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

By the Numbers

Despite an estimated $30–40 billion in enterprise generative AI spending, MIT Project NANDA's July 2025 study found 95% of enterprise generative AI pilots produced no measurable P&L return — the sharpest documented gap between capital deployed and financial return in the current cycle.

This is the quantitative anatomy of the generative AI cycle in 2026: several hundred billion dollars a year in hyperscaler and vendor capital expenditure, near-universal enterprise pilot activity, and a documented gap between adoption and measurable financial return. The figures span three distinct layers — infrastructure capex, foundation-model company financials, and enterprise-side ROI/failure surveys — that are frequently conflated in vendor narrative but measure fundamentally different things.

The generative AI capital story and the generative AI returns story are two different datasets measuring two different populations, and conflating them is the single most common analytical error in coverage of this cycle. The capital side is dominated by hyperscaler infrastructure spending, which has scaled from a few hundred billion dollars to a level Goldman Sachs Research's consensus estimate put at $527 billion for 2026 as of the third-quarter 2025 earnings season, up from $465 billion at the start of that earnings season. Multiple trade outlets tracking the same underlying earnings calls converged on figures in the $600–700 billion range for combined 2026 capex across Amazon, Alphabet, Meta, Microsoft, and Oracle, with one tracker citing the Big Four's plans at up to $630 billion, a 62% increase from the $388 billion in 2025 spending it cites. Bloomberg Intelligence's June 2026 Generative AI Outlook separately put hyperscaler capex approaching $750 billion in 2026 alone. These are all analyst estimates and trade-press aggregations of company guidance, not audited actuals, and the spread between $527 billion and $750 billion for the same year illustrates how much the figure depends on which companies and which spending categories (AI-specific versus total capex) are counted. What is not in dispute is the trajectory: Goldman Sachs Research noted consensus capex estimates have proven too low for two years running, with actual growth exceeding 50% in both 2024 and 2025 against roughly 20% consensus forecasts at the start of each year.

Nvidia's own reported financials anchor the demand side of this capex wave in verifiable, audited terms. Nvidia's data center revenue for its fiscal 2026 (ended January 25, 2026) reached $193.7 billion, up 68% year-over-year, on total fiscal 2026 revenue of $215.9 billion, up 65%. The pace continued into fiscal 2027: first-quarter data center revenue hit a record $75.2 billion (up 92% year-over-year), and second-quarter data center revenue reached a record $89.0 billion, up 117% year-over-year and up 18% sequentially, per Nvidia's own 8-K filings. This is capability-and-deployment-confirmed revenue — actual chip shipments booked as revenue — not a market-size projection, and it is the clearest evidence that infrastructure capex is translating into real chip demand, whatever the eventual return on that infrastructure for the buyers.

On the foundation-model company side, the two leading labs show diverging revenue and valuation trajectories, both reported through funding-round and press disclosures rather than audited public filings. OpenAI closed a $122 billion funding round on March 31, 2026 at an $852 billion post-money valuation, with an annualized revenue run rate reported near $24–25 billion in February 2026. Anthropic overtook OpenAI on valuation with a $65 billion Series H closed May 28, 2026 at a $965 billion post-money valuation, and its revenue run rate was reported at $47 billion by May 2026, up from a $10 billion annual figure the prior year — with Sacra's estimate showing growth from $9 billion at year-end 2025 to $47 billion by May 2026. Both companies are pre-IPO; Anthropic confidentially filed a draft S-1 on June 1, 2026 with a tentatively planned October 2026 listing. These are late-stage private valuations set by funding rounds, not public-market prices, and revenue run-rates (annualizing a single month or quarter) are not the same measurement as trailing twelve-month or audited annual revenue — a distinction that matters given how fast both figures are moving.

The enterprise-adoption and return-on-investment layer is where the sourced data most directly contradicts the capital-deployment narrative. MIT's Project NANDA published its July 2025 report, The GenAI Divide: State of AI in Business 2025, based on 52 structured executive interviews, a survey of 153 leaders, and analysis of more than 300 public AI deployments (a later account from the report's own promotional framing cites 150 leader interviews and 350 employee surveys — the two citations conflict on exact sample size, a discrepancy worth flagging rather than resolving). The study found that despite an estimated $30–40 billion in enterprise investment, 95% of generative AI pilots produced no measurable P&L return, while just 5% of integrated pilots were extracting significant value. The study is explicitly preliminary, not peer-reviewed, and has drawn criticism for its short measurement window (pilots were assessed roughly six months post-deployment) and its narrow definition of P&L impact — a pilot that improved worker speed without moving a financial line item still counted as a 'failure' under this methodology. Independently, S&P Global Market Intelligence's 2025 Enterprise AI Survey, covering over 1,000 enterprises across North America and Europe, found the share of companies abandoning most of their AI initiatives had jumped to 42%, up from 17% the prior year, with the average organization scrapping 46% of AI proofs of concept before they reached production. Gartner separately predicted in July 2024 that at least 30% of generative AI projects would be abandoned after proof of concept by end of 2025, and its research cited elsewhere in this cycle found the average company invested $1.9 million in GenAI projects in 2024, with less than 30% of CEOs satisfied with the return. These three independent survey organizations converge on the same qualitative finding — high adoption paired with low financial transformation — even though their exact percentages differ, which is the strongest evidentiary signal in this entire dataset precisely because the sources are methodologically independent of one another and of MIT NANDA.

Adoption breadth itself is not in question and is the one metric category showing unambiguous, converging growth across sources. A 2026 Plug and Play enterprise survey (skewed toward Fortune 500 and Forbes Global 2000 respondents) found 74% of the world's largest enterprises run at least one AI solution in production and 93% are piloting or further along, while half of production-stage companies still cannot consistently measure ROI — adoption and measurement capability are advancing on different timelines. On the agentic AI subset specifically, Gartner's Q1 2026 enterprise survey found 80% of respondents report at least one production application now embeds an AI agent, up from a 33% share two years earlier, while separate 2026 data from Forrester and Anaconda found 88% of agent pilots fail to graduate to production — again illustrating the divide between deployment breadth and pilot-to-production conversion. Market-size forecasts for where this all nets out vary by an order of magnitude depending on methodology and scope: Bloomberg Intelligence's June 2026 outlook projects a $2.3 trillion generative AI market by 2032 (up from its March 2025 forecast of $1.8 trillion), representing a projected 22% of total technology spending, while MarketsandMarkets projects $1,658.97 billion by 2033 from a $185.45 billion 2026 base, and Statista's broader AI (not generative-AI-specific) market figure was around $255 billion in 2025 rising to a projected $1,218 billion by 2030. These are vendor and market-research forecasts, not measured outcomes, and the wide dispersion across firms using different scope definitions (infrastructure-only versus infrastructure-plus-software-plus-advertising) means none should be treated as more than directional.

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

August 31, 2026

August 31, 2026

August 31, 2026