Can AI Justify a $3 Trillion Bill?

A Number Reshaping the Industry

Three years ago, David Cahn, a partner at Sequoia Capital, ran careful calculations that revealed the scale of the bet Silicon Valley companies were making on artificial intelligence. Starting in 2023 from Nvidia's GPU revenue figures, Cahn estimated that the industry would need to generate $200 billion in revenue to recoup the capital spent on infrastructure. It was both a warning and an explicit call to entrepreneurs to build products and services capable of leveraging that massive infrastructure and covering its costs.

Today, three years into an era of explosive expansion, Cahn returns with a new and far more sobering estimate: he believes spending on AI infrastructure will reach $1.5 trillion in 2026 alone, and that the industry as a whole must generate at least $3 trillion in revenue to justify the accumulated expenditure. He himself acknowledges that figure may be conservative, as memory costs escalate and specialized inference chips proliferate, continuously raising the financial break-even threshold.

The Gap Between Spending and Returns

Meanwhile, the current revenues of the sector's leading companies remain far from reassuring for investors. Anthropic's annualized recurring revenue is estimated at around $60 billion, while OpenAI reported revenues exceeding $13 billion during 2025, with projections to reach $20 billion by year's end. These are striking figures, yet they remain modest relative to the scale of the gap that needs to be bridged.

Despite these challenges, analysts note that the tech giants — Google, Meta, Microsoft, and Amazon — are anticipating a significant jump in their free cash flows by 2028, predicated on their massive AI investments beginning to bear fruit. However, these projections carry a considerable degree of risk.

Economic Concerns That Extend Beyond the Sector

Torsten Slok, Chief Economist at Apollo Global Management, identifies a number of risks that could disrupt this optimistic picture, most notably:

• A rapid shift by many enterprises toward cheaper open-weight models — particularly Chinese models — in place of offerings from leading labs.
• A continuous decline in token prices, with OpenAI CEO Sam Altman announcing that the company's latest models are 54% more efficient on coding tasks.
• The possibility that this efficiency may not translate into a meaningful increase in overall usage, thereby compressing profit margins for AI infrastructure companies.

Growing efficiency is a double-edged sword: while it lowers costs for users and facilitates the spread of AI agents, it simultaneously threatens to erode the returns of companies whose business models are built on selling vast quantities of compute tokens.

A Risk to the Broader Economy

Slok goes beyond sector-specific concerns, warning that a failure by major tech companies to meet their financial targets could trigger wide-ranging macroeconomic consequences. Given the concentration of market index weight in a small number of dominant companies, disappointment in their results would not be confined to the tech sector — it could push the U.S. economy toward recession and expose the S&P 500 to a sharp correction.

This reality confronts investors and decision-makers with a fundamental question that cannot be deferred: will the real-world applications of artificial intelligence prove sufficient to justify these enormous bets? Or will the gap between ambition and returns continue to widen until the moment of truth arrives at the end of this decade?

✦ بقلم فريق دروب أيديا