Make this analysis yours. Add it to a free WorldbyFlow workbench to run follow-ups, ask questions, and re-check it as events move.

By the Numbers

Four US hyperscalers plan ~$725 billion in combined 2026 AI capex (Financial Times data via Tom's Hardware, April 2026) against just 11% of S&P 500 firms showing deep AI integration in their 10-K filings for 2025 (MIT FutureTech/Carnegie Mellon, arXiv, July 2026) — a capital-to-deployment ratio with no historical precedent.

This compiles the load-bearing figures on both sides of enterprise AI's central 2026 tension: hyperscaler capital expenditure racing toward $700-750 billion annually against enterprise-level evidence that deep, value-generating AI integration remains concentrated in a small minority of large firms. The numbers show capacity being built years ahead of proven demand.

The supply side of this gap is the more measurable of the two. American technology companies are executing what one Tier 1 source calls history's largest investment surge: America's biggest technology companies plan to spend $900bn on artificial intelligence infrastructure in 2025, rising to $1.4trn by 2027 in history's largest investment surge (The Economist, July 28, 2026). Company-reported guidance compiled from first-quarter 2026 earnings calls puts the four largest hyperscalers — Google, Amazon, Microsoft, and Meta — at $725 billion on capex in 2026, up 77% from last year's record $410 billion, according to first-quarter earnings compiled by the Financial Times. Credit analysts have since revised that figure upward: analysts raised estimates to ~$750 billion for the top 5 hyperscalers—a massive 67% YoY increase and the third consecutive year of 60%+ growth driven by AI infrastructure investments (CreditSights). Looking further out, Goldman Sachs projects a combined $5.3 trillion of capex spending for the four largest hyperscalers — Meta, Microsoft, Amazon, and Alphabet — from fiscal year 2025 to fiscal year 2030, up from $4.5 trillion prior to first-quarter earnings. At the market-wide level rather than just hyperscaler capex, Gartner's most recent estimate has worldwide spending on AI forecast to total $2.59 trillion in 2026, a 47% increase year-over-year, a figure Gartner has revised upward twice in 2026 (from an initial $2.52 trillion estimate in January).

The demand side — actual production deployment and value capture — tells a starkly different story, and the discrepancy between adoption surveys and rigorous accounting-based measures is itself a critical data point. The most methodologically credible figure comes from a joint MIT FutureTech/Carnegie Mellon study using SEC 10-K disclosures rather than self-reported surveys, reasoning that securities laws prohibit companies from making materially false or misleading statements, which the researchers used to build a measure of deep AI adoption distinct from AI hype. That study found in 2025, 11% of S&P 500 enterprises had AI deeply integrated into their business processes, and a further 10% were using AI in the production of goods and delivery of services — meaning 21% combined have reached production-level deployment or above, but only about half of that group has achieved full operational integration. This is up sharply from a low base: this represents a quadrupling of integration since 2022, when deep integration stood at just 5%. Adoption is also heavily concentrated: technology companies accounted for roughly two-thirds of firms that had reached deep AI integration by 2025, with 62% of technology firms scoring at the two highest adoption levels, compared with much lower rates elsewhere. Crucially, the 10-K-based measure found no clean payoff yet at the balance-sheet level: the underlying paper notes regressing adoption on financial outcome measures recovers 'J-curve' effects among firms achieving deep integration in business processes on profitability but with little impact on capex and productivity.

Survey-based measures paint a more optimistic adoption picture but converge on the same value-capture problem. The Census Bureau's business survey — a much broader, self-reported measure covering the whole economy rather than just large caps — found 19.8% of U.S. enterprises had used AI in any business function in the prior two weeks, with 23% expecting to be doing so within six months as of late April 2026. A separate Plug and Play pulse survey of Fortune 500 and Forbes Global 2000 companies, cited by Forbes, found 74% of the world's largest companies now run AI in production, yet half still can't prove it delivers business value. That ROI-measurement problem intensifies at the most common deployment stage: among companies running AI in a single function, the earliest and simplest stage of production, three-quarters say ROI is too early to measure or is not tracked at all. A KPMG survey of AI agent deployments found a similar cost-discipline problem farther along the maturity curve, reporting established ROI of only 7% against continued planned spending of roughly $188 million across the surveyed leader cohort, with nearly half of respondents cutting agent rollouts once costs outran value. MIT's earlier and more widely cited Project NANDA study, using a different qualitative methodology, found an even starker split: based on 52 executive interviews, surveys of 153 leaders, and analysis of 300 public AI deployments, 95% of pilots delivered no measurable P&L impact, with only 5% of integrated systems creating significant value (July 2025).

Reconciling these figures requires care because they measure fundamentally different things at different points in the funnel — a discipline several of the sourced analyses explicitly flag. As one industry breakdown notes when comparing the major AI-spending trackers, Gartner's total measures procurement spend across the full AI stack, IDC's infrastructure tracker counts only AI-optimized servers and storage, and Stanford's figure tracks capital flowing into AI companies — none of them are additive or interchangeable. Similarly, the 21-percentage-point gap between the Census Bureau's broad usage figure and MIT's narrower deep-integration figure reflects both definitional differences and the fact that large-cap firms are ahead of the broader business population. What is consistent across every methodology — accounting-based (MIT/CMU 10-K analysis), survey-based (Plug and Play, Census BTOS), and case-study-based (MIT NANDA) — is that capital deployment has scaled by an order of magnitude faster than verified production value capture, and that the companies capturing measurable returns remain a minority even among large enterprises three years after ChatGPT's public release triggered the current investment cycle.

The claims behind this analysis, each with its verification status — including what is contested, unverified, or could not be established.

August 6, 2026

August 9, 2026

August 9, 2026