Howard Marks asks Claude to grade AI predictability, and frontier labs score 2 out of 10

Marks, who has 'absolutely no knowledge' of AI, outsourced technical explanations to Claude and focused on investment judgment. The chatbot rated its own maker as close to unknowable, assigning a 2 out of 10 score to frontier AI labs, including OpenAI and Anthropic.

Claude’s predictability map placed the overall AI technology at 3 out of 10, with the big-picture direction scoring between 6 and 7. Speed of adoption, profitability, and which companies win all received a 2 out of 10, reflecting deep uncertainty in these areas.

The memo breaks down the AI stack further, with chips scoring 4 out of 10 and hyperscalers 5 out of 10. Leased data centers fared best, scoring between 5 and 6 out of 10, as their revenues are tied to contracts and existing businesses.

Debt amplifies risk

Marks highlights projections of trillions of dollars in AI infrastructure spending and cites Goldman Sachs data showing nearly $600 billion in AI-related debt issued globally in 2026. Equity investors can absorb a bad guess and move on, but lenders underwriting a 2-out-of-10 business face a fixed upside and a less fixed downside.

The memo points out that frontier labs emerged quickly, require enormous capital, and lack clear competitive advantages. Their valuations depend on outcomes that are close to guesswork, making them the least predictable segment of the AI trade.

Long-term view

Marks doesn’t stop at caution, recalling a 1979 Warren Buffett observation that uncertainty can work in favor of the long-term investor. The memo references earlier exchanges with Claude in early 2026 and an AI-focused memo from December 2025, showing Marks has been tracking the topic for months.

For investors, leased data centers and hyperscalers sit in the middle of Claude’s scale, offering more stable revenue streams. Chips and frontier labs remain at the bottom, with their futures tied to outcomes that are difficult to predict.