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

Original Source: Wall Street CN

From the last week of June to early July, the entire AI track was collectively repriced.

Anthropic struck first with Sonnet 5, slashing prices to $2/$10 in a limited-time offer. Over a week later, xAI, OpenAI, and Meta launched a flurry of releases 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 moved. On July 15th, Altman added a note on X: Sol is already half the price of Fable 5, "we'd be happy to deliver at a quarter of the price." A further 75% cut.

The company cutting prices lost $9.3 billion in one quarter. The one charging a premium started turning a profit.

This doesn't look like a price war should. Three almost identical historical battles have already scripted the template for this one.

The Expensive One is Actually Making Money

Anthropic's annualized revenue surged from $9 billion at the end of last year to $47 billion by May this year. OpenAI topped $30 billion during the same period. In Q1, OpenAI reported $5.7 billion in revenue, while Anthropic brought in $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 posted an operating loss of $9.3 billion in Q1.

The one cutting prices is bleeding money. The one selling at a premium is now profitable.

The reason lies buried in the revenue structure. Approximately 85% of Anthropic's revenue comes from enterprise clients — the number paying over $1 million annually has already surpassed 1,000. These clients buy stability, security, and compliance, not the lowest token price. A 75% price cut won't make them buy a single token more; a 75% price hike won'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 shouted "another 75% cut" on X on July 15th, Anthropic didn't reply — it had its executives publicly urge companies "not to reduce AI usage due to cost concerns." It's not a non-response; it's a response in a different language.

Thirty Years Ago, Intel Tried This Move

In 1998, AMD's market share jumped from 8% to 16%, posing the first real threat to Intel. Intel's response: price cuts described by semiconductor analysts as "unprecedented" — not just one round, but years of scorched-earth assaults. Every time AMD launched a competitive chip, Intel unleashed another wave of price cuts. The logic was simple: 13 fabs against AMD's 2 created a cost structure competitors couldn't replicate.

AMD survived. It was forced to spin off its fabs, but it did something Intel didn't anticipate: it stopped imitating and redefined the product. The Ryzen launch in 2017 turned AMD around through product superiority, and its market cap once surpassed Intel's.

Intel used its cost advantage to keep AMD pinned to the floor for years. But cost advantages have an expiration date; product advantages don't.

What Altman is doing is no different from Intel in 1998. He lists the technology to compress reasoning costs by 50% as "core trade secrets." The Information, citing insiders, reported: "They don't even want other employees inside OpenAI to know, because if it leaks, other labs will quickly adopt it." The key word isn't "secret," it's "adopted" — once competitors master the same efficiency tools, his only weapon becomes burning cash. Intel didn't have to burn cash back then; its price cuts relied on having more fabs.

Flywheel or Pump?

From 2006 to 2018, AWS proactively cut prices over 100 times. S3 storage costs dropped 85% over 12 years. But every time the price cut announcement was made, AWS was profitable.

The flywheel turns for one simple reason: every penny saved came from real efficiency improvements — 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 even lower. That's a flywheel.

Altman wants to tell exactly this story. "50% compression in reasoning costs," "core trade secrets" — every phrase sends the same signal: this isn't burning cash to grab market share; it's returning the cost savings from technological progress to customers.

But one number punctures the narrative. OpenAI's Q1 operating loss was $9.3 billion; for every $1 of revenue, it lost $1.6. AWS never lost money like this while cutting prices. The flywheel premise is that the price curve and cost curve move down in sync, or at least the cost curve moves down faster. If the cost curve hasn't kept pace — if every price cut uses funding to cover the spread — it's not a flywheel; it's a pump.

At the pump's intake, Altman has the $122 billion funding round completed in March and $73 billion in cash on hand; at the outlet, an IPO. If the pump stops, the water runs dry. The bigger problem: by the end of 2025, OpenAI's total commitments for cloud computing power procurement are as high as $665 billion, spanning through 2030 — whether AI demand grows as expected or not, that money must be paid out.

The Cash Burners Have No Way Out

In 2014, the subsidy war between Didi and Kuaidi lasted less than half a year and burned through $2.4 billion, losing tens of millions daily at its peak. They announced a merger on Valentine's Day 2015. In October of that same year, Meituan and Dianping merged — the most candid line in Wang Xing's internal email on merger day: "Yesterday we fought fiercely; today we shake hands."

It wasn't the founders who drove the merger; it was the investors. Sequoia was a Series A 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 continuing the fight meant neither would survive to an IPO.

The price wars of Didi and Meituan both ended in mergers, sharing one common precondition: neither side was making money.

The dynamic between OpenAI and Anthropic is different. Anthropic is already profitable; its refusal to join the price war isn't because it can't compete, but because it doesn't need to. OpenAI is more like Didi — large scale, huge cash burn, eager to go public, and must stop the bleeding before an IPO. Didi's solution was a merger, but OpenAI doesn't have that option. Microsoft and Amazon are their respective largest investors, but what they really care about is how much computing power their own cloud platforms sell — AWS Bedrock distributes Anthropic's models, skimming off a significant slice of profit. A merger won't happen: customer lock-in is too deep, it would create antitrust nuclear bomb, and Anthropic, already profitable, has no incentive to merge.

Which Path is Altman Walking?

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

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

If Altman's reasoning cost advantage is real — not accounting tricks, not selective disclosure, but systematically compressing unit reasoning costs to below one-third of Anthropic's — then this isn't a price war; it's an AWS-style reshaping of the landscape. No matter how much Anthropic emphasizes "enterprise clients don't care about price," it can't infinitely maintain a premium when a competitor is three-quarters cheaper. Enterprise clients may not care about price, but CFOs care about costs.

If the "savings" aren't enough — if the price cuts mainly rely on cash infusion — the capital market will ask him the question he can't answer: Your price is already half your competitor's, so why are you still losing money?

Anthropic bets that enterprises buying AI is different from buying cloud — the depth of Claude Code embedded in development workflows, the trust dividend from Anthropic's refusal to remove safety guardrails for the Pentagon, and "a few dollars per million tokens" have never been the same dimension of competition. OpenAI bets on scale — 900 million weekly active users, the default model in the Microsoft 365 suite, the largest developer community. When prices hit a critical point and token consumption expands exponentially, it bets on capturing the largest share of that incremental growth.

Both bets are about to be tested.

Time Will Answer for Him

Over the coming months, these wagers will be verified one by one.

August 31st is the first marker. Anthropic's limited-time offer on Sonnet 5 expires, with prices returning from $2/$10 to $3/$15. If Anthropic doesn't extend the offer or signal further price cuts, it's telling the market: I'm not following. If Q3 enterprise client renewal rates and average contract values remain stable, it's proving with numbers that the very question of "to follow or not to follow" is invalid.

October is the second. Anthropic's IPO window — the AI industry's first public market yardstick. A valuation of $965 billion or $1 trillion directly determines all subsequent valuation leverage for OpenAI's negotiations. OpenAI's IPO has already been advised to be postponed until 2027. The $122 billion funding round in March is secured, but if further financing is needed, the implicit IPO timeline pressure and valuation commitments in the terms will reveal how many "saved" chips Altman really has.

Then come the earnings reports. Anthropic's profitability sustainability will be scrutinized; OpenAI's loss narrowing will be watched.

Finally, the customers. If a Fortune 500 company switches from Anthropic to OpenAI, even a single case, the market will interpret it as the first signal that the price war is breaking the premium logic.

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