For most of this year the question hanging over the AI industry was how high token prices could go. This week it flipped. According to a Wall Street Journal report picked up across the financial press, OpenAI is considering drastically lowering what it charges for tokens, the metered units that big customers use to pay for AI, in a bid to win business back from Anthropic. The company reportedly expects Anthropic to cut prices in response. In other words, the two most valuable AI startups in the world are preparing to race each other to the bottom, at the precise moment both are asking the public markets to believe in their margins.
The trigger is a customer revolt that has been building for months. Sam Altman admitted at a recent OpenAI event that usage costs had gone "from an issue that never came up" to "a huge issue" in the space of half a year, joking that it had become a meme for companies to exhaust their annual AI budgets in the first quarter. The anecdotes back him up. Uber's technology chief said the company had burned through its entire 2026 budget for agentic AI by April. Axios reported one CFO accidentally racking up half a billion dollars in Claude usage fees in a single month. The practice of "tokenmaxxing," using as many tokens as possible and treating consumption as a proxy for productivity, has curdled into its opposite: reports of Amazon and Meta employees inflating usage numbers to satisfy internal pressure, while finance teams struggle to connect any of it to results.
What makes the timing remarkable is the IPO calendar. Anthropic and OpenAI both confidentially filed to go public within the past two weeks, with Anthropic's $965 billion valuation now topping OpenAI's $852 billion. On Friday, SpaceX showed them what is at stake, debuting on the Nasdaq in the largest IPO in history and surging past a $2 trillion market value. The exuberance is real, but so is the scrutiny that comes with public filings. A price war would hit both companies where they are weakest: they already lose billions of dollars a year running their models, and the case for their valuations rests heavily on revenue growth that cheaper tokens would directly undercut.
The sources frame the same facts very differently. Forbes and Futurism present the cuts as a competitive response to Anthropic's surge, driven by Claude Code's popularity with enterprise software teams. Zero Hedge reads it as the bubble's first audible hiss, noting that a token price index has been falling for days and arguing that Anthropic's headline revenue figures annualised an unsustainable spending spree that customers are now unwinding. An MSNBC opinion column worries about what happens to everyone else, pointing out that the AI giants, their cloud providers, and their chip suppliers are so densely cross-invested that a stumble by one could cascade through the markets the IPOs are joining.
Between the optimists and the doomers sits a more ordinary commercial story. Customers got their first real bills, balked, and the sellers are responding the way sellers always have. The open question is whether cheaper tokens expand usage enough to pay for themselves, or whether the industry has just discovered that its product, however transformative, is one customers will only buy at a loss-making price. The answer will land in public quarterly reports rather than private board decks, and sooner than either company might like.