OpenAI's AI Model Price War: History Has Already Written the Ending

Original Source: Wall Street CN

From the last week of June to the beginning of July, the entire AI track underwent a collective repricing.

Anthropic struck first with Sonnet 5, offering a limited-time discount slashing prices to $2/$10. A little over a week later, xAI, OpenAI, and Meta launched their offerings within 48 hours — Grok 4.5, three tiers of GPT-5.6, and Muse Spark 1.1. Four companies, ten calendar days, and every price tier was adjusted. On July 15th, Altman added a line on X: Sol is already half the price of Fable 5, "would be happy to deliver at a quarter of the price." Another 75% cut.

The company cutting prices lost $93 billion in a single quarter. The one charging a premium started turning a profit.

This isn't what a price war is supposed to look like. Three almost identical historical battles have already written the script for this one.

The Expensive One is Making Money

Anthropic's annualized revenue surged from $9 billion at the end of last year to $47 billion in May this year. OpenAI surpassed $30 billion in the same period. In Q1, OpenAI's revenue was $5.7 billion, and Anthropic's was $4.8 billion — OpenAI still held the lead. But Anthropic projects Q2 revenue of $10.9 billion, nearly doubling quarter-over-quarter, and expects to achieve its first operating profit of $559 million. OpenAI reported an operating loss of $9.3 billion in Q1.

The one cutting prices is bleeding money. The one charging a premium is turning profitable.

The reason lies in their revenue structure. Approximately 85% of Anthropic's revenue comes from enterprise clients — over 1,000 now pay over $1 million annually. These clients buy stability, security, and compliance, not the lowest token price. A 75% price cut wouldn't make them buy a single extra token; a 75% price increase wouldn't make them buy less. OpenAI is the opposite — over 60% comes from consumer subscriptions, and each price cut stimulates price-sensitive C-end users and small developers.

Anthropic doesn't face a price-sensitive market. Altman does. When Altman declared "another 75% cut" on X on July 15th, Anthropic didn't reply — instead, it had its executives publicly urge enterprises not to reduce AI usage due to cost concerns. It wasn't a lack of response; it was a response in a different language.

Intel Tried This Three Decades Ago

In 1998, AMD's market share jumped from 8% to 16%, posing the first real threat to Intel. Intel's response: what semiconductor analysts called an "unprecedented" price cut — not a single round, but a sustained, scorched-earth campaign. Whenever AMD launched a competitive chip, Intel would hammer it with another round of price cuts. The logic was simple: 13 fabs versus AMD's 2 fabs meant a cost structure competitors couldn't replicate.

AMD survived. It was forced to sell off its fabs, but did something Intel hadn't anticipated: it stopped copying and started redefining the product. When Ryzen launched in 2017, AMD turned things around on the back of its product, even surpassing Intel in market cap at one point.

Intel used its cost advantage to keep AMD pinned to the ground for years. But cost advantages expire; product advantages don't.

What Altman is doing is essentially no different from what Intel did in 1998. He has labeled the technology to compress inference costs by 50% as a "core secret." The Information, citing sources, reported: "They don't even want other employees inside OpenAI to know, because if it leaks, other labs will adopt it quickly." The keyword isn't "secret"; it's "adopted" — if rivals master the same efficiency tools, his only remaining weapon becomes burning cash. Intel didn't need to burn cash back then; its price cuts were backed by owning more fabs.

Flywheel or Pump?

From 2006 to 2018, AWS proactively cut prices over 100 times. The price of S3 storage dropped 85% over 12 years. But every time AWS announced a price cut, it was already profitable.

The flywheel spins for one simple reason: every dollar saved came from real efficiency gains — the Graviton self-developed processor offers 40% better price-performance than x86, and Nitro offloads virtualization overhead to dedicated hardware. No matter how low prices go, costs go lower. That's a true flywheel.

This is the story Altman wants to tell. "Inference cost compressed by 50%", "core secret" — all the phrasing signals the same thing: this isn't burning cash to grab market share, it's returning cost savings from technological progress to customers.

But one number punctures the narrative. OpenAI posted an operating loss of $9.3 billion in Q1, losing $1.60 for every dollar of revenue. AWS never lost this much while cutting prices. A flywheel requires the price curve and the cost curve to move down in tandem, or at least for costs to fall faster. If the cost curve hasn't kept pace — if every price cut is funded by burning financing — then it's not a flywheel; it's a pump.

What Altman feeds into the pump's intake is the $122 billion funding round completed in March and $73 billion in cash on hand. The outlet is an IPO. If the pump stops, the water runs dry. More troublingly, as of the end of 2025, OpenAI's compute procurement commitments with cloud providers total as much as $665 billion, spanning through 2030 — whether AI demand grows as expected or not, that money has to be paid out.

No Way Out for Those Burning Cash

In 2014, the subsidy war between Didi and Kuaidi lasted less than six months, burning through $2.4 billion, with daily losses in the tens of millions at its peak. They announced their merger on Valentine's Day 2015. In October of the same year, Meituan and Dianping merged — the most blunt line from Wang Xing's internal email on the day of the merger: "Yesterday we fought bloody battles; today we shake hands in friendship."

It wasn't the founders driving the mergers; it was the investors. Sequoia was an A-round investor in both Meituan and Dianping — the harder they fought, the more money both sides burned. The logic was simple: neither could kill the other, and if they kept fighting, neither would survive to IPO.

The price wars between Didi and Meituan ended in mergers, sharing a common premise: neither side was making money.

The landscape between OpenAI and Anthropic is different. Anthropic is already profitable. Its refusal to engage in a price war isn't because it can't win; it's because it doesn't need to. OpenAI is more like Didi — large scale, heavy cash burn, wanting to go public, and needing to stop the bleeding before an IPO. Didi's path was a merger, but that path isn't available to OpenAI. Microsoft and Amazon are the largest investors in each, respectively, but what they truly care about is how much compute their own cloud platforms sell — AWS Bedrock distributes Anthropic's models, skimming off a large chunk of profit. A merger won't happen: customer lock-in is too deep on both sides, it would create an antitrust nuclear bomb, and Anthropic is already profitable with no incentive to merge.

Which Path is Altman Taking?

Three historical precedents point to the same question: Did the price-cutter ultimately win?

Intel won for a decade but lost to a better product. AWS won because its efficiency was real. Didi merged because both sides were losing money — but Anthropic is already profitable, so that path doesn't exist.

If Altman's inference cost advantage is real — not accounting optimization, not selective disclosure, but a systematic reduction of unit inference costs to below one-third of Anthropic's — then this isn't a price war, it's an AWS-style landscape shift. No matter how much Anthropic emphasizes that "enterprise clients don't care about price," it can't indefinitely maintain its premium when a competitor offers a 75% discount. Enterprise clients may not care about price, but CFOs care about costs.

If the "savings" aren't enough — if the price cuts are funded primarily by financing injections — the capital market will ask him the question he can't answer: Your price is already half of your competitor's, so why are you still losing money?

Anthropic bets that enterprise AI procurement differs from cloud procurement — the depth of Claude Code's integration into development workflows, the trust dividend accumulated by Anthropic's refusal to remove safety guardrails for the Pentagon, and the "a few dollars per million tokens" cost are fundamentally different competitive dimensions. OpenAI bets on scale — 900 million weekly active users, the default model in the Microsoft Office suite, and the largest developer community. When prices hit a critical point and token consumption expands exponentially, it bets on capturing the largest share of incremental demand.

Both bets will face their tests.

Time Will Answer for Him

In the coming months, these bets will be tested one by one.

August 31st marks the first checkpoint. Anthropic's limited-time discount on Sonnet 5 expires, with prices reverting from $2/$10 to $3/$15. If Anthropic doesn't extend the offer or signal further cuts, it's telling the market: I won't play. If enterprise client renewal rates and average revenue per customer remain stable in Q3, it will be proving with numbers that the question of "whether to follow" is itself invalid.

October is the second. Anthropic's IPO window — the first public market yardstick for the AI industry. A price tag of $965 billion or $1 trillion will directly determine all subsequent valuation negotiation space for OpenAI. OpenAI's own IPO has been advised to be postponed until 2027. The $122 billion funding round completed in March has been secured, but if further funding is needed, the implied IPO timeline pressure and valuation commitments in the terms will expose just how much "saved" leverage Altman actually has.

Then come the earnings reports. Anthropic will be watched for sustained profitability; OpenAI, for narrowing losses.

Finally, customers. If a Fortune 500 company switches from Anthropic to OpenAI, even a single case will be seized upon by the market as the first sign that the price war is breaking the premium logic.

Altman's real adversary isn't Anthropic; it's time.