AI Has Become Major Capital Formation Cycle, Says Altimeter

Key Concepts

• Capital Formation Cycle: The shift of AI from a mere technology product cycle to a massive engine for capital allocation and infrastructure investment.
• Capex Divide: The strategic distinction between companies receiving capital expenditure (compute, energy, memory, networking) and those spending it (AI labs/model builders).
• Scaling Laws: The empirical observation that increasing compute, data, and algorithmic resources leads to higher levels of intelligence.
• Switzerland Strategy: A business model (e.g., xAI) that acts as a neutral provider of infrastructure (compute) while simultaneously developing its own models and consumer products.
• Picks and Shovels: The foundational hardware and infrastructure layer (Nvidia, SK Hynix, etc.) essential for AI development.

---

1. The State of the Capital Markets

The current market is defined by a massive "capital formation cycle." There is a clear bifurcation in the industry:

• The Receivers (Infrastructure): Companies providing the "picks and shovels"—compute, memory, optics, and networking—are seeing massive capital inflows. Examples include Nvidia’s $80 billion stock buyback and SK Hynix’s $8 billion capital return program.
• The Spenders (Labs): Companies like OpenAI, Anthropic, and Alphabet are deploying massive amounts of capital to train frontier models. Alphabet recently upsized an $84 billion equity offering, while OpenAI has secured historic funding rounds.

2. Strategic Business Models and "The Switzerland Approach"

The speaker argues against the "binary" view that companies must be either consumer-focused or enterprise-focused. Instead, successful firms are becoming multifaceted:

• OpenAI: Operates across consumer (nearly 1 billion ChatGPT users), enterprise (5 million Codex users), and hardware/robotics.
• xAI: Described as the "Switzerland" of AI. By building data centers and selling compute while simultaneously developing its own models and coding products, it hedges against the risk of model commoditization.
• SpaceX: Beyond its core launch and Starlink business, it is evolving into a "hyperscaler" by selling compute, effectively positioning itself as a major player in the AI infrastructure layer.

3. The "Long/Short" Framework for Innovation

The speaker utilizes a "Long/Short" mental model to evaluate industry trends, often used in his Stanford classes:

• The "Long" Side: Industry leaders are overwhelmingly bullish on compute and energy. Energy is identified as the primary bottleneck for scaling AI at the global level.
• The "Short" Side: The consensus is bearish on incumbents that fail to innovate. The speaker notes that businesses vulnerable to "scaling laws"—those that cannot adapt as intelligence levels rise—are in a precarious position.

4. Market Outlook and Liquidity

Regarding the "wall of public offerings" (IPOs) expected in the AI sector, the speaker remains optimistic:

• Cyclicality: IPOs are inherently cyclical, and the market has been waiting for access to the AI "super cycle."
• Demand: The speaker believes there is significant pent-up demand for these companies. He points to the rapid advancement in capabilities—specifically the jump in coding model performance (e.g., Opus 4.5)—as evidence that the technology is ready for broader market participation.

5. Synthesis and Conclusion

The core takeaway is that the AI industry is currently defined by a massive infrastructure build-out. Investors should focus on the Capex Divide: identifying whether a company is a beneficiary of the massive spending on compute and energy or if it is a consumer of that capital. The most resilient companies are those that are "holistic"—integrating across consumer, enterprise, and infrastructure layers—and those that are not threatened by the inevitable rise in intelligence dictated by scaling laws. As the speaker notes, "If you're a business that is hurt by this empirical observation that intelligence is going to go up, that's a tough place to be."