The Enterprise Lock-In Bet Behind OpenAI and Anthropic’s IPOs

Claude Code, Anthropic’s agentic coding platform, is generating over $2.5 billion in annual run-rate revenue as enterprise subscriptions have quadrupled since January. Anthropic filed confidentially for an initial public offering (IPO) at a $965 billion valuation on June 1, days after closing a $65 billion funding round. OpenAI is running the same race through Codex, and in mid-May merged its ChatGPT and Codex product teams under president Greg Brockman into a single organization. Both companies switched from flat-fee subscriptions to per-token billing in the past eight months, the pricing shift that their public listings need to generate defensible margins.

That shift is producing the first real enterprise blowback. Uber burned through its entire 2026 AI coding budget in four months, then put a $1,500 monthly cap on each engineer’s tool spending. Walmart deployed a model-agnostic coding assistant built to route requests across multiple AI providers, avoiding any single vendor’s billing structure. The lock-in thesis both companies are selling to Wall Street is running into organized corporate cost controls.

Claude Code Is Already a $2.5 Billion Business

For most of the past decade, AI companies measured success by benchmark scores and research publications. The metrics that move investors in 2026 are quarterly revenue figures and enterprise renewal rates. Claude Code, publicly available since May 2025, is the product that changed what gets measured.

• Over $2.5 billion in annual run-rate revenue as of February 2026, more than doubled since January
• Business subscriptions: quadrupled since the start of 2026
• Enterprise customers spending $1 million or more annually: more than 1,000, doubled from roughly 500 in February to above 1,000 by April, per Futurum Research, an enterprise technology research firm
• Fortune 10: 8 of the 10 largest US companies are now Claude customers

Smartsheet, the workflow software company, told Anthropic that engineers using the platform ship three times more code and merge 31% more pull requests than peers on the same teams. Those figures came from Anthropic’s Series G funding announcement in February, which disclosed enterprise metrics the company had no prior obligation to share. Once an IPO filing is in progress, the case for customer retention has to become visible to prospective public investors.

The Meter Goes On

Both companies needed the IPO math to work, and flat-fee seats couldn’t produce the margins a public listing demands. Anthropic shifted its enterprise pricing from seat-based contracts to API-level token billing in November 2025, a change that existing customers discovered only when renewing, according to The Information. OpenAI completed the same transition for all existing ChatGPT Enterprise plans by April 23, 2026. Both companies simultaneously released new frontier models with higher per-token costs: OpenAI’s GPT-5.5 at twice the prior API rate, Anthropic’s Opus 4.7 at roughly 1.4 times, per independent developer Simon Willison’s May analysis of the AI pricing shift. For enterprise teams in multi-year contracts, the November 2025 Anthropic change arrived at renewal time rather than on a published date, leaving some to discover it mid-cycle.

Sam Altman, OpenAI’s chief executive, described the direction in a March interview: “We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter.”

Anthropic’s projected operating profit for Q2 2026 is $559 million on $10.9 billion in expected revenue, a margin of roughly 5 percent. Per-token billing is the mechanism that makes that margin expandable: a developer running an AI coding agent for eight hours a day consumes vastly more compute than one running a weekly query, and those costs now flow directly to the vendor’s revenue line. For enterprise teams accustomed to annual flat-rate contracts, the shift amounts to an open-ended commitment where usage, not headcount, sets the final bill.

When the Bill Comes Due

By April, Uber’s chief technology officer (CTO) Praveen Neppalli Naga told The Information that the company had exhausted its entire 2026 AI coding budget in four months. Individual software engineers had been generating monthly token bills between $500 and $2,000 each. About 11% of Uber’s live backend code updates were being written entirely by AI agents. The company’s internal productivity rankings had tied scores to token consumption, a practice critics labeled “tokenmaxxing,” which drove usage regardless of whether output improved.

The company’s response was a $1,500 monthly spending cap per tool for each engineer, per Bloomberg, plus an internal dashboard for real-time cost tracking and a formal review process for exceptions. Andrew Macdonald, Uber’s president and chief operating officer (COO), put the underlying problem plainly on the Rapid Response podcast:

If you’re not actually able to draw a direct line to how many useful features and functionality you’re shipping, that trade becomes harder to justify.

Macdonald’s comments, made on the Rapid Response podcast, centered on the gap between rising AI spend and measurable product output. The company had incentivized adoption, engineers adopted aggressively, and the question of whether the output justified the cost remained open.

Walmart arrived at structurally similar terrain from a different direction. Michael Pfaffenberger, a developer, built Code Puppy as an open-source coding agent designed to route requests across OpenAI, Anthropic, and local models, avoiding commitment to any single pricing structure. The tool spread through the retail giant’s workforce, with tens of thousands of employees using it. In June, Walmart moved to per-employee token allocations after usage ran above projections, per Retail Systems.

A Gartner forecast puts global AI agent software spending at nearly $207 billion for 2026, up 139% from $86.4 billion in 2025. The companies generating that growth are the same ones now installing controls on how much of it reaches any single vendor.

A Thirty-Day Window to Switch

On May 13, OpenAI launched a 30-day “Switch to Codex” promotion offering two months of free enterprise access to any organization migrating from a competing platform before mid-June. The offer covered only developers not previously on a ChatGPT Business or Enterprise plan. Anthropic raised usage limits by 50% for all paid tiers within hours, a counter-offer running through July 13.

The tactical exchange sits on top of an architectural difference that matters for enterprise procurement decisions.

Claude Code’s local execution model is a practical advantage in regulated industries. Finance and healthcare organizations routinely face data residency requirements that restrict what code can leave a company’s own network. Codex’s cloud architecture enables asynchronous parallel work, queuing tasks to run without a developer waiting for output, but requires that code move through OpenAI’s infrastructure for each job. Both products carry the same enterprise base price of $20 per seat per month plus token costs by usage, as of 2026.

In March, Microsoft launched Copilot Cowork, an enterprise agent built on Anthropic’s technology and distributed through Microsoft’s existing corporate channels. OpenAI’s longest-running financial backer was simultaneously marketing a competing product. The fragmentation inside individual corporate relationships is accelerating faster than the public competitive positioning suggests.

Beyond the Engineer’s Terminal

Claude Code and Codex both launched as tools for software developers. The expansion target is every other role in the enterprise.

Anthropic launched Claude Cowork on January 12 as a research preview for macOS desktop users. Where the coding platform runs in a terminal, Cowork operates on the desktop and handles tasks that knowledge workers spend their days on: document review, data extraction, report building, calendar management. The enterprise-wide release followed on February 24, adding connectors for Google Drive, Salesforce, Gmail, DocuSign, and FactSet. Eleven open-source plugins shipped with the platform, covering:

• Sales operations (Salesforce Agentforce connector)
• Contract and document workflows (DocuSign integration)
• Financial data analysis (FactSet plugin)
• Legal review processes
• Healthcare and life sciences (under HIPAA compliance for enterprise customers)

PwC, the professional services firm, announced a partnership with Anthropic to build industry-specific Cowork plugins for finance and healthcare clients. When PwC embeds a client’s institutional workflows into a plugin configuration, switching AI platforms means migrating both the tool and the operational logic encoded inside it.

OpenAI went a step further in mid-May, merging its ChatGPT and coding agent teams into a single product organization under Brockman, with the goal of a unified “super app” handling conversation, code execution, and file management in one interface. The move was first reported by Wired. OpenAI reaches roughly 900 million weekly active ChatGPT users; Anthropic’s approximately 134 million monthly active users skew enterprise. Enterprise accounts represent roughly 80% of Anthropic’s total business, per CNBC.

Code Puppy’s architecture, routing requests across multiple providers from a marketplace of more than 65 options, shows what a deliberate counter-strategy to that platform expansion looks like at the enterprise level.

What a $965 Billion Valuation Has to Prove

Anthropic’s audited financials become public only after the Securities and Exchange Commission (SEC) completes its review of the confidential S-1 filed June 1. No listing date has been set. The company expects to report $10.9 billion in Q2 2026 revenue and an operating profit of $559 million, which would be its first profitable quarter. Its annualized run rate was roughly $4 billion in July 2025 and $9 billion in January 2026; it has told investors the rate will exceed $50 billion by the end of July, an approximately 80-fold increase over two years. OpenAI is preparing its own filing and targeting a fall 2026 debut at approximately $852 billion, per The Next Web.

OpenAI generated $5.7 billion in Q1 2026 revenue, per The Information. CFO Sarah Friar told Bloomberg the company sees “a vertical wall of demand.”

Samuel Colvin, chief executive of Pydantic, an AI developer-tools startup, told Business Insider that both companies are “doing their very best to find ways of locking people in that are not related to model quality,” pointing to coding agents as the primary mechanism. A codebase built with an AI coding agent grows faster than a human engineering team can manage independently, creating sustained reliance on whatever tool built it. Enterprises that let the tool write the infrastructure find themselves needing the same tool to maintain it.

Anthropic’s enterprise concentration gives it more predictable revenue than a consumer-heavy competitor. It also means Anthropic has more exposure to exactly the buyers who now install spending dashboards when AI bills start arriving.

The S-1 will disclose what that 5% margin looks like at scale, and the enterprises running $1,500 token caps have already started voting with their dashboards.

Disclaimer: This article is for informational purposes only and does not constitute investment advice. It discusses companies preparing for or expected to file public listings; figures cited reflect publicly reported estimates as of publication. Consult a qualified financial professional before making any investment decisions. Numbers are accurate as of June 7, 2026.