OpenAI’s Revenue Model Hits a Wall as Enterprises Cut AI Spending

Published by James Harris on June 26, 2026June 26, 2026

Corporate AI budgets hit a predictable wall. After a year of encouraging, then mandating, then quietly celebrating token consumption as a proxy for innovation, major enterprises are now discovering that burning through API credits at scale does not automatically produce business value.

The correction has arrived with some speed. Amazon scrapped an internal leaderboard that tracked employee AI usage after concluding it was generating busywork rather than output. Uber exhausted its entire 2026 AI coding budget in four months and has since capped employee spending on AI tools at $1,500 per month. Meta flagged an “exponential increase” in internal AI costs to roughly 6,000 employees. Accenture, which previously warned staff they could risk losing promotions without AI adoption, is now trying to stop those same employees from using it on trivial tasks. The pattern is consistent enough to have a name: the industry calls the overconsumption phase “tokenmaxxing,” and the hangover is hitting providers directly. OpenAI and Anthropic built their revenue projections on the assumption that enterprise token consumption would keep compounding, and both are burning through cash on compute and hiring while waiting for profitability that depends on that growth continuing.

IBM’s framing is the sharpest diagnosis here: token minimization is not the fix, because it just replaces one bad metric with its opposite.

The actual pressure on AI providers is structural, not cyclical. Enterprises are already routing complex tasks to expensive flagship models while shifting routine workloads to cheaper alternatives or open-source models running on their own infrastructure, which carry no per-token cost at all. IDC projects that by 2028, 70% of leading AI-driven enterprises will run multiple models rather than depending on a single provider, which would compress margins across the sector as capability differences narrow and price competition accelerates. That dynamic puts OpenAI’s $1 trillion self-valuation and Anthropic’s comparable figure under genuine scrutiny, since both assume sustained, concentrated enterprise spend rather than a fragmented, cost-optimized multi-model market. Even Sam Altman has acknowledged that AI costs have become a “huge issue” for customers this year, which is a notable admission from a CEO whose company is still spending aggressively on the infrastructure those customers are now trying to avoid.

IBM’s alternative framing, which it calls “valuemaxxing,” measures completed tasks, time saved, and rework avoided rather than tokens consumed. Whether enterprises actually adopt that framework or simply cut budgets and move on is the question that will determine whether this is a temporary correction or the beginning of a longer repricing of what enterprise AI is actually worth.