Enterprise AI spend tripled while token prices cratered 98%

Token prices are down 98%, but the bills are not. The Next Web ties that reversal to a set of incidents that read like a stress test for enterprise AI procurement: Uber blew through its entire 2026 AI coding budget by April, and Microsoft revoked its developers' Claude Code licenses six months after enabling them.

The painful part for decision-makers is not that AI can get expensive. It is that the economics started with the promise of getting cheaper, and the stories suggest the practical spend can still spike fast, even when underlying token pricing collapses. One company reportedly ran up a $500 million Claude bill in a single month after forgetting to set usage limits. And a Priceline employee, speaking to TechCrunch, described a routine Cursor contract renewal coming back four to five times higher than expected.

So what actually happened when “tokens got cheaper”? In AI services, token cost is only one variable. Usage caps, routing, model choice, tool behavior, and how developers deploy agents or assistants matter as much as the per-token price. A forgotten usage limit is basically the AI version of leaving a credit card on the stove. If safeguards are missing, or if teams test features without constraining consumption, costs can accelerate regardless of headline token rates.

These incidents also point to a governance issue that boards and CFOs recognize instantly: enablement without controls. Microsoft enabling Claude Code and then revoking licenses six months later suggests a shift from experimentation to enforced spend management. When a large vendor clamps down after an early rollout, it is usually because usage behavior and cost exposure are not aligning with the assumptions used to launch the program. Uber’s “by April” budget burn is another sign that planning for AI as a normal line item is breaking down. A coding budget that gets consumed in the first part of the year implies either underestimated adoption, underestimated intensity, or both.

At the same time, enterprise AI demand is accelerating. The headline from The Next Web frames the pattern as “Enterprise AI bills tripled,” which tells you this is not isolated chaos. Once a company integrates AI into workflows, spending tends to rise with it. Developers want to use the tool more, faster, and for more tasks. Teams also start to expect quality improvements from additional context, longer generations, or more iterations. Even if the cost per token drops, total tokens processed can surge because the workflow changes around the tool. That is the mismatch executives are trying to explain.

Now layer in what regulators and standard setters generally care about: predictability, transparency, and comparability. Regulators do not need to police every bill line item to see the risk. If companies cannot understand why costs moved, they cannot audit vendors, they cannot forecast internally, and they cannot defend procurement decisions. That is why The Next Web says the industry wants a standards body to explain why the numbers do not match the price signals. A standards body can define how usage is measured, how billing terms map to consumption, how limits should be communicated, and what reporting should look like across providers.

For enterprises, the second-order implication is brutal: even when pricing looks like it should improve, the control plane can still fail. A missing usage limit turns “cheaper tokens” into “uncapped outcomes.” A revoked license can stall teams and force replatforming. A four to five times renewal jump can trigger vendor disputes, contract renegotiations, and internal credit headaches. In other words, the financial story is becoming an operational story, and operational costs tend to arrive without warning.

For peers in the C-suite, the stake is whether AI budgeting becomes a repeatable system or a recurring fire drill. If token prices keep falling but bills keep climbing, the industry will treat cost volatility as a feature. Standards would not just help in spreadsheets. They could enable better governance, clearer accountability with vendors, and more disciplined rollout strategies across engineering, procurement, and finance. The executives who get this right will not be the ones who chase the lowest token rate. They will be the ones who can explain, measure, and control end-to-end spend.