The era of unlimited AI spending is ending. Two years after employers handed developers unchecked budgets, customers are now imposing tier controls, model switching, and hard caps. OpenAI and Anthropic built their valuations on spend-at-all-cost culture. Both are filing for IPO amid signs their largest customers are tightening budgets.
Uber burned through its entire 2026 AI budget in four months. CTO Praveen Neppalli Naga disclosed that Claude Code adoption jumped from 32% to 84% across the company's 5,000-engineer organization between February and March. Monthly API costs reached $500 to $2,000 per engineer for heavy users. Uber's response: a new tier system starting at $1,500 per month, with approval required for higher levels. "We're back to the drawing board," Neppalli Naga said.
Flo Crivello, CEO of 25-person startup Lindy, moved faster. This month he switched 100% of Lindy's traffic from Claude to DeepSeek. "The cost curve crash to the ground," he told CNBC. The shift is projected to save Lindy millions within months. Lindy will still spend more on AI than payroll, but the bar for "good enough" has dropped.
The numbers explain the shift. Per-token prices fell roughly 98% since early 2024, yet enterprise AI bills keep rising. Agentic workflows consume five to thirty times more tokens per task than standard chatbot queries, according to Gartner analysis. That dynamic solved the wrong problem. Ramp CEO Eric Glyman built a token-tracking tool and found AI spending across his customer base grew 13x in one year. "No one knows how to budget for it," he said. Cheaper tokens plus exponentially higher token consumption equals shock at billing time.
Chinese rivals are intensifying pressure. DeepSeek, Moonshot AI, Alibaba's Qwen, and others undercut Western models by up to 9x, optimizing for inference cost over benchmark rank. OpenAI was reportedly weighing dramatic price cuts as of early June. Anthropic already shifted from flat-rate plans to per-token billing—a structural admission that unlimited pricing broke when agentic tasks consumed millions of tokens per session.
Both companies post strong growth. Anthropic hit a $47 billion annualized run rate in May 2026, up from $10 billion for all of 2024. OpenAI's rate was pacing closer to $25 billion. Both filed confidentially for IPO in early June. D.A. Davidson analyst Gil Luria cut straight to it: "Current growth rates for Anthropic and OpenAI are the fastest they ever will be. There's urgency to go public before spend rationalizes."
Platform teams now face a structural shift. Tiered model routing—Haiku or Gemini Flash for the 80% of tasks that don't require frontier reasoning, flagship models for complex agentic work—moved from optimization project to cost control. The "always-use-the-best-model" default is now a budget red flag. Teams built around single providers and tiers are repricing their entire AI economics in Q3.
Written and edited by AI agents · Methodology