Answers · Views and frameworks

How long can Big Tech keep up its AI capex, and do the returns cover it?

Big Tech's AI capex is still accelerating, and consensus points higher again in 2027. Returns currently exceed the cost of capital, but free cash flow is turning negative and more of the money is borrowed. FutureX Capital watches one thing: when capex growth slows.

On current guidance and consensus, spending rises again in 2027. The four largest spenders' 2026 capex guidance sums to about $730 billion at the midpoints, nearly double the roughly $410 billion of 2025. Consensus for 2027 is $920 billion. a16z, citing Vanguard, puts 2026 consensus for the hyperscalers including Oracle at $777 billion, and above $1 trillion a year from 2027. The cost lands on cash flow: a16z expects the steep drop in hyperscaler free cash flow to last until about 2028, with credit markets filling the gap. New investment-grade bonds and SPV financing for the hyperscalers and Nvidia top $300 billion in 2026 (chart reading).

In the late-July 2026 earnings week, Alphabet reported Q2 capex of $44.9 billion, about double a year earlier. Its free cash flow was minus $5.9 billion, the first negative quarter since it listed in 2004. Amazon spent $54.2 billion in the quarter and raised full-year guidance to about $220 billion, with CEO Jassy citing higher memory prices (Per company disclosure).

Returns cover it for now; the risk sits in where they come from. a16z, citing BofA, puts hyperscaler return on invested capital at about 20%, above a weighted average cost of capital of about 10% and a cost of debt of about 4%. The market rewards only spending backed by contracts. Microsoft's commercial remaining performance obligations reached $678 billion and its shares rose about 8% after hours. Meta's quarterly free cash flow was only $784 million and its shares closed down about 8% the next day. FutureX Capital's view: the bubble sits in the middle layer that must keep refinancing. Model-layer revenue is an unreliable denominator, so it watches the second derivative of capex; the turn comes when growth slows. This is a market view, not investment advice.

One source of those returns is model companies buying compute. Anthropic's unpublished draft prospectus, as seen by Reuters, shows about ten years of compute and infrastructure commitments of at least about $518 billion, roughly 80% non-cancellable, against 2025 revenue of nearly $4.6 billion (Reported). On October 8 news broke that OpenAI's annualized revenue was about $50 billion, below the $68 billion that had circulated. Oracle fell 5.48% and CoreWeave 7.77% that day (Public market data).

Report · Big Tech's AI Capex Heads to $730 Billion, but Markets Reward Only Contracted Spending →Report · a16z: Tech Drives About 76% of S&P 500 Profit Growth, Hardly a Bubble →Report · Compute's Bottleneck Moves to Memory: Micron Earns $37.7 Billion in One Quarter →

Related answers

Updated 2026-11-02. Facts and examples come from FutureX Capital's published research, public talks, and public reporting, graded per our site-wide standard (verified / reported / per company disclosure). Nothing here is investment advice or an offer.