Two of the world’s most valuable private companies are heading toward public markets at the same time — and the race between them is already reshaping how Wall Street thinks about AI. OpenAI, the maker of ChatGPT, and Anthropic, the company behind Claude, both filed confidential IPO paperwork with the US Securities and Exchange Commission in June 2026. Their combined target valuations — up to $1 trillion each — would put them among the largest stock market debuts ever attempted.
But as the listings take shape, questions are getting louder about whether their pay-per-use business model can hold up under public market scrutiny. Palantir CEO Alex Karp told CNBC on July 1, 2026 that “something has gone completely wrong” with how frontier AI labs sell access to their models. Enterprise customers from Uber to Meta have started rationing AI spending. And cheaper open-weight competitors from China are catching up fast. Here is what you need to know.
When you send a message to ChatGPT or Claude, the text is broken into small chunks called tokens — roughly 3–4 characters each. Every word you type and every word the AI generates costs tokens. OpenAI and Anthropic charge companies a price per million tokens processed.
For a single conversation, the cost is negligible. But multiply that across thousands of employees using AI for hours a day, and the bills become enormous — fast. Agentic AI workflows, where models run multi-step tasks autonomously, can consume millions of tokens in a single session.
OpenAI’s GPT-5.5 is currently priced at $5 per million input tokens and $30 per million output tokens at standard rates. Priority pricing reaches $12.50 and $75 respectively. Anthropic’s Claude Sonnet 5 is at introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026.
A new term entered the corporate vocabulary in 2026: “tokenmaxxing” — what happens when employees, under pressure to show they are using AI, use it far more than the task actually requires, with no oversight on spend.
Uber burned through its AI budget in four months. Microsoft discontinued some AI tool integrations internally. Salesforce and Meta both put limits on employee AI usage. The common thread: the token-payment model made costs spiral before any ROI was measurable.
The irony is that this pressure arrives just as both OpenAI and Anthropic are trying to convince public market investors that enterprise AI spending is sustainable and growing. OpenAI projects a $14 billion net loss in 2026 even as revenue surges. Anthropic remains unprofitable, with profitability pushed out to 2028. Both companies are spending heavily on compute — OpenAI plans to spend $50 billion on computing infrastructure in 2026 alone.
Palantir CEO Alex Karp launched the sharpest public attack on OpenAI and Anthropic’s business model on July 1, 2026, two days after Palantir announced a joint AI infrastructure deal with Nvidia for US government agencies.
Karp’s argument: enterprises are paying per token for AI output, but getting little measurable return while potentially exposing sensitive business data to third-party model providers. He argued that open-weight models — where the company owns and runs the model itself on its own hardware — give enterprises ownership of their compute, data, and outputs.
Karp also warned that the US AI industry should not dismiss Chinese open-weight models. Beijing startup Z.ai’s GLM-5.2 is now ranked among the top 10 large language models globally by Artificial Analysis, and second-best for web development on Code Arena. The open-weight model is reported to be four to six times cheaper than comparable frontier AI offerings from US labs.
Karp’s remarks coincide with his company’s commercial interests in Palantir’s Artificial Intelligence Platform — context worth keeping in mind when weighing his critique.
Some enterprise customers have already begun switching away from OpenAI and Anthropic to cheaper alternatives. Coinbase and other firms have reported testing Chinese-developed models like DeepSeek to reduce costs.
Meanwhile, the Financial Times reported that OpenAI has held talks with the Trump administration about giving the US government a 5% stake — a proposal that hinges on other US AI labs, including Anthropic, agreeing to similar terms. Export control concerns add further complexity: in June 2026, Anthropic’s two most advanced models, Fable 5 and Mythos 5, were taken offline worldwide following a Commerce Department export-control order. The restriction was lifted on June 30, 2026, and the models returned on July 1.
Against this backdrop, both companies need to demonstrate to public market investors that revenue growth is durable, enterprise AI spending is sticky, and their models remain sufficiently ahead of cheaper alternatives to justify premium token pricing.
Both Anthropic and OpenAI have filed confidential IPO paperwork with the SEC — Anthropic on June 1, 2026, and OpenAI one week later. Anthropic is targeting an October 2026 listing on Nasdaq at a valuation near $965 billion. OpenAI, whose CEO Sam Altman has called anything below a $1 trillion valuation unacceptable, is now expected to list in 2027 rather than late 2026. Goldman Sachs, Morgan Stanley, and JPMorgan are lined up as underwriters for both.
The filings arrive at a moment of genuine pressure on the business model that built these companies. Enterprise customers have started rationing AI spending. Competitors from China are offering comparable models at a fraction of the cost. And the question of whether pay-per-token pricing can sustain trillion-dollar valuations under the discipline of quarterly public market reporting remains open.
Both companies continue operating as private entities until any listing is finalised. The coverage above drew from OpenAI’s funding announcement, reporting from The Information, public SEC filing confirmations, and CNBC’s July 1, 2026 interview with Palantir CEO Alex Karp. For further context on the AI infrastructure build-out, see our coverage of Nvidia’s cloud revenue model and the broader AI tooling landscape in 2026.
