Runaway AI token bills force a corporate cost reckoning
As AI coding budgets blow past forecasts, firms are racing to cap token spend before the bill swallows the savings. The new battleground is FinOps.
The News
The bill for the generative-AI boom is landing, and it is bigger than anyone budgeted for. According to TechCrunch, Uber burned through its entire 2026 AI coding allowance by April, while one unnamed company ran up a single Claude bill of $500 million. Priceline says its contract renewal came in 4-5x more expensive than the previous term.
The pressure is forcing a rapid shift in how finance teams think about software. J.R. Storment, executive director of the FinOps Foundation, said companies began telling him in April and May that they were running three times over their full-year token budgets. His organisation now counts 180 vendors selling tools to track and curb the spend.
The usage curve is steep. Faros AI, which tracks roughly 20,000 developers, found per-developer token consumption rose 18.6x in nine months. The heaviest users ship about twice as much code but burn ten times the tokens. Goldman Sachs projects global token usage will multiply 24x by 2030.
Why It Matters
For two years the prevailing instinct was to spend freely and chase speed. That era is closing. Vitaly Gordon, chief executive of Faros AI, captured the unease when he said one of his engineers spent $40,000 on tokens in a single month and he genuinely did not know whether to stop him. Priceline's Chris Reed was blunter, likening cheap trial credits to a dealer giving away the first hit.
The deeper problem is measurement. Cloud bills generate hundreds of millions of rows a month; token bills generate trillions. As Salesforce's Nishant Gupta put it, token economics is more abstract and opaque than anything finance teams have managed at this scale before. That opacity is precisely what FinOps, the discipline that grew up taming runaway cloud spend, now has to solve again.
The pattern rhymes with history. When Amazon Web Services made server capacity a metered utility in the late 2000s, costs ballooned before a whole industry of cost-control tooling emerged to rein them in. Tokens are the new compute unit, and the same correction is arriving faster.
Indian Angle
Nowhere is the maths more sensitive than in India. The country's IT services giants and the 1,700-plus global capability centres operating here employ engineering headcounts in the hundreds of thousands. When token pricing is set in dollars and salaries are paid in rupees, a weakening currency quietly inflates every AI coding bill. A rupee hovering near record lows against the dollar turns a flat US price list into a rising local cost.
That changes the build-versus-buy calculation for Indian enterprises and startups. Cost-conscious teams at firms like Razorpay, Zerodha and Freshworks have every reason to meter usage aggressively, route cheaper models for routine tasks, and reserve frontier models for high-value work. It also strengthens the case for home-grown models from Sarvam and Krutrim, whose pitch of lower-cost, India-hosted inference looks more compelling when the alternative is an unpredictable dollar invoice.
There is an opportunity here too. India's deep base of finance and analytics talent is well placed to build the token-FinOps tooling the world suddenly needs, much as Indian firms once rode the cloud-cost-management wave. For Indian CFOs, the lesson is immediate: treat AI tokens as a metered utility from day one, not a flat subscription.
FAQ
Why are AI token costs rising so sharply?
Usage is climbing far faster than per-token prices are falling. Faros AI measured an 18.6x rise in per-developer token consumption over nine months, as coding assistants take on bigger tasks and run more autonomously. The result is bills that outpace budgets even when unit prices drop.
What is FinOps and why does it matter now?
FinOps is the practice of giving finance and engineering shared visibility and control over variable cloud spending. It matured taming runaway server bills, and is now being applied to AI tokens. The FinOps Foundation already lists 180 vendors building tools for this.
How does a weak rupee affect Indian AI budgets?
Most frontier-model pricing is dollar-denominated. When the rupee falls against the dollar, the same token usage costs more in local terms, squeezing margins for Indian startups and service firms even if the headline US price never changes.
Where can I read the original report?
The full reporting, including the company examples and executive quotes, was published by TechCrunch on 5 June 2026. The link is in the attribution note below.
This story was reported by TechCrunch. Read the full original coverage at TechCrunch.