Denise Dresser took over as OpenAI's Chief Revenue Officer in late 2025. On Sunday, April 12, she sat down to write an internal memo laying out the company's Q2 strategy. By Monday morning, The Verge had the full text.
What reporters found inside was not the usual all-hands note. It was a blueprint for a company trying to publicly break with its most important partner, publicly undercut its closest competitor, and publicly reposition itself as something bigger than a model vendor, all at the same time.
The memo named names. It attacked accounting. It revealed a product roadmap. And within hours of hitting the press, it triggered a wave of questions from OpenAI's own investors about whether the company's $852 billion private valuation still makes sense.
The $8 Billion Accusation
The most explosive line in Dresser's memo was a direct accusation against Anthropic. She told staff that Anthropic's widely reported $30 billion annualized run rate is wrong.
The gap, by her calculation, is roughly eight billion dollars. The reason, she wrote, is that Anthropic books gross revenue on sales routed through Amazon Web Services and Google Cloud, rather than netting out the cloud providers' cut.
Translated: when an enterprise buys Claude access through AWS Bedrock, Anthropic records the entire invoice as its own revenue, including the portion AWS keeps. OpenAI, by contrast, reports its Microsoft-routed sales net of Redmond's share.
Both methods comply with US GAAP. They produce very different numbers.
The dispute matters because both companies are sprinting toward IPOs. Anthropic targeted a $60 billion IPO in October 2026, and OpenAI told Wall Street last week it would reserve IPO shares for retail investors. The opening price on both offerings will turn on whose revenue number the market believes.
If Dresser's claim holds, Anthropic's real run rate sits closer to $22 billion.
That would put the company behind OpenAI's roughly $24 billion run rate, rather than ahead of it. Bloomberg first reported Anthropic crossing the thirty billion threshold earlier this year.
Fortune noted on Monday that both companies now self-report monthly revenues of about two billion dollars each.
Anthropic pushed back through the Financial Times. A person close to the company told the FT that Anthropic "recognises gross revenue on sales through partners because it is the principal in the transaction and its cloud partners are the distribution channel," a standard justification for gross recognition.
The accounting is technical. The timing is not. Both companies are racing toward dual IPOs, and every dollar of reported run rate will be priced into their opening valuations.
The Quote That Went Nuclear
Dresser did not stop at accounting. She characterized Anthropic's public positioning in language usually reserved for political opponents.
"Their story is built on fear, restriction, and the idea that a small group of elites should control AI." — Denise Dresser, Chief Revenue Officer, OpenAI (internal memo, April 12, 2026, via The Verge)
For a CRO to write that line in a memo that will obviously leak is not an accident. Anthropic has built its enterprise pitch around safety, interpretability, and limited deployment of its most powerful systems. Dresser was telling her own sales team, and any reporter who happened to see the memo, that OpenAI now plans to frame that positioning as gatekeeping rather than responsibility.
It is a sales playbook written in public.
Microsoft, The "Foundational" Partner, Got Named Too
The second target was OpenAI's most important backer. Dresser wrote that the Microsoft partnership has been "foundational to our success" but "has also limited our ability to meet enterprises where they are," which, she added, for many customers now means Amazon Web Services' Bedrock platform.
She described demand from OpenAI's newly expanded Amazon deal as "frankly staggering." She told staff the company is "firing on all cylinders to establish this as a scaled distribution channel."
The context behind those sentences matters. Amazon committed up to $50 billion in investment and compute to OpenAI in the deal that was first reported on April 1, and struck terms to become the exclusive third-party cloud distribution provider for Frontier, OpenAI's new agent platform. That pact had already strained the Microsoft relationship. The Sunday memo made the strain official.
The Irish Times reported Monday that OpenAI investors are now openly questioning the company's $852 billion valuation, citing both the Anthropic dispute and the public cooling toward Microsoft.
The Product Reveal Hidden in the Memo
Sandwiched between the attacks was a roadmap. Dresser revealed the existence of a new OpenAI model codenamed Spud, which she called "an important step in the intelligence foundation for the next generation of work." Early customer feedback, she wrote, points to stronger reasoning, better handling of intentions and dependencies, and more reliable production results.
She named three other products: Frontier, the agent platform Amazon now distributes; DeployCo, a deployment engine OpenAI is building to sit alongside "Frontier Alliance" partners; and an expanded ChatGPT for Work offering that would bundle with Codex, the company's software development surface.
The through-line was explicit. Dresser told staff: "We should stop thinking like a company with separate product lines. That is how we move from product vendor to operating infrastructure."
How It Unfolded
What This Actually Changes for Practitioners
For engineers building on top of these platforms, the memo is not gossip. It is a forward indicator.
Three things shift if the strategy in the memo holds.
- Where enterprise AI gets deployed. Until this quarter, Azure OpenAI Service was the default path for regulated industries to get GPT access. Dresser's framing, that Microsoft "limited" OpenAI's reach, signals AWS Bedrock will get first-class access to Spud and Frontier. Teams that standardized on Azure may find parity lagging behind Bedrock on new model releases.
- What Spud actually is. The description in the memo, "intelligence foundation for the next generation of work," maps onto what industry reporting has been calling GPT-5.5 or GPT-6. Industry outlets now point to an April 16 preview window. Whether that holds or slips, the product name is public and the positioning is reasoning plus long-horizon agent work.
- How agent platforms compete. Frontier is OpenAI's bid to make itself the default enterprise agent runtime. Anthropic's Model Context Protocol crossed 97 million installs in March. Google is pushing the Agent Development Kit. Meta is pitching Llama Stack. The space that looked settled six months ago is now openly contested.
The Other Side
Not every industry voice read the memo as a power move. Several commentators described it as a tell.
Analysts at The Next Web and The Irish Times pointed out that OpenAI chose to publicly attack a rival's accounting in the same week its own investors began questioning its $852 billion valuation. The usual playbook when a company is winning is to let the numbers do the talking. Writing a memo that obviously leaks does the opposite.
The Financial Times' source near Anthropic defended the gross revenue method as standard for principal-in-transaction accounting under ASC 606 and IFRS 15. Under that framework, a company that controls the service before it is transferred to the customer, and that sets the price, recognizes the full invoice, even when payment flows through a reseller.
Axios' Ina Fried framed the sharpest read: "OpenAI rips Anthropic, distances itself from Microsoft." That is not the posture of a company confident in its quarter. It is the posture of a company trying to reset the narrative before an IPO window closes.
Microsoft and Amazon both declined to comment on the memo. Anthropic has not publicly responded to the "fear, restriction, and elites" framing.
The Bottom Line
A CRO who writes a Sunday memo attacking a competitor's accounting, a partner's limitations, and previewing a model codename at the same time is not sloppy. She is running a strategy. The strategy is to reset three conversations at once: that Anthropic is ahead on revenue, that Microsoft is still the center of OpenAI's distribution, and that OpenAI is a product company.
By Monday, all three conversations had shifted. By Tuesday, investors were asking whether a company that needs to leak its own playbook is the same company worth $852 billion.
The cleanest read on what happened this week came from the memo itself. Dresser wrote: "That is how we move from product vendor to operating infrastructure."
The industry is about to find out whether it believes her.
Sources
- OpenAI rips Anthropic, distances itself from Microsoft (Axios, April 13, 2026)
- OpenAI touts Amazon alliance in memo, says Microsoft has 'limited our ability' to reach clients (CNBC, April 13, 2026)
- OpenAI revenue chief accuses rival Anthropic of goosing revenue projections (Fortune, April 14, 2026)
- OpenAI's leaked memo says new "Spud" model will make all its products "significantly better" (The Decoder, April 13, 2026)
- OpenAI Memo Accuses Anthropic of $8B Run-Rate Inflation (Implicator AI, April 14, 2026)
- OpenAI investors question $852bn valuation as strategy shifts (The Irish Times, April 14, 2026)
- OpenAI Exec Reveals New Strategy in Leaked Memo: Attack Anthropic (Gizmodo, April 14, 2026)
- OpenAI Says Microsoft 'Limited' Its Reach, Slams Anthropic For 'Inflated $30B Revenue' (Benzinga, April 14, 2026)