Uber Ran Out of Tokens in April. IPOs Start in October.

Enterprise customers just staged a billing revolt against the token model OpenAI and Anthropic are betting their IPOs on — and the loudest critic is selling the alternative.

On July 1, Palantir CEO Alex Karp told CNBC's Squawk Box that "something has gone completely wrong" with how frontier labs sell AI. He said U.S. businesses now assume they'll burn tokens, get no value, and hand over their IP — while OpenAI and Anthropic prepare megacap listings built on that same meter.

The receipts landed first:

• Uber exhausted its entire 2026 AI coding-tools budget by April — four months into the year — after rolling Claude Code to roughly 5,000 engineers in December, Fortune reported May 26, citing Uber CTO Praveen Neppalli Naga's April remarks to The Information. Adoption hit 84% by March; heavy users ran $500 to $2,000 a month.
• Uber COO Andrew Macdonald told the Rapid Response podcast: "That link is not there yet" between token consumption and shipping more useful consumer features, per Business Insider. CEO Dara Khosrowshahi said Uber is slowing hiring to offset AI spend on the May earnings call.
• Microsoft is canceling Claude Code licenses across its Experiences + Devices division — Windows, Office, Teams, Surface — by June 30, The Verge reported. EVP Rajesh Jha cited "benchmarking-then-convergence" onto GitHub Copilot CLI; sources cited fiscal-year cost pressure on the same deadline.

Karp's pitch — and the conflict:

The CNBC appearance followed a June 29 Palantir-Nvidia deal to deploy open Nemotron models in air-gapped sovereign environments — the product Karp was arguing for when he said customers want control over "compute, their models, their data stack, and their alpha." His line: "The jig is up."

Palantir shares rose roughly 8% on July 1, per Yahoo Finance. A day earlier, Palantir posted a nine-point "AI sovereignty" manifesto on X warning: "Data retention is your treasure. Transfer it at your own peril."

Karp insisted he wasn't attacking Sam Altman or Dario Amodei personally — "there's nothing more fun than debating Dario in private" — but said frontier models have been "completely, irresponsibly, oversold." Co-anchor Becky Quick told him he sounded angry; he reframed the fury as "the voice of American business."

The IPO clock doesn't care:

• Anthropic confidentially filed a draft S-1 on June 1 and closed a $65 billion Series H at a $965 billion valuation in May, per Anthropic's announcement. Reports target an October Nasdaq debut.
• OpenAI, valued at $852 billion in March, filed its own confidential S-1 and is expected to follow in 2027, The Daily Upside reported July 6.
• Salesforce and Meta have also rationed employee AI access as token bills outran returns, per the same reporting. Google CEO Sundar Pichai told I/O attendees CIOs are "so concerned about how much their companies are blowing through budgets."

Why "tokenmaxxing" broke:

Engineers coined the term for burning tokens to prove AI adoption — leaderboards, uncapped API access, no ROI guardrails. Uber's internal usage rankings accelerated the spike. Microsoft adopted Claude Code in December 2025; six months later the tool was too popular for the budget.

Cheaper open-weight rivals are filling the gap. Beijing's Z.ai GLM-5.2 ranks in Artificial Analysis's top 10 and costs four to six times less than frontier APIs, per The Daily Upside. Enterprise customers are switching to DeepSeek and imposing hard per-employee caps.

Convina's view: Karp is selling sovereign AI while trashing token billing — but he's not wrong about the timing. Uber and Microsoft didn't reject AI; they rejected a pricing model that punishes success. OpenAI and Anthropic are about to ask public markets to price infinite token consumption at a moment their best customers are installing hard caps. The IPO roadshow will pitch scarcity; the enterprise inbox is sending cancellation notices.