The AI price war that analysts spent a year predicting has finally arrived, and it is uglier than almost anyone forecast. OpenAI is cutting the cost of its frontier models by as much as 90 percent to pull marquee customers away from Anthropic, and it is doing so just weeks before its rival tries to pull off the largest technology listing in history. The AI price war is no longer a thesis on a venture capitalist's slide. It is a line item on the invoices of some of the most closely watched companies in software.
By Blumefield | October 25, 2026
A Discount War Breaks Into the Open
For most of the past two years, the two leading American laboratories behaved like a duopoly that preferred not to compete on price. Both sold intelligence at a premium, both pointed to runaway demand, and both let the other hold its ground. That truce has broken. According to reporting that surfaced over the weekend, OpenAI has been quietly offering discounts of 30 percent and more to lock in enterprise contracts, and in at least one case the savings on offer reached the 80 to 90 percent range. The AI price war is being fought contract by contract, in private rooms, with terms that would have been unthinkable a year ago.
The trigger appears to have been a strategic decision at OpenAI to buy market share even where that means selling below cost. The company has paired genuine efficiency gains in its newer Sol and Luna model families with aggressive, deal-specific concessions aimed squarely at Anthropic's book of business. Sam Altman framed it in characteristically breezy terms, saying the company would find "a lot of ways we can help people get more value for less spend." Stripped of the optimism, the message to customers is simpler. Switch, and your bill falls through the floor.
What makes this AI price war different from the usual enterprise haggling is where it strikes. Anthropic built its recent surge on coding workloads, the exact territory OpenAI is now carpet bombing with discounts. Claude Code turned a research lab into a revenue machine. OpenAI's Codex push, long dismissed as a laggard, has become the battering ram.
The Customers Switching Sides
The defections are already visible, and they are not small accounts. Replit moved the bulk of its AI spending to OpenAI after the company offered more than 30 percent off a contract reported to be worth over 100 million dollars a year. Rippling, the workforce platform run by Parker Conrad, saw OpenAI's share of its coding workload climb from 16 percent to 52 percent while Anthropic's collapsed from 57 percent to 18 percent. DataCamp's chief executive said bluntly that the new pricing cut his company's AI costs by 80 to 90 percent.
The concessions extend well beyond token prices. One American bank with more than 7 billion dollars in assets reportedly doubled its ChatGPT deployment to 500 employees at no additional charge, paying under 7 dollars per user against a list price north of 20. These are the sort of terms a company offers when it has decided that owning the customer relationship matters more than this quarter's gross margin. In an AI price war, the land grab comes first and the economics are argued about later.
For enterprises, the timing could hardly be better. Chief technology officers have spent 2026 watching their model bills detonate. Uber's technology chief reportedly burned through the company's entire annual AI budget by April. A JP Morgan memo bluntly titled "AI Bills Are Out of Control" circulated this month, and the bank's own data chief noted that some employees now spend more on AI tools than they earn in salary. Palantir's Alex Karp compared the habit to an addiction. Against that backdrop, a vendor offering to halve the bill is not a salesperson. It is a lifeline.
Why Chinese Models Set the Floor
The uncomfortable truth beneath the AI price war is that neither American lab is fully in control of the price. The floor is being set thousands of miles away, by Chinese laboratories that keep open-sourcing models good enough to matter. DeepSeek, Moonshot's Kimi, Zhipu's GLM and others now publish open weights that independent inference providers can host and serve at a fraction of closed-model prices. On several coding benchmarks these models trade blows with Claude Opus at roughly one-thirteenth of the cost.
That dynamic rewrites the math for everyone. As Delphi Ventures analyst Tommy Shaughnessy put it, the model is the single largest cost an inference provider carries, and Chinese labs are handing it over for free. When the most expensive input in the supply chain drops to zero, prices drift toward zero with it, and the premium that OpenAI and Anthropic have charged starts to look less like a moat and more like a temporary mispricing. The AI price war between the two Americans may simply be the moment they both admit the Chinese competition is real.
There is an escape hatch. If Chinese labs reversed course and went closed-source, the pricing pressure would ease and the American incumbents would breathe again. For now the opposite is happening. The open-weight flood continues, and every new release gives enterprise buyers another credible reason to walk into a renewal negotiation and ask why they should pay full freight for anything.
The Margin Problem Nobody Solved
None of this would sting so much if the underlying businesses were healthy. They are not. OpenAI posted an adjusted operating margin of negative 122 percent in the first quarter of 2026, meaning it lost 1.22 dollars for every dollar of revenue it booked. The famous 20 dollar monthly subscription was always a loss leader, priced below what heavy users actually cost to serve. Push those users onto metered pricing and the credits vanish faster, which is precisely why the AI price war looks less like a path to profit and more like a race to see who can subsidize demand the longest.
Anthropic entered this fight in better shape, which is part of what makes the assault so pointed. Its annualized run rate reportedly leapt from 9 billion dollars at the end of 2025 to 47 billion by the middle of 2026, a rise of more than 400 percent driven overwhelmingly by coding. For the first time, more companies tracked by the Ramp AI Index now pay for Anthropic than for OpenAI, and the second quarter was said to be the company's first profitable one. OpenAI's discounting is an attempt to break that momentum before it hardens into a durable lead.
The optimistic case for both firms rests on a Western oligopoly thesis. If only two or three trusted laboratories can serve regulated industries at scale, the argument goes, they can eventually charge what they like and repair their margins at leisure. The pessimistic case is that open weights have already punctured that story, and the AI price war is the sound of the air leaking out. Which reading is correct will decide whether these companies are the next cloud platforms or the next telecom carriers, hauling enormous revenue and almost no profit.
What It Means for the November IPO
The stakes sharpen because Anthropic is reportedly weeks away from a public listing targeted for mid-November at a valuation discussed in the range of 1.8 to 2 trillion dollars. An AI price war that visibly slows its growth, at the exact moment bankers are setting a price, is close to a worst-case scenario for the roadshow. Investors buying into a hypergrowth story will want to know whether the growth survives a rival willing to give the product away. You can follow Blumefield for continuing coverage as the pricing campaign collides with the IPO calendar.
There are already cracks worth watching. Two engineers from the Claude Code team departed in the weeks before the listing, and an investor memo reportedly flagged a deceleration between the second and third quarters. None of that is fatal on its own. A high-profile customer defection narrative during an IPO window, however, is the kind of thing that turns a confident book build into an anxious one. OpenAI understands that calendar perfectly, and the timing of its discount blitz does not look accidental.
For buyers, the near term is a windfall. Compute that cost a fortune in the spring is on sale this autumn, and the smart procurement teams are signing multiyear deals while the giants are desperate for logos. The longer question is who absorbs the losses, and for how long. An AI price war funded by venture capital and pre-IPO enthusiasm can run for a while, but not forever, and the day the subsidies stop is the day enterprises learn what intelligence actually costs. Both OpenAI and Anthropic are betting they will be the one left standing when that bill comes due.