The consensus says AI investment compounds from here. The data says otherwise for the median enterprise. AI spend will plateau in 2026 and only scale when three preconditions land.
The position
The prevailing narrative is that enterprise AI spending is on an exponential curve. Vendor talk tracks reinforce the narrative. Analyst reports project compound growth rates that imply a near-doubling within two years. Our position is contrarian. For the median enterprise, not the leading 10 percent, AI spend will plateau in 2026 and re-accelerate only when three preconditions land. Misreading the next 12 months as continued exponential growth leads to over-allocation by vendors, stranded capacity by buyers, and disappointment that calcifies into AI skepticism in the buying committees.
The leading 10 percent of enterprises will continue to invest aggressively because their platforms have started to compound. They are a meaningful part of the market by spend and a small part by count. The headline numbers will look healthy through 2026, sustained by the leaders. The median firm experience will be the plateau, which is not what the headline numbers will say.
Why the plateau is coming
The first driver is pilot fatigue. Three years of pilots have produced modest scaled output for the median firm. Boards lose patience after the third unscaled cycle. Budgets get re-scrutinized. Programs are asked to defend not just their roadmap but their existence. The defense gets harder when the unit economics are honestly measured for the first time.
The second driver is unit-cost reality. As programs measure the real cost of an answer, including retrieval, evaluation, and override, the ROI math gets harder. Some use cases that looked profitable retract. The honest measurement is, on net, a good thing, because it allows the program to focus capital where it earns. The short-term effect is a slowdown in commit while the portfolio is rebalanced.
The third driver is talent constraint. The bottleneck has shifted from model access to operating talent. There are not enough domain translators, system engineers, and quality stewards to staff the programs the boards approved. Talent supply grows slowly. The programs that cannot staff cannot scale, regardless of how much capital is available. This is the constraint that most boards are not yet seeing clearly.
The three preconditions for re-acceleration
Re-acceleration requires three preconditions. The first is platform compounding. Firms that built shared platform components in 2024 and 2025 will see use-case unit cost fall fast enough to reignite spend. The platform investment becomes self-funding through use-case savings. Firms that did not build platform have no on-ramp to this dynamic.
The second is decision-rights reform. AI's value materializes when decisions move closer to the data and to the moment the data was created. Without decision-rights reform, the value stays trapped in dashboards that nobody acts on quickly enough. Most enterprises have not yet started this reform, which is why the value has been disappointing.
The third is outcome-tied capital. Funding by quarterly outcome rather than annual envelope is the operational discipline that lets scaling decisions actually get made. Without it, the program is committed for a year to the portfolio that looked right at the budget meeting, regardless of what the evidence later showed.
What to do if this position is right
Protect platform investment in the next budget cut. Cut individual use cases that have not earned scale rather than the platform that makes use cases cheaper. Move funding to quarterly outcome gates now, before the budget cut forces the change under duress. Expect the median competitor to plateau. Position the firm to compound through the trough, because the gap between the leaders and the median will widen substantially in 2026 and the cost of catching up grows with the gap.
Closing
The exponential story is comforting because it removes the need to do the unglamorous work. The plateau story is uncomfortable because it makes the work visible. The firms that do the work compound through the plateau. The firms that wait for the exponential to resume will discover that the resumption was happening at competitors who never stopped doing the work.