Companies Spent $2.6 Trillion on AI in 2026. MIT Says 95% of It Produced Nothing.

Budgets are still climbing. Boards are still approving. But the people actually measuring return on that spend are finding almost none of it shows up on the P&L.

Global enterprise AI spending hit an estimated $2.59 trillion in 2026, according to Gartner — a 47% jump from the year before, and one of the fastest capital-allocation shifts in the history of enterprise technology spend. By any normal measure of corporate technology adoption, that's an extraordinary number, comparable in scale to entire national GDPs. It's also, according to the people who actually studied where the money went, mostly not working.

MIT's NANDA initiative, which tracked enterprise AI pilots across industries, found that 95% of them delivered zero measurable impact on profit and loss. Not "modest" impact. Zero. Read that number slowly, because it's easy to skim past: nineteen out of every twenty AI pilots studied produced no detectable financial return at all.

That is not a small-sample fluke or a single skeptical study designed to generate headlines. It rhymes closely with what enterprise leaders are saying, in survey after survey, about their own programs.

The confidence gap has a number attached to it

A KPMG Global AI Pulse survey of more than 2,145 C-suite and senior business leaders across 20 countries put it plainly: only 7% of leaders report having established AI ROI. Nearly a quarter — 24% — say they're facing active investor pressure to prove the spending is worth it. And 42% admit they have only partial visibility into how their AI costs are even accumulating across systems, vendors, and business units.

Read those three numbers together and a picture forms: most large organizations are spending heavily on AI, can't fully account for what they're spending, and can't demonstrate what they got for it. That's not a technology problem. That's a governance problem wearing a technology costume — and it's the kind of problem that tends to get expensive before it gets fixed.

The free-spending era of enterprise AI is ending — not because AI stopped being useful, but because CFOs finally started asking the question that should have come first: what exactly did we buy, and what did it return?

By the numbers

$2.59 trillion — global enterprise AI spending in 2026 (Gartner)

47% — year-over-year increase in that spending

95% — share of enterprise AI pilots showing zero measurable P&L impact (MIT NANDA)

7% — share of leaders who report having established AI ROI (KPMG)

24% — share facing active investor pressure to prove value

42% — share with only partial visibility into their own AI spend

Nobody's slowing down, though

Here's the part that makes this genuinely strange rather than just another cautionary tale: spending intent isn't cooling off at all. Bain's most recent CFO survey found 56% of finance executives are increasing enterprise-wide AI investment by more than 15% this year, and looking two years out, 83% plan increases above 15%, with 42% expecting jumps above 30%.

In other words, the spending curve and the results curve have decoupled from each other. Companies are betting bigger while the evidence that the last round of bets paid off gets weaker, not stronger — a pattern that, in almost any other category of corporate capital allocation, would trigger immediate board-level scrutiny.

Why boards keep approving budgets anyway

Some of this is legitimately rational, not just momentum or fear of missing out. AI capital expenditure has increasingly been treated as a strategic position — the cost of not having invested if a competitor's bet pays off is judged worse than the cost of an individual pilot failing. That's a defensible framework in theory, though it becomes considerably less defensible the longer the "pays off eventually" story goes unverified.

Some of it is more mundane: once AI line items are embedded across cloud contracts, software licensing, and headcount-avoidance calculations, they're genuinely hard to isolate and therefore hard to cut, even when a CFO wants to. The spending has, in a sense, gone into hiding inside categories that were never designed to be measured separately.

And some of it is simply that "95% of pilots fail" isn't the same claim as "95% of AI spending is wasted." A single pilot that scales successfully across an entire finance or operations function can outweigh a dozen failed experiments in dollar terms, since successful deployments tend to concentrate spend rather than distribute it evenly. The MIT finding is about pilot-level hit rate, not enterprise-level capital efficiency — a distinction that gets lost every time this stat gets forwarded around a Slack channel without its context attached.

What actually separates the 5% that works

The organizations showing up in the minority that do report measurable P&L impact tend to share a pattern: AI embedded directly into existing, owned workflows and systems of record, rather than standalone tools bolted on next to them. Generic copilots disconnected from a company's actual data and controls are the ones most likely to produce impressive demos and forgettable balance sheets — a distinction that matters far more than which vendor or model a company chose.

The successful minority also tends to share something less technical: a specific, narrow, already-quantified process being targeted, rather than a broad mandate to "use more AI." Vague mandates produce vague results. Specific targets — cut this cycle time, reduce this error rate — produce results you can actually verify against a baseline.

What this means for your portfolio

If you're evaluating companies partly on their AI narrative, the useful question has shifted. "Is this company using AI" answers nothing anymore — virtually everyone can say yes. The question worth asking is whether a company can point to a specific, measured process improvement with a before-and-after number attached, versus a general statement about AI transformation. Earnings calls are increasingly easy to sort this way: management teams with real results tend to cite specific metrics unprompted; management teams without them tend to speak in terms of "positioning" and "capability building."

For sectors with heavy AI capex — financials, software, and increasingly industrials — the KPMG visibility gap (42% with only partial tracking of their own AI spend) is also worth watching at the balance-sheet level. Companies that can't cleanly separate AI spend from general technology spend today are the ones most likely to face a difficult writedown conversation once the ROI question stops being optional.

What we're watching next

Whether the 2026 budget cycle is the one where "prove it" finally becomes a real gate rather than a talking point. If KPMG's 7%-established-ROI number is still in single digits a year from now, that's the signal this isn't a temporary measurement lag — it's a structural mismatch between how AI gets sold internally and how it actually gets used, and one that will eventually show up in either write-downs or a sharp reallocation of budget toward the narrow use cases that actually work.

Sources

1. Fortune / Yahoo Finance — “MIT report: 95% of generative AI pilots at companies are failing”

2. MIT NANDA — “The GenAI Divide: State of AI in Business 2025” (as covered by Virtualization Review)

3. KPMG Global AI Pulse survey(as covered by Beri)

4. Bain & Company — “CFOs funded the AI revolution. Now they are joining it.”

Illustrative figures and third-party research cited above; not investment advice.