Sam Altman makes stunning admission about AI

Sam Altman has spent years telling anyone who would listen that artificial intelligence usage would grow beyond what most people could imagine.

On June 2, at an OpenAI enterprise livestream, he revealed a number that made his own case for him and then undermined it in the same breath.

The number was 100 billion tokens per month. The complication was that even that figure, which would have been unthinkable six years ago, is no longer the record.

What Altman revealed about AI token consumption and why it matters

Speaking at OpenAI's "Intelligence at Work" enterprise event on June 2, OpenAI CEO Sam Altman disclosed that the company's top internal token user now consumes approximately 100 billion tokens every month.

"The token leader at OpenAI uses about 100 billion tokens a month," he said. "To my embarrassment, that's not the token leader in the world. We found someone that used even more," according to Axios.

More AI:

• Micron sits at the center of a red-hot chip rally
• IBM CEO sends blunt message on AI and quantum computing
• Anthropic CEO makes shocking admission about AI

The context makes the number more striking. Six and a half years ago, OpenAI's heaviest user consumed roughly 100,000 tokens per month, a figure then likely the highest in the world. Today, 100,000 tokens per month is the global per capita average.

The top user inside OpenAI is now consuming one million times more than that former record. Altman projects another million-fold expansion in token usage ahead.

Why Altman also admitted AI costs have become a huge problem

The same event that produced the 100 billion token disclosure also produced something less expected from the CEO of the world's most valuable AI company: a candid acknowledgment that cost management has become a serious issue for enterprise customers.

Altman said cost concerns are now the second most common complaint OpenAI receives from enterprise customers. He referenced a meme circulating among corporate clients: "The company spent its entire 2026 budget in Q1. Can you make it more efficient?"

He acknowledged the issue had arrived "all of a sudden" and that early in 2026 "nobody cared about costs. Everyone was happy with their spending," according to Axios.

That has changed sharply. Uber blew through its entire planned 2026 AI budget in four months, CEO Dara Khosrowshahi told The Verge. As I previously reported, on June 2 that Uber capped all employees at $1,500 per month per agentic coding tool, including Claude Code and Cursor.

Uber COO Andrew Macdonald went further, saying publicly that the company cannot yet draw a clear line from its rising token spend to consumer-facing product improvements.

How Microsoft, Amazon, and Walmart are all hitting the same AI cost wall

Uber is not an isolated case. The cost problem Altman described is playing out simultaneously across some of the largest technology companies in the world.

Microsoft canceled most of its internal Claude Code licenses by mid-May and redirected engineers across its Experiences and Devices division to GitHub Copilot CLI by June 30, the end of its fiscal year, as I previously reported. The move came after internal spending on AI coding tools exceeded what Microsoft had budgeted.

Separately, the same tension between AI investment and workforce costs was visible when Scout AI agent launched alongside internal documents describing the product as designed to make users "addicted" before expanding features.

Amazon scrapped its internal AI token leaderboard after a senior executive told staff to stop using AI for its own sake.

Walmart has also capped how much its employees can use its in-house AI agent, after previously providing workers with unlimited tokens, according to PYMNTS.

The pattern is consistent: companies encouraged maximum AI adoption, set budgets based on 2025 usage rates, and discovered in early 2026 that the actual cost of agentic AI tools was running far ahead of those projections.

The same dynamic was visible when Amazon engineers that the company was spending $200 billion on AI infrastructure while cutting 30,000 jobs , a collision between capital deployment and workforce cost that is now being replicated at the individual tool level inside companies trying to manage their monthly AI bills.

Key figures from Altman's June 2 enterprise event and the broader AI cost picture:

• OpenAI's token usage has grown one million-fold in six and a half years: the top user went from 100,000 tokens per month in 2019 to 100 billion today; 100,000 tokens per month is now the global per capita average, according to Axios
• OpenAI maintains an internal token leaderboard and has a culture of employees competing to consume the most tokens; Altman himself described the leaderboard and said some employees post their totals publicly on X, according to TechCrunch
• Peter Steinberger, a developer of the OpenClio app, publicly disclosed consuming 603 billion tokens over 30 days; a New York Times report identified one OpenAI employee who used 210 billion tokens in a single week, both exceeding the 100 billion monthly figure Altman cited, according to PYMNTS
• Uber's COO Andrew Macdonald said on the Rapid Response podcast in May 2026 that the company "cannot yet draw a line" from its rising Claude Code token spend to consumer-facing features for riders and drivers, a direct acknowledgment that heavy AI spending is not yet translating into measurable product improvements, according to Bloomberg
• Altman also disclosed that Anthropic has overtaken OpenAI in corporate AI spending among enterprise customers, making it the first time a rival has led OpenAI on that metric; he framed it as evidence that the enterprise AI market is expanding rather than a competitive setback, according to Axios

What the AI cost crisis means for the companies building and buying AI infrastructure

The cost dynamic Altman described has two distinct audiences. For companies buying AI tools, the message is that the cost curve moved faster than anyone planned for and the budget problem is now a mainstream corporate concern rather than an early-adopter edge case.

For companies building AI infrastructure, including Nvidia, the major cloud providers, and OpenAI itself, the message is more complicated.

Rising token consumption at the scale Altman described is ultimately good for infrastructure companies. Every token consumed is revenue for OpenAI, compute revenue for cloud providers, and demand for Nvidia chips.

The one million-fold growth in usage over six and a half years, if it repeats, implies a market for AI infrastructure that is orders of magnitude larger than what exists today.

The tension is that the companies funding that infrastructure buildout, Uber, Microsoft, Amazon, and Walmart, are simultaneously learning that token costs are harder to control than they expected.

If budget pressure leads companies to cap or reduce usage, the growth trajectory Altman is projecting runs into a commercial ceiling. The 100 billion token figure is impressive. The meme about companies burning through their annual budgets in Q1 is a signal that the next phase of the AI boom will be defined as much by cost discipline as by capability.

Related: Microsoft CEO sends shocking message to employees

The Arena Media Brands, LLC THESTREET is a registered trademark of TheStreet, Inc.

This story was originally published June 7, 2026 at 3:37 PM.