TheModelDesk

David Cahn, a partner at Sequoia, has updated his financial projections regarding the massive investments in AI infrastructure. While his 2023 analysis estimated a need for $200 billion in revenue to justify early spending, his 2026 projection sees infrastructure outlays reaching $1.5 trillion. To offset these costs, the AI industry would need to generate $3 trillion in revenue.

Cahn suggests this figure may be conservative, as the cost of construction and memory, along with the shift toward inference-specific chips, has increased the required revenue per gigawatt of capital expenditure. Current earnings show a significant gap to reach this target; for instance, Anthropic is estimated to have reached $60 billion in annual recurring revenue (ARR), while OpenAI reported $13 billion in 2025 and claimed $20 billion ARR in late 2025.

Torsten Slok, chief economist at Apollo, observes that hyperscalers like Amazon, Google, Meta, and Microsoft expect a major surge in free-cash flow by 2028. However, Slok warns that this payoff is at risk. The trend toward cheaper open-weight models—particularly those from China—and a general decline in token prices could undermine the revenue goals of frontier labs. Even efficiency gains, such as OpenAI's latest model being 54% more token-efficient for coding, may reduce the total spending required from users, further complicating the return on investment.