Optimists vs Pessimists

The AI trade can justify very large valuations if compute demand, monetization and productivity gains continue, but its investment case is highly sensitive to margins, chip depreciation and the cost of financing capacity.

• AI exposure reaches global growth, equities, rates, credit, commodities, infrastructure and real estate, so a loss of confidence could destroy trillions of dollars of portfolio value.
• Anthropic is expected to IPO at a $2tn valuation; its latest filing reported $65bn of annual recurring revenue and a small operating profit, while its current compute base is close to 3GW.
• At a 30x earnings multiple, a $2tn valuation requires about $70bn of annual profit; CFM estimates that roughly 15GW of installed compute could support this under its baseline assumptions, including five-year depreciation and a 30% Anthropic margin.
• Compute remains scarce and expensive: Anthropic reportedly pays SpaceX $50m per year per megawatt for inference capacity, while Anthropic and OpenAI are expected to secure half of a possible 100GW of incremental compute in 2028 under Dylan Patel's bull case.
• The funding model is evolving as labs issue debt at roughly 8%, Nvidia-backed securitization attracts institutional interest, and CME lists compute futures; this could create a scalable asset class around GPU capacity.
• The economics remain vulnerable to competition, open-source models, lower margins and faster depreciation, although software improvements such as OpenAI's reported 2x increase in token throughput and Meta's AI-driven advertising gains could extend chip lives and raise returns.

Takeaway: Investors should treat AI as a capital-intensive theme whose upside depends on sustained compute demand and improving productivity, while testing valuations against weaker margins, faster depreciation, infrastructure constraints and the possibility that scientific breakthroughs arrive more slowly than expected.