The C-suite has spent two years approving AI budgets on faith. The invoice has now arrived — and the returns column is mostly blank.
A new Infosys survey of more than 1,000 senior executives, reported by CIO Dive this week, puts hard numbers on what many boards have only whispered: 72% of enterprises have successfully scaled less than a quarter of their AI pilots. Two-thirds cannot measure AI's return on investment at all. And yet global enterprise AI spending is projected to hit $64 billion in 2026 — a 63% increase on last year.
Read those three numbers together and the picture is uncomfortable. The C-suite is accelerating spend on a technology most organisations cannot yet scale or measure.
The failure pattern is consistent. Pilots are green-lit by innovation teams with no owner in the P&L. Success is defined in demo terms ("the model works") rather than business terms ("cycle time fell 30%"). And nearly three-quarters of executives told Infosys that pressure to prove short-term ROI is actively inhibiting the longer-term initiatives that would produce real returns — a self-defeating loop where impatience kills the only projects worth being patient for.
Only about half of the surveyed organisations have a balanced KPI framework for evaluating AI value. That is the tell. A company that cannot define what AI success looks like in business terms has not made an AI decision — it has made a fashion decision.
If you are a CEO or CFO reviewing AI line items this quarter, three moves follow directly from this data. First, force every AI initiative to name a single P&L owner and a single business metric before renewal — no metric, no budget. Second, separate the measurement problem from the value problem: two-thirds of firms can't measure ROI, which is not the same as having none; fund the instrumentation before killing the initiative. Third, resist the urge to demand quarterly payback on platform investments — the Infosys data shows that exact pressure is why pilots die at the scaling stage.
There is also an important signal in who published this. Infosys — an Indian IT-services major with a front-row seat on Global 2000 transformation budgets — is telling its own client base that deployment discipline, not model selection, is the bottleneck. For Indian IT leaders, that is the services opportunity of 2027: the money is moving from "build me a pilot" to "make my twenty pilots pay."
The scaling gap bites hardest where the spending is heaviest — and the majority of that $64 billion sits in US corporate budgets. American CFOs face an added complication their peers elsewhere don't: activist investors and equity analysts now ask AI-productivity questions on earnings calls, which pushes companies to announce deployments before they can measure them. That is precisely the short-term-proof pressure the Infosys data identifies as the pilot-killer. For Canadian executives the exposure is different but real: Canada's enterprise AI adoption trails the US while its regulatory scrutiny (AIDA's successor framework, provincial privacy law) arrives on the same schedule — meaning Canadian firms risk paying compliance costs on AI estates that haven't yet produced returns. In both markets, the arbitrage for Indian-origin IT services leadership is identical: North America is where the twenty unscaled pilots per client actually live.
The honeymoon phase rewarded announcements. The next phase rewards operators.