Can AI answer the $3 trillion question?

Three years ago, David Cahn, a partner at Sequoia, made headlines by quantifying the financial implications of Silicon Valley's substantial investments in AI infrastructure. Fast forward to 2023, as Nvidia reported an impressive annual revenue of $50 billion from GPUs, Cahn revisited his calculations. By factoring in operational costs for data centers and expected margins, he estimated that a staggering $200 billion in revenue would be necessary to recoup initial investments. As the AI landscape continues to evolve, Cahn has revised his projections for AI infrastructure spending, predicting it will hit $1.5 trillion by 2026. This figure comes with an alarming conclusion: the AI sector must generate $3 trillion to validate its extensive expenditures on chips and data centers. He cautions that this estimate may be conservative, given the escalating costs associated with memory and the increasing reliance on specialty chips designed for specific tasks. Meanwhile, companies like Anthropic are reportedly nearing $60 billion in annual recurring revenue (ARR), while OpenAI is believed to have generated $20 billion ARR as of late 2025, though it initially reported $13 billion. Despite these successes, a significant revenue gap remains to be filled. Torsten Slok, Apollo's chief economist, highlights the expectations of major tech players—Google, Meta, Microsoft, and Amazon—who foresee a surge in free cash flow by 2028 as a payoff for their AI investments. However, there is an underlying concern: what if these expectations are not met? Slok points out a growing trend of organizations opting for more affordable open-weight models, often sourced from China, rather than those developed by leading labs, which could impact revenue streams. In response to user concerns about costs, OpenAI's latest model boasts a 54% increase in token efficiency for coding tasks. While this development benefits users, it poses challenges for companies focused on token generation if overall usage does not significantly rise. Slok warns that if hyperscalers fail to achieve their cash flow objectives, the repercussions could ripple through the economy. He suggests that a slower-than-anticipated return on investment could not only affect the AI sector but might also trigger a recession and a decline in the S&P 500. This evolving landscape serves as a critical reminder for stakeholders to monitor the financial dynamics as they navigate the complexities of AI-driven initiatives.