Enterprise AI hits an inflection point as companies rein in spending and demand real results

Enterprise AI has officially exited the pilot stage. OpenAI reported in April that business customers now represent more than 40% of its total revenue, with that segment on pace to reach parity with its consumer business by the end of 2026, according to a note published by Denise Dresser, the company's Chief Revenue Officer. The company's APIs are processing more than 15 billion tokens per minute, and its coding tool Codex has reached 3 million weekly active users.

But even as adoption accelerates, a hard correction in AI spending is taking hold across corporate America, creating a tension that neither OpenAI nor Anthropic can ignore, particularly as both reportedly filed confidentially for IPOs in early June, according to CNBC.

The tokenmaxxing era gives way to accountability

For much of the past two years, enterprise AI spending operated on a spend-first, optimize-later logic. Developers were encouraged to pump as many tokens as possible into AI models, a phenomenon CNBC has labeled tokenmaxxing. Employers built internal leaderboards rewarding heavy AI usage with little scrutiny of costs or outcomes.

That dynamic is changing fast. Uber disclosed this month that it has introduced monthly spending tiers on certain AI tools, beginning at a $1,500 base level per employee, with higher access available on request. The policy came after Uber's CTO Praveen Neppalli Naga revealed in April, as reported by The Information and cited by CNBC, that the company had exhausted its entire annual AI budget within four months.

The pressure is not limited to large corporations. Flo Crivello, CEO of AI startup Lindy, told CNBC he moved 100% of his company's traffic from Anthropic's Claude models to DeepSeek, a Chinese provider of lower-cost open-weight models. He said the switch will save Lindy millions of dollars within months. Even so, he expects AI costs to still exceed the roughly 25-person company's payroll bill. "It's a matter of survival for the business," Crivello told CNBC.

Analysts flag a structural growth ceiling

The efficiency shift has caught the attention of Wall Street. Gil Luria, an equity analyst at D.A. Davidson, told CNBC that current growth rates at OpenAI and Anthropic are likely the fastest either company will ever post, calling it largely a matter of basic math. He specifically flagged the risk that large enterprise customers would begin imposing limits on what he described as out-of-control token spend.

Anthropic's annualized revenue run rate stood at $47 billion as of May, according to CNBC. Both Anthropic and OpenAI have seen valuations approach $1 trillion on the back of explosive demand, and both are now navigating how to sustain that momentum as budget scrutiny intensifies. Open-source alternatives, along with competing efficiency-focused offerings from Microsoft, Amazon, and Google, are giving enterprise buyers more leverage than they have had at any point in the AI boom.

OpenAI's answer: a unified agent layer, not just more tools

OpenAI is not standing still in the face of these pressures. Dresser's April note outlined a two-part enterprise strategy built around what the company calls OpenAI Frontier, a platform designed to let organizations deploy and manage AI agents across their entire business rather than in isolated product pockets. Customers including Oracle, State Farm, and Uber are already using Frontier to move agents across internal systems and data sources, according to OpenAI.

The company has also formalized a network of alliance partners to help enterprises integrate its technology into existing infrastructure. McKinsey, Boston Consulting Group, Accenture, and Capgemini are among the consulting firms involved, alongside cloud and data platform providers AWS, Databricks, and Snowflake. A jointly developed Stateful Runtime Environment with AWS is designed to allow agents to retain context and memory across complex, multi-step tasks, according to OpenAI.

The second pillar of the strategy is what OpenAI describes as a unified AI superapp, a single interface where employees interact with agents throughout the workday. The company says Codex, which it reports has grown more than five times since the start of 2026, is an early signal of that shift, with customers including GitHub, Nextdoor, and Notion building multi-agent systems capable of executing engineering work end to end.

What comes next

The two stories playing out simultaneously, rapid enterprise adoption on one side and a cost-driven pullback on the other, are not necessarily contradictory. Companies are not abandoning AI; they are becoming more deliberate about where and how they deploy it. The question OpenAI, Anthropic, and their rivals must answer is whether they can build offerings that satisfy both the ambition of enterprise transformation and the financial discipline now being demanded by CFOs and boards.

For OpenAI, the immediate measure will be whether enterprise revenue actually reaches parity with consumer revenue by December, as Dresser projected. For Anthropic, the IPO process itself will force a reckoning with how the company characterizes growth trajectories to public market investors. Both timelines arrive before the year is out.