OpenAI internal memo leaked: directly calls out Anthropic’s fear-based narrative and “compute is behind, revenue is inflated”

AI competition has entered a white-hot stage. According to an internal memo obtained by The Verge, Denise Dresser, OpenAI’s revenue chief, told employees: “This is the most intense competitive market I’ve ever seen.” Spanning four pages, the document not only reveals OpenAI’s strategic shift, but also unusually launches systematic criticism at its main rival, Anthropic, showing that the AI industry has escalated from a model race to a full-scale war over platforms and compute power.

From product to platform: OpenAI wants users to be “unable to leave”

The memo’s core message is very clear: OpenAI must strengthen its enterprise market while also increasing user stickiness. Dresser noted that in the current environment, users can switch among different models almost anytime, which makes it difficult for a single product to build a long-term advantage. As a result, the company’s strategy is shifting toward building a “moat,” increasing switching costs through integration across multiple products.

She emphasized, “Using multiple products will make it harder for us to be replaced,” and added that OpenAI should no longer operate with an independent-product mindset, but instead transition into “an enterprise platform with multiple entry points, but integrated as one.” This means that the core of AI competition is shifting from model performance to platform integration capabilities.

Internal memo blasts back: Anthropic’s “fear narrative” and strategic missteps

In the memo, OpenAI’s criticism of Anthropic is quite direct, even rising to the level of ideas and storytelling. Dresser pointed out that Anthropic’s arguments are built on “fear and limitations,” and they implicitly hold the view that AI should be controlled by a small group of elites. By contrast, OpenAI advocates a more proactive and open approach, including building powerful systems, establishing safety safeguards, and expanding AI usage permissions so that more people can create value.

Beyond the difference in philosophy, OpenAI also directly calls out Anthropic’s strategic missteps, especially shortcomings in its compute-power strategy and deployment. The memo says that failing to secure sufficient compute resources early has been directly reflected in product experience, including throttling, declining availability, and stability issues—differences that enterprise customers can clearly feel.

OpenAI emphasized that it had already recognized the trend of compute scaling exponentially earlier and laid the groundwork in advance, giving it a tangible advantage in the current competition.

Anthorpic’s single-point advantage is hard to sustain in the platform war

On product strategy, OpenAI also points out Anthropic’s structural limitations. The memo states that Anthropic initially entered the market through its coding capabilities, successfully establishing an early advantage. However, in platform competition, a single-product positioning will become a long-term burden.

As AI applications expand from developers to every department in enterprises, work workflows, and industry scenarios, a strategy focused solely on coding will struggle to support comprehensive competition. Dresser clearly stated that in a platform war, you don’t want to be a company with only one product.

OpenAI accuses Anthropic of “inflating” revenue by $8 billion

Besides product and technology, OpenAI also questions Anthropic’s financial figures. The memo says that Anthropic’s reported annualized revenue (run rate) to the outside has “inflation” issues—making the numbers look higher than actual performance through accounting treatment. This includes revenue-sharing arrangements with cloud partners (such as Amazon and Google) being recorded using “grossing up.”

OpenAI’s internal analysis believes this approach causes its annualized revenue to be overestimated by about $8 billion. In comparison, OpenAI’s partnership with Microsoft records revenue on a “net basis,” and the company believes this better aligns with the accounting standards required for companies preparing for a future IPO.

The compute war’s key gap: 30GW vs 8GW

In another investor memo, OpenAI further emphasizes its compute advantage. The company expects to reach 30GW of compute power by 2030, while its estimate for Anthropic is about 7 to 8GW by 2027. OpenAI noted that even under a high-estimate scenario, its expansion speed is still clearly ahead.

The company also presents an argument for a “compounding advantage”: stronger infrastructure can train higher-capability models, and model optimization along with hardware progress continues to reduce costs, creating a positive feedback loop.

Currently, both OpenAI and Anthropic are reported to have plans to kick off IPOs this year. As pressure from capital markets increases, the two companies not only need to prove technical leadership, but also need to convince investors that their business model is sustainable.

This article reveals an internal OpenAI memo: it directly targets Anthropic’s fear narrative and “compute lag, revenue inflation.” First published on Chain News ABMedia.