Quick Answer
Enterprises are putting hard ceilings on AI spend. Accenture now centrally funds $29 a month per person for Claude and charges anything above that to a project budget code. Uber used up its 2026 AI budget in four months and restricted access. A cap stops waste and useful work at the same rate. The alternative is seeing AI cost by team, by project and by outcome, then funding what returns.
Companies spent the past year telling employees to use more AI. Now a lot of them are putting a lid on it.
A Goldman Sachs report covered by The Economic Times on October 19, 2026 said “many enterprises are currently capping AI spend, given elevated AI costs.” Uber was the starkest example; it used up its 2026 AI budget in four months and restricted access in response.
Accenture is one of the latest to set a limit. Since October 1, 2026, the company has centrally funded just $29 a month per person for Claude, 404 Media reported. Anything above that goes to a Work Breakdown Structure, Accenture’s budget code for a project or team. No more free-for-all: every extra dollar has to go somewhere. Before the cap, some employees were hitting $1,000 a month. A spokesperson said Accenture still plans to “invest in AI capabilities, skills and enablement,” according to The Next Web.
At Gartner’s IT Symposium/Xpo in Orlando on October 20, 2026, vice president analyst Alicia Mullery urged CIOs to write new rules for AI economics with their CFOs, starting with which teams are responsible for token costs.
Why finance can’t tell good AI spend from bad
When you can’t tell good spend from bad, you limit all of it. Any CFO who has frozen travel or hiring knows how this works.
A cap can’t tell the difference. It shuts off the runaway agent and the engineer building something customers will pay for at the same rate. Your provider’s invoice won’t help. It lists seats, tokens and models. It doesn’t say which project the spend belonged to, who asked for it, or what it produced.
CloudZero’s new survey report, Why Finance Caps What It Can’t See, puts numbers on that blind spot. Of 252 finance leaders surveyed, 54% can see AI spend only as a rough category split or a single line on a bill. The most common change they plan for FY2027 is new approval gates (27%). Only 18% are tying funding to demonstrated ROI.
Among teams choosing between added sign-off requirements and AI ROI, the ones with weak visibility picked the gate 71% of the time. “Gates work without visibility,” the report says, “which is exactly why 27% of the market is reaching for them.”
2026 State of AI Spend Report
Why Finance Caps What It Can’t See
Learn how finance leaders are managing AI spend, and how they’re approaching FY2027
What the friction costs
Almost no finance team got through the year clean. In CloudZero’s survey, 86% of finance leaders ran into at least one real consequence: an AI initiative killed or paused, a hiring delay, a rejected investment ask, a mid-year budget shuffle, or an uncomfortable board conversation. Two-thirds hit more than one.
Among companies that can see only a total or a rough split in their AI spend, 21% had an investment ask stalled or rejected. At companies with better visibility, it was 9%, less than half as many.
Teams that rely on approvals took the heaviest hit. Compared with teams that tie funding to ROI:
- They killed or paused AI initiatives at nearly three times the rate (20% vs. 7%).
- A third delayed hiring, against 19% of everyone else.
- 55% hit at least one growth-slowing consequence in 2026 (a killed initiative, a hiring delay, or a blocked investment request), versus 28% of the ROI group.
Both groups planned the same increases in AI spend. In the report’s words, gate teams “absorb more damage to control the same number.”
Then there’s the money that never got spent. 30% of teams that added approval gates came in under their AI budget, compared with 11% of teams that tie funding to ROI. The report calls that “hesitation rather than discipline.” It isn’t savings. Someone approved that budget because they expected a return, and the work that would have delivered it never happened.
The report is careful to say these things travel together, and its data doesn’t mean that pre-approvals result directly in slowdowns or other consequences. Still, the pattern is hard to miss. Gates, in the report’s words, “become a permanent tax when nothing is funded to replace them.” And 78% of the teams adding gates say better data would change their decisions, yet only 28% expect FY2027 to require it. Friction in AI spend doesn’t show up on the P&L as savings. It shows up as slower growth.
| Measure | Teams adding approval gates | Teams tying funding to ROI |
|---|---|---|
| Killed or paused an AI initiative | 20% | 7% |
| At least one growth-slowing consequence in 2026 | 55% | 28% |
| Came in under AI budget | 30% | 11% |
| Delayed hiring † | 33% | 19% † |
| Planned FY2027 increase of 20% or more | 39% | 39% |
Source: CloudZero, Why Finance Caps What It Can’t See, n=252 finance leaders, fielded August to September 2026. † The hiring row compares teams adding gates against all other respondents, not against the ROI group. Every other row is a direct gate-versus-ROI comparison.
Limit the waste, not the work
That takes seeing AI cost by team, by project, and by outcome. A budget code points the right way: usage past the line lands on the project that benefits, and someone close to the work decides whether it was worth it. It only works if every dollar traces to an owner.
Some companies can already see that far. In our survey of 252 finance leaders, 23% can allocate AI spend systematically and produce ROI a board would accept. Compared with everyone else, that group:
- Failed to secure an AI investment at 3%, versus 19%.
- Hit a growth-slowing consequence in 2026 at 27%, versus 47%.
- Planned the same FY2027 AI increases.
| Measure | Can allocate AI spend systematically (23% of respondents) | All other respondents |
|---|---|---|
| Failed to secure an AI investment | 3% | 19% |
| Hit a growth-slowing consequence in 2026 | 27% | 47% |
| Planned increase in AI spend, FY2027 | No material difference | No material difference |
Source: CloudZero, Why Finance Caps What It Can’t See, n=252 finance leaders, fielded August to September 2026.
Most finance leaders want to get there: 74% say linking spend to outcomes would change specific FY2027 decisions. Only 12% can break AI spend down to the customer, product, or cost-center level and act on it inside the quarter.
Uber hit its 2026 ceiling in four months. Plenty of companies will send out a cap memo of their own. The useful ones will name the budget code, the owner, and what the work returned, so the next cap lands on a line someone can actually defend.
