Uber spent its entire 2026 AI budget in four months. Not on a single flashy initiative, on Claude Code, the coding assistant its own engineers were using so heavily that monthly costs per engineer ran between five hundred and two thousand dollars for power users, across a base of roughly five thousand engineers. Uber's president and COO, Andrew Macdonald, said publicly that the company could not yet draw a clear line between that surge in usage and measurably better products for riders and drivers. Ninety five percent of Uber's engineers use AI tools monthly, and seventy percent of committed code is now AI generated. The usage is real. The return on it, by Uber's own admission, is not yet provable.
The industry now has a name for this pattern: tokenmaxxing, token consumption surging well ahead of any demonstrated return, treated as an innovation cost rather than a governed line item. Uber is the most visible example because the numbers are public and the admission was candid, but the underlying pattern, adoption outpacing the ability to measure or govern its cost, is showing up across enterprises that treated AI tooling budgets as separate from, and less scrutinised than, the rest of their technology spend.
Why Token Spend Escalates Faster Than Traditional Software Costs
Traditional enterprise software costs scale with seats or usage tiers that are relatively easy to forecast and cap. Token based AI tooling costs scale with how much an individual user actually chooses to use the tool, which varies enormously between a casual user and what the industry is now calling a power user, and that variance is exactly what caught Uber's finance models off guard. A budget built assuming moderate, evenly distributed usage will be blown through quickly once a meaningful share of users become power users, and there is currently very little friction stopping that shift from happening quickly once a tool proves genuinely useful.
Why "Just Cap Spend" Is Only a Partial Fix
Uber's response, a cap of fifteen hundred dollars per employee per month on agentic coding tools, is a reasonable emergency measure, but it treats the symptom rather than the underlying measurement gap. A spending cap controls cost without answering the question that actually matters, whether the spend is producing proportionate value. An organisation can be perfectly within budget and still have no idea whether its AI tooling spend is a good investment, which is a different and arguably more important problem than the budget overrun itself.
What Proportionate Governance Actually Looks Like
The organisations managing this well are not treating AI tooling spend as a separate innovation budget insulated from the financial governance applied to any other enterprise software category. That means tracking token consumption against specific, measurable outcomes, not just usage volume, setting tiered access or budgets based on role and demonstrated value rather than a flat cap applied uniformly, and building the same kind of usage dashboards and cost attribution that already exist for cloud infrastructure spend, which faced an almost identical governance maturation several years ago.
Why This Should Not Be a Reason to Restrict Adoption
The temptation after a story like Uber's is to respond by restricting access to AI tools to control cost, which risks solving the wrong problem. Gartner's own forecasting suggests AI coding assistant costs will become a standard, significant line item by 2028, not a temporary anomaly to be stamped out. The organisations that will be ahead by then are the ones building proper cost governance and outcome measurement now, not the ones that restricted adoption in 2026 and have to rebuild both the capability and the governance discipline simultaneously later.
What This Means for Your Organisation
What we see across implementation engagements is that organisations rarely lack the appetite to measure AI tooling ROI properly, they lack the instrumentation to do it at the point a tool is rolled out, and by the time cost becomes visible enough to prompt a response, as it did at Uber, the organisation is reacting to a budget problem rather than managing a planned one. Building cost attribution and outcome tracking into an AI tool's rollout from day one is a considerably smaller job than retrofitting it after usage has already scaled past what anyone budgeted for.
Key Takeaways
- Uber spent its entire 2026 AI budget within four months on Claude Code usage across roughly five thousand engineers, with its COO publicly stating there was no clear link yet between that spend and measurably better products.
- The industry term for this pattern, tokenmaxxing, describes token consumption surging well ahead of any demonstrated return, often treated as a separate innovation budget rather than a governed cost line.
- A spending cap, which is Uber's current response, controls budget overrun but does not answer whether the spend is actually producing proportionate value.
- Gartner forecasts AI coding assistant costs will be a standard, significant enterprise budget line by 2028, meaning proper cost governance and outcome measurement built in now is preferable to restricting adoption reactively.
How Trusenta Can Help
AI Strategy Enterprise builds token spend governance and outcome measurement into an AI tool's rollout from day one, rather than retrofitting it after costs have already scaled past budget.
Custom AI Development implements the usage dashboards and cost attribution needed to track AI tooling spend against measurable outcomes, not just consumption volume.
Fractional CIO provides the ongoing financial oversight to apply tiered access and budget discipline to AI tooling spend without restricting genuine adoption.
Conclusion
Uber's four month budget burn is a useful, public data point precisely because the admission was so candid, usage without provable return, at meaningful scale, from a household name. The lesson is not that AI tooling spend needs to be restricted. It is that it needs the same financial governance discipline every other significant enterprise software category has already been through, and the organisations building that discipline in now will not be the ones publicly explaining their budget overrun later.
