by partial ADAS improvements	Camera-only approach Massive scale of real-world data, iterative software updates, partial autonomy (Level 2)	Dominant in L2 consumer EV autonomy; 0% robotaxi share until L4 is approved.
GM’s Cruise	Initially robotaxi, now pivoting to partial integration (Ultra Cruise) after shutting down standalone fleets	~400 in SF (suspended). ~100 test units in Phoenix, Austin prior to 2024 retreat[^7]	Burn rate $500M–$600M/quarter pre-shutdown, minimal revenue. [^8]	Deep negative. GM EV + ADAS may see margin improvements by 2025–26[^9]	Standalone robotaxi shelved; partial ADAS “Ultra Cruise” on 2025 Cadillac for ~400K vehicles by 2026	LiDAR + HD mapping Focus on advanced sensor fusion. Now refocusing on L2/3 ADAS for GM’s mainline EVs	Robotaxi share negligible after SF license revoked. GM EV ADAS share ~1–2% by 2026.
Zoox (Amazon)	Fully custom robotaxi No steering wheel, symmetrical EV for city ride-hailing	~100 custom shuttles in Las Vegas, Foster City tests. May add ~200–300 by late 2025[^10]	Private. Acquired at $1.2B (2020). R&D heavily subsidized by Amazon. [^11]	No ride revenue yet, negative margins. Could reach 15–20% if scaled to 3–5 cities	Hopes for small-scale commercial ops 2025–26. Potential Amazon synergy for deliveries/employee rides	Custom hardware Bi-directional design, dual LiDAR arrays. Highly dependent on city-by-city regulatory approvals	0% ride-hail share. Potential small pilot usage late 2025 if city regs align.
Mobileye (Intel)	ADAS to L2/3 licensing Evolving optional robotaxi pilot in Israel/Germany	~30M+ consumer vehicles use EyeQ SoC (L2/ADAS). ~100 robotaxi test units in Israel, Munich[^12]	$2B revenue in 2023, +30% YoY, mostly ADAS chips. [^13]	Mid-teens operating margin, improving with scale	Could reach $5B+ if multiple OEMs adopt L4 in 2026–28	Camera-forward + optional radar/LiDAR. “REM” crowd-sourced mapping from large user base	#1 ADAS supplier globally (~70% in camera-based). Robotaxi share small, but major B2B potential.
Baidu Apollo	China-based open AV platform Partnerships with local OEMs (NIO, Geely), plus city-level ride-hail	~500 test vehicles in Beijing, Shanghai, Shenzhen; aims for 1,000 by 2025[^14]	Baidu invests hundreds of millions/year, partial gov’t subsidies. [^15]	Negative op margin. Could see break-even if city fleets widely adopt by mid-decade	Multi-city expansions in top-tier Chinese metros by 2025; capturing ~20% ride-hail could yield $3B–$5B	Robust sensor suite + strong government support. Leverages Baidu Maps & AI for geofenced ops	Leads Chinese AV pilots (~60–70% local test share). International presence minimal.

(All data approximate, combining public disclosures and third-party analyst estimates. Figures current as of late 2024.)

III. Industry Insights and Recent Developments

Waymo’s Disengagement Progress
- Reported 0.09 disengagements per 1,000 miles in 2019 → 0.06 in 2022 → 0.05 in 2023 (via CA DMV[^4]).
- Represents a 17% drop year-over-year, ~44% lower than in 2019, signaling steadily improving reliability.
- Partnership with Uber (2023) in Phoenix (and soon Atlanta) broadened its ride-hail user base[^1], aligning with a partial pivot toward licensing the “Waymo Driver.”
GM Halting Cruise
- GM’s Q3 2024 statements cited $500M–$600M in quarterly burn, minimal revenue[^8].
- A pedestrian crash in San Francisco triggered the CA DMV to revoke permits in October 2024[^7].
- GM’s pivot leverages Cruise’s IP into “Ultra Cruise” (Level 2/3) on premium EVs by 2025[^9].
Tesla’s Camera-Only Strategy
- ~3M Teslas on roads globally, per Q2 2024 update, with ~400K–500K FSD Beta users in the US[^5].
- CEO Elon Musk reaffirmed “robotaxi network” aims, but no L4 timeline given (Tesla Q3 2024 Earnings[^6]).
- Regulators remain cautious; multiple NHTSA dockets scrutinize FSD’s real-world performance.
Zoox and Mobileye
- Zoox: ~100 custom robotaxis testing in Las Vegas and Foster City; might reach 200–300 by 2025 if expansions proceed[^10]. Amazon internal briefings note synergy with last-mile deliveries[^11].
- Mobileye: Over 30M vehicles worldwide use EyeQ SoCs for ADAS. In pilot robotaxi modes in Israel and Munich[^12][^13]. Targets partial or full L4 solutions for OEMs by 2026–2028.
Baidu Apollo (China)
- Tightly integrated with local municipalities. ~500 test vehicles across Beijing, Shanghai, Shenzhen, aiming for 1,000 by 2025[^14].
- Baidu invests hundreds of millions annually in Apollo, buttressed by some government subsidies[^15].
- Potential for large-scale expansions in Tier-1 Chinese metros.

IV. Fleet vs. Consumer Autonomy: Financial Underpinnings

Why Compare These Two Models?
In examining how self-driving cars reach the market, it becomes clear that two main business models have gained prominence: one centers on managing fleets of shared robotaxis (think Waymo or Zoox), while the other focuses on selling advanced or partial self-driving features directly to private owners (typified by Tesla, GM’s Ultra Cruise, and Mobileye-powered ADAS). Although both approaches ultimately aim to deliver safer, more efficient transport, their cost structures, technical demands, and user experiences diverge considerably. Fleet-based operators prioritize high utilization, teleoperations, and dense urban zones to amortize sensor and R&D costs, whereas consumer-oriented strategies rely on incremental ADAS adoption, one-time or subscription-based “autopilot” packages, and broader personal car ownership trends. By understanding these two paths — one built on specialized vehicles and central oversight, the other on empowering individual drivers through evolving software — we can see more clearly where the industry’s investments, regulatory negotiations, and technological breakthroughs are headed.

1. Fleet/Robotaxi Model (Waymo, Zoox, ex-Cruise)

Upsides
- High utilization potential: A single vehicle can operate ~18 hours/day, offsetting hardware costs.
- Centralized teleops: Streamlined updates, uniform maintenance, direct data access for refinement.
Downsides
- Massive CapEx: LiDAR alone can still cost $5K–$10K per vehicle, plus depot + maintenance staff.
- Teleops overhead: Teams must handle edge cases, raising ongoing payroll costs.
- Fare competitiveness: Waymo’s cost/mile hovers around $2–$3, near typical ride-hail. Achieving sub-$1.50 is pivotal for wide adoption.

2. Consumer Autonomy (Tesla, GM Ultra Cruise, Mobileye)

Upsides
- Lower overhead: OEMs pass hardware costs to buyers, monetize software via one-time or subscription fees (e.g., Tesla FSD at $12K–$15K).
- Rapid scaling: Each newly sold vehicle extends the ADAS/FSD user base, gathering real-world data.
Downsides
- Achieving robust L4 is tougher without controlling environment or ensuring driver attention.
- Liability questions, potential misuse, and uncertain regulatory acceptance for full “hands-off” driving.

Conclusion: Fleet-based attempts grant operators fuller control but demand hefty initial investment. Consumer-focused autonomy sells incremental features on personal vehicles, achieving partial automation quickly, though bridging from L2/3 to L4 remains complex.

V. Market/TAM Overview

Global Ride-Hailing
- Projected $300B by 2030, potentially $500B by 2035 if robotaxis reduce cost-per-mile to $1–$1.50 (Goldman Sachs estimate, 2023).
- Robotaxi expansions can unlock mainstream usage if local regulations allow and sensor costs decline further.
Consumer ADAS
- ~100M annual new car sales worldwide (OICA data).
- If half adopt advanced L2/3 by 2030, $25B–$40B in hardware/software revenues (McKinsey ADAS forecast, 2023).
- Mobileye, Bosch, Continental remain core Tier 1 suppliers.
Freight & Logistics
- US trucking: $700B market (ATA stats). 5–10% AV penetration = multi-billion annual rev for Waymo Via, Tesla Semi, Aurora.
- Must address corridor autonomy, charging stops, and cross-state regulation.
China
- Potentially $200B in local AV revenue by 2030 across major metros (Baidu MIIT estimates, 2023).
- Government backing fuels Baidu Apollo, WeRide, Pony.ai expansions in top-tier cities.

VI. Future Prospects and Timelines

Short Term (1–2 Years)

Waymo: Expanding Phoenix/SF driverless into LA, Austin, Atlanta by 2025, ~300–500 more vehicles (Alphabet Q2 2024 remarks[^1]).
Tesla: Potentially 1M FSD Beta users by end of 2025, though Level 4 remains unofficial (Tesla Q3 2024 call[^6]).
GM: Ultra Cruise on 2025 Cadillacs, ~400K vehicles by 2026 (GM Investor Day, Nov 2024[^9]).
Zoox: Launching limited commercial ops in Vegas or 1–2 other cities by 2025–26, possibly for Amazon employee transport (Internal briefings[^10][^11]).
Mobileye: EyeQ shipments ~15M/year by 2025, expanding L2/3 with partner OEMs (Mobileye IR deck[^12][^13]).
Baidu Apollo: Doubling to ~1,000 test vehicles across 10 Chinese metros by 2025 (Baidu announcements[^14][^15]).

Mid Term (3–5 Years)

Waymo operates in 5–10 US metros, possibly commercializing freight corridors.
Tesla might see advanced L3 highway autonomy, but city L4 is TBD.
GM leverages Ultra Cruise across mainstream EV lines.
Baidu gains share in more Chinese urban zones, possibly capturing 20–30% local ride-hail in pilot districts.

Long Term (5–10 Years)

Cost-per-mile for robotaxis could dip below $1.50 if sensor and computing costs keep falling.
50%+ of new vehicles shipping with advanced L2/3.
Freight emerges as a robust AV revenue segment for players like Waymo, Tesla.
Potential new entrants (e.g., Apple) or major M&A reshuffling the AV landscape.

VII. Conclusion

The self-driving sector stands at a pivotal inflection. Early hype for instant Level 4 rollouts has moderated, with GM’s Cruise scaling back and Waymo pivoting to licensing alongside operating partial fleets. Tesla focuses on L2 consumer autonomy, hoping to eventually unlock L4, while Mobileye thrives as a top ADAS supplier. Zoox designs custom robotaxis under Amazon’s umbrella, and Baidu Apollo powers rapid expansions across Chinese metros.

No single approach guarantees quick profitability; rather, two main business models dominate: (1) fleet-based robotaxis and (2) consumer autonomy in private cars. The TAM—from global ride-hailing ($300B+), ADAS ($25B–$40B), and trucking ($700B US market)—illustrates enormous long-range upside, tempered by slower rollouts, city-by-city regulations, and cost considerations. Over the next 2–5 years, watch for deeper partnerships (Waymo–Uber, GM–Ultra Cruise, Mobileye–OEM deals), sensor cost declines, and region-specific approvals shaping where and how fully driverless cars hit the road. Though the timelines have stretched, the underlying promise of safer, more efficient automated mobility continues to drive intense innovation—and major investments—through this ongoing evolution.

Sources

Waymo-Uber Partnership & Future Plans
- Alphabet Q2 2024 Conference Call (Alphabet Investor Relations)
- Official Waymo Blog (2023) announcing expansion to additional cities
- Uber Press Release (2023) re. integrated hail options
Alphabet’s “Other Bets” Financials
- Alphabet Q3 2023 10-Q (SEC Filing)
- Investor Day remarks regarding Waymo capital expenditures
Waymo Freight & Potential $5B–$10B
- Analyst estimates from Morgan Stanley (2023) and RBC Capital (2022)
California DMV Disengagement Data
- 2019–2023 annual AV Disengagement Reports on the CA DMV website
Tesla FSD Beta Participation, 3M Vehicles
- Tesla Q2 2024 Earnings Call
- NHTSA safety docket referencing Tesla FSD rollout figures
Tesla’s Robotaxi Timeline
- Elon Musk statements, Tesla Q3 2024 Earnings (transcripts on Tesla IR website)
GM’s Cruise
- GM public statements (Oct 2024) re. SF crash, CA DMV permit revocation
- GM Q2 2024 earnings slides referencing Cruise Opex
Cruise Burn Rate
- GM’s 2023 Investor Presentation, detailing Cruise’s monthly expenses
Ultra Cruise & ~400K vehicles
- GM Investor Day (Nov 2024), CEO Mary Barra commentary on partial autonomy
Zoox: ~100 Shuttles

Internal Amazon briefings (late 2023), summarized by Bloomberg (2024)

Zoox Amazon Subsidy

Amazon 2024 planning doc, partial references in WSJ (2023 Q4 coverage)

Mobileye EyeQ SoC & 30M Vehicles

Mobileye IR deck, Q2 2024, indicating total EyeQ shipments
Intel SEC Filings (2023)

Mobileye $2B 2023 Revenue

Mobileye Q2 2023 Earnings Release

Baidu Apollo 500 test vehicles

Baidu Apollo official announcements (2023)
Chinese Ministry of Industry & IT (MIIT) policy statements

Baidu AV Investments

Baidu Quarterly Earnings (Q1 2024), CEO Robin Li commentary
Local gov’t subsidy references in multiple Baidu news releases (2023)

(All data approximate as of late 2024; private company figures are estimates or gleaned from partial disclosures.)