Earlier this year, Silicon Valley CEOs competed to push AI usage as far as it would go. Giant token budgets became a status symbol. Developers were encouraged to run AI on everything. One anonymous CTO told researchers his engineer spent $40,000 on tokens in a single month, and he genuinely was not sure whether to stop him or tell everyone else to follow his lead.
That was then. The bill has arrived, and enterprises are scrambling to figure out whether their AI spending is actually producing returns.
The shift is the subject of a recent TechCrunch Equity podcast interview with Tiffany Luck, a partner at New Enterprise Associates (NEA), one of the world’s largest venture capital firms with $25 billion in assets under management. Luck told the podcast that the enterprise AI conversation has pivoted from “what can it do” to “what visibility do you have, what auditability do you have, what token controls do you have?”
TechCrunch coined the term “tokenmaxxing” in April to describe the practice of treating large AI token budgets as a status symbol. The data behind the trend tells a striking story.
According to Jellyfish, a firm that analyzed 7,548 engineers in Q1, the largest token budgets delivered roughly 2x throughput at 10x the token cost. “The tools are generating volume, not value,” the report concluded. GitClear found that AI users averaged 9.4x higher code churn than non-AI counterparts, meaning far more code was deleted or rewritten shortly after being written. Faros AI, studying 20,000 developers, reported that code churn increased 861% under high AI adoption.
Then the overspend hit. Uber blew through its entire 2026 AI coding budget by April. Priceline saw a routine Cursor contract renewal come back 4x to 5x more expensive. One company faced a $500 million Claude bill after forgetting usage limits. Microsoft revoked Claude Code licenses months after enabling them.
“The whole conversation shifted from tokenmaxxing and ‘go fast’ to ‘we need guardrails, how do we control this?'” J.R. Storment of the FinOps Foundation told TechCrunch.
A Market in Transition
The macro numbers confirm the pattern. A PwC survey of 4,454 CEOs across 95 countries found that 56% said AI had not yet produced revenue or cost benefits. Only 12% said AI both increased revenue and lowered costs. An MIT study found that roughly 95% of enterprise generative AI pilots had little to no measurable effect on profitability.
The data is not all bad. Morgan Stanley found that 25% of S&P 500 companies cited quantifiable AI impact in Q1 earnings calls, up from 13% a year earlier. But the direction of travel is clear: procurement has shifted from innovation-focused to finance-focused. Buyers now demand spreadsheet-ready ROI, not transformation narratives.
Where the Opportunity Is
Luck’s central thesis is that value exists at every layer of the AI stack, not just at the model layer. She sees opportunity for startups that help enterprises measure and manage AI spending.
A new ecosystem is forming around this problem. Pure-play cost management tools (Pay-i, Paid), engineering intelligence platforms adding AI monitoring (Jellyfish, Waydev, Faros AI), and platform expansions (Datadog, New Relic adding token-level observability) are all competing for the same customer. Factory, an enterprise AI agent startup, launched a model router that auto-picks the cheapest model per task.
Alexander Embiricos, OpenAI’s head of enterprise, described the shift in a recent interview: “Six months ago, I would have a conversation with a customer and it would be all about ‘What can it do? Is it good enough?’ Now the conversations are about, ‘hey, we’re spending so much. What visibility do you have? What auditability do you have?'”
Nicholas Arcolano of Jellyfish added a counterintuitive finding: “The best ROI comes from moving the broad middle from low to moderate usage, not pushing heavy users higher.” The most cost-effective AI deployment might not come from the teams spending the most.
The Bigger Picture
The tokenmaxxing hangover is a natural correction after a period of uncontrolled experimentation. Enterprises adopted AI tools at a pace that outstripped their ability to measure outcomes. Now they are building the governance infrastructure they should have had from the start.
Goldman Sachs projects global token usage will multiply by 24x by 2030. That growth will not be sustainable without measurement, guardrails, and cost controls. The startups that build those tools, and the enterprises that deploy them wisely, will be the ones that capture AI’s real value.
For now, the industry is in a painful but necessary transition. As Faros AI CEO Vitaly Gordon put it: “Maybe we created a steam engine. We still haven’t figured out the assembly line.”
Sources: TechCrunch (June 17, 2026); TechCrunch — Tokenmaxxing (April 17, 2026); TechCrunch — Token Bill Comes Due (June 5, 2026); Startup Fortune analysis (June 2026)