AI Spending Hits $2.59 Trillion as Most Firms Miss the ROI

Companies worldwide are on pace to spend $2.59 trillion on artificial intelligence this year, a 47% jump from 2025, according to Gartner. Most of that money will not turn into profit anytime soon.

Gartner’s own researchers found that fewer than one in five firms are seeing real benefit from the buildout, and just 12% have scaled AI beyond a handful of pilots. That gap keeps widening, and Gartner already has a name for the companies pulling ahead: breakaway firms.

A Small Tier of “Breakaway Firms” Is Pulling Away

Only 19% of firms are seeing real benefits from their AI investments, and just 12% have scaled the technology across their business, according to research Gartner presented at its Finance Symposium/Xpo in National Harbor, Maryland, in late May. Gartner’s own summary of that keynote states the gap is growing.

The firms already ahead are not a little ahead, they are capturing significant and compounding gains from AI while most organizations stay stuck in early deployment. Faster returns fund further investment, which stretches the lead again. Gartner calls this small group breakaway firms, and everyone else is effectively financing the learning curve.

Where the $2.59 Trillion Is Actually Going

The spending is not slowing to match those disappointing returns. Gartner forecasts worldwide AI spending will reach $2.59 trillion in 2026, with AI infrastructure alone accounting for more than 45% of that total. Servers, chips and networking gear built for AI workloads make up the largest single category, and Gartner expects spending on AI-optimized servers to triple over the next five years.

John-David Lovelock, a distinguished vice president analyst at Gartner, says the enterprise wallet has barely opened. “Up to this point, AI spending has primarily been driven by technology companies and hyperscalers,” Lovelock said. “Enterprises have yet to really flex their spending potential. That is coming, and 2026 will be the inflection year.”

He added that most buyers are still playing it safe. “Currently, organizations show limited appetite for using AI to drive disruptive enterprise change,” Lovelock said. “Instead, they favor tactical AI initiatives with incremental improvements in efficiency and productivity.”

• 45%+ – share of 2026 AI spending going to infrastructure such as servers, chips and network gear, per Gartner.
• $206.5 billion – what Gartner expects enterprises to spend on AI agent software alone in 2026.
• 40% – the share of agentic AI projects Gartner expects companies to cancel before the end of 2027, citing cost and unclear value.

That last figure matters as much as the spending total. Gartner is forecasting a record buildout and a mass cancellation wave in the same twelve months, which is exactly the shape of a market still working out who actually benefits.

Independent Surveys Keep Landing on the Same Fraction

Gartner is not the only one finding this. Several research groups, using different methods and different definitions of success, keep arriving at a similar range: somewhere between one in twenty and one in four companies are getting meaningful value from their AI spending.

The exact figure depends on what is being measured, whether it is profit and loss impact, EBIT contribution, or a subjective sense of meeting expectations. WRITER’s 2026 survey found that 59% of companies now invest at least $1 million a year in AI, yet only 29% report significant returns on that spending. Different methodologies, the same shape of gap.

MIT’s Year-Old Warning Still Holds

MIT warned about exactly this a year earlier. Researchers at MIT’s NANDA initiative published a study in the summer of 2025 that found 95% of generative AI pilots failed to deliver measurable financial return, based on interviews with 150 executives, a survey of 350 employees and an analysis of 300 AI deployments across industries.

MIT’s researchers traced the failures to misallocated budgets and shallow integration between AI tools and daily workflows. More than half of generative AI budgets went to sales and marketing tools, yet the strongest returns showed up in back office automation, things like cutting outsourced work and agency spending.

Workers found value even where employers had not built it in officially. MIT’s researchers described a shadow AI economy in which most staff used personal chatbots for work tasks even though only a minority of companies had bought official subscriptions. The tools worked. Companies simply were not set up to capture or measure what they produced, and a year of record spending has not closed that gap.

Denmark Leads, Romania Trails, and the EU Average Sits at 20%

Zoom into Europe and the same divide shows up in a different shape. The European Central Bank’s latest survey of euro area firms, covering the fourth quarter of 2025, found that 38% had reached an advanced stage of AI adoption, meaning significant or moderate use, while 33% remained stuck at an early, experimental stage.

National figures vary far more widely than that euro area average suggests. Eurostat’s most recent enterprise survey put the EU27 average at 20.0%, up from 13.5% a year earlier, but Denmark leads enterprise AI adoption in the EU at more than double that average.

Eurostat counts a firm running a single customer service chatbot the same way it counts one that rebuilt its operations around AI, so the ranking measures presence, not depth. That is the same distinction Gartner draws at the top of the market. Plenty of companies can say they use AI now. Far fewer can say it changed anything that shows up on a balance sheet.

What the Leaders Do Differently

AI does not follow one cost curve, and it does not produce one uniform type of value.

Twisha Sharma, a senior principal analyst in Gartner’s finance practice, told CFOs at the Gartner Finance Symposium/Xpo in Sydney in March that CFOs need to stop looking for a single ROI formula and build a portfolio instead, one that blends routine productivity gains with a few bigger, riskier bets.

The firms actually pulling ahead have already done the harder work behind that advice. Advisory firm CFGI, in its 2026 CFO outlook, found that workflow redesign, not which model a company buys, carries the strongest correlation with earnings impact. Yet only about one in five organizations using generative AI have actually redesigned how work gets done, which is a large part of why returns stay so inconsistent.

Gartner sees the same divide in talent. Only about 30% of finance staff currently qualify as digital talent, meaning they can build a tech solution when they hit a problem, while breakaway firms are targeting 90% or more. Clement Christensen, a vice president analyst at Gartner, said finance leaders need to democratize technology work and empower their people because hiring enough digital talent simply will not be possible.

How Should Companies Budget for AI in 2026?

Gartner recommends splitting AI spending into three buckets rather than chasing one big bet: productivity tools that save time now, targeted process fixes with a clear owner, and a small slice for bigger transformational projects. Founders and finance chiefs can copy that structure at any size, judging each bucket on its own timeline instead of one blended ROI target.

• Productivity use cases – drafting, research and support tools that save time within weeks and should absorb most of the budget.
• Targeted process improvements – forecasting, procurement or reporting fixes with a named owner and a defined check-in date.
• Selective transformational bets – the smallest slice, reserved for the handful of ideas with a real shot at changing the business.

Deloitte’s 2026 enterprise AI survey found the same gap between ambition and delivery. Two thirds of organizations reported real productivity and efficiency gains from AI so far, but only 20% are already growing revenue through AI, even though 74% hope to get there eventually.

Gartner’s own long-range forecast still lands on the optimistic side: by 2029, companies that get the portfolio approach right could add 10 points of margin growth. Getting there starts with admitting how much of this year’s $2.59 trillion will not.

Frequently Asked Questions

What Is a Realistic Timeline for AI Investments to Pay for Themselves?

Timelines vary sharply by project type. One CFO-focused analysis of payback models found a gap between the 7 to 12 months executives often expect and the 2 to 4 years many transformational projects actually take to break even. Insurance broker Gallagher’s 2026 adoption survey landed in a similar range, with organizations reporting an average of 28 months before AI value outweighs the upfront cost.

Do Younger and Smaller Companies Get Better Returns From AI Than Big Enterprises?

Often, yes, at least on adoption depth. The European Central Bank’s late 2025 survey of euro area firms found 56% of companies under five years old had reached an advanced stage of AI use, compared with 45% among large, listed or venture backed firms overall. Younger firms appear to move faster because they carry less legacy process to redesign.

Which Industries Are Actually Seeing AI Change Their Bottom Line?

Very few, according to MIT’s 2025 research. Only two of the nine major sectors researchers studied, technology and media, showed material business transformation from generative AI. Most other industries, including retail, healthcare and manufacturing, reported pilots that stayed limited to small, isolated use cases rather than company-wide change.

How Reliable Are AI Agents on Complex, Multi-Step Tasks?

Not very, yet. Advisory firm CFGI’s 2026 CFO outlook found that AI agents handling complex enterprise tasks succeeded on the first attempt only about 24% of the time, improving mainly when a human stayed in the loop to check the work. That is one reason Gartner expects a large share of agentic AI projects to be scrapped before the end of 2027.

Is It Better to Buy AI Tools From a Vendor or Build Them In-House?

Buying tends to work better right now. MIT’s research on enterprise AI pilots found that externally built AI tools reached successful deployment roughly twice as often as systems built entirely in-house, about 67% versus 33%. Vendors that specialize in one workflow have usually already solved integration problems that internal teams are tackling for the first time.