In May, Axios reported something that should have made every CFO in the country sit up: a large enterprise had spent $500 million on AI services in a single month — and internal finance hadn’t signed off on it beforehand. It sounds like an anomaly. It isn’t. It’s the single clearest illustration of what’s actually happening to enterprise AI spending in 2026: the story is shifting from how fast can we grow to can you explain where the money went.
The Numbers Are Almost Too Big to Process
Start with the scale. Gartner now expects global AI spending to hit $2.59 trillion this year — a 47% jump over 2025. The four largest hyperscalers, Amazon, Microsoft, Alphabet, and Meta, are on track to pour somewhere between $700 billion and $725 billion into capital expenditure in 2026 alone, based on their own public guidance — a 77% increase year over year.
Break that down further and the individual numbers get stranger. Amazon has committed roughly $200 billion, most of it going toward AWS data centers and custom chip development. Meta, notably, has no cloud business generating recurring revenue to point to as justification — and it still raised its full-year capex guidance to somewhere between $125 billion and $145 billion back in the first quarter. Whatever advertising-side AI gains Meta is banking on, its next earnings call is going to need to make the math work in public.
None of that spending, on its own, tells you whether the investment is paying off. And that’s precisely where the disagreement starts.
Finance Is Finally Asking the Question Nobody Wanted to Answer
Through most of 2024 and 2025, AI purchasing decisions largely bypassed the normal cost-benefit gauntlet that governs the rest of an enterprise IT budget. CTOs and AI strategy teams made the calls, and the reasoning was simple: move now or get left behind. Nobody wanted to be the one demanding a business case in a market moving that fast.
That grace period is over. Forrester now expects enterprises to push 25% of planned AI spending out into 2027 as financial scrutiny catches up with the buying. A separate Gartner survey found that fewer than one in three corporate decision-makers could name a specific financial result tied to their AI investment. Deloitte’s 2026 State of AI in the Enterprise report — based on interviews with 3,235 leaders — found two-thirds of organizations reporting real productivity gains. That’s a genuine number. But it sits awkwardly next to a PwC finding that more than half of CEOs (56%) describe their AI initiatives as having produced essentially no measurable return. MIT’s widely circulated GenAI Divide research goes further still, putting the failure rate for generative AI pilots at 95% — meaning the overwhelming majority never make it out of the experimental phase.
It would be easy to read all of this as evidence of a bubble deflating. That’s probably the wrong conclusion. What it actually looks like is a technology investment finally going through the same maturation cycle every major enterprise technology eventually faces. Cloud spending went through an almost identical reckoning a decade ago, which is why FinOps emerged as a discipline in the first place — a structured way to bring runaway cloud bills under control. AI spending is now due for its own version of that same discipline, and Forrester’s postponement figure reads less like a vote against AI and more like a vote for governance that should have existed already.
Adoption Hasn’t Slowed — It’s Just Concentrating
Set the ROI skepticism aside for a moment, because the adoption data tells a genuinely different story. McKinsey’s latest State of AI survey puts the share of organizations using AI in at least one business function at 88%. Separate research found that 72% of organizations now have at least one AI workload actually running in production — up 17 percentage points from just 55% in 2024. This is no longer a pilot-stage technology at most large companies. It has moved into production, and it’s staying there.
The more interesting shift is in how companies are getting there. Menlo Ventures’ 2025 enterprise research found that 76% of enterprise AI use cases are now purchased off the shelf rather than built in-house — a near-total reversal from a year earlier, when 53% of use cases were homegrown. That’s a meaningful signal: AI is being treated less like a bespoke engineering project and more like a standard procurement line item, which is exactly the kind of shift you’d expect once finance departments start paying closer attention. The same research found AI deals converting into production deployments at nearly double the rate of traditional SaaS contracts — 47% versus 25% — and enterprise spending on AI coding tools rocketing from $550 million in 2024 to $4 billion in 2025, making it the single largest category of departmental AI spend.
The vendor map has also been redrawn. Anthropic’s share of enterprise LLM API spend has more than tripled, climbing from 12% in 2023 to 40% today, while OpenAI’s share slid from 50% down to 27% over the same stretch. For any CIO currently negotiating a multi-year model contract, that’s a reminder that the landscape they’re locking into today looks nothing like it did two years ago — and may look just as different two years from now.
Agentic AI Is About to Repeat the Same Cycle, Faster
If generative AI has graduated from experimentation into governed, production deployment, agentic AI is standing exactly where generative AI stood a year or two ago — full of momentum, thin on guardrails. Gartner expects 40% of enterprise applications to embed task-specific AI agents by the end of 2026, up from under 5% in 2025. At the same time, Gartner is projecting that more than 40% of agentic AI projects will be canceled by 2027, citing spiraling costs, unclear business value, and weak risk controls. Only 21% of organizations currently have anything resembling a mature governance framework for these systems.
That combination — agents getting embedded everywhere at once, alongside a wave of projected cancellations — is the clearest warning sign available to any CISO or CIO building a 2027 plan right now. Agentic AI is tracing the same adoption curve generative AI already walked, just compressed into a much shorter window and carrying meaningfully higher stakes. A chatbot that drafts bad text is an inconvenience. An agent that acts autonomously on bad judgment is a different category of risk entirely.
The Real Choice Facing Enterprise Leaders
Nothing in the Q2 2026 data points to enterprise AI spending slowing down, and nothing in it suggests companies have soured on generative AI’s underlying value. What it does show is a market splitting into two distinct camps. One camp is managing AI spend as a governed portfolio — tracking usage and outcomes with the same discipline applied to cloud cost management, and able to answer the board’s ROI question without flinching. The other camp is still operating on the enthusiasm-driven buying habits that defined 2024 and 2025, and it’s this group that will absorb the brunt of Forrester’s 25% spending postponement and Gartner’s coming round of agentic AI project cancellations.
For the CTOs, CISOs, and CFOs building budgets for the second half of 2026, the question worth sitting with isn’t the size of the capex figure making headlines. It’s whether their organization can give a straight answer to what that spending is actually returning. If it can’t, that’s the real risk hiding in this year’s budget.
