OpenAI chief revenue officer Denise Dresser told employees Sunday that rival Anthropic is overstating its $30 billion run rate by roughly $8 billion through gross revenue accounting, according to a four-page internal memo viewed by CNBC and The Verge. The accusation turns on how each company books cloud-partner sales: Anthropic counts the full amount billed through AWS, Azure and Google Cloud as top-line revenue, while OpenAI reports its Microsoft share net of Redmond's cut. Both approaches comply with US GAAP, but the gap would reshape the scoreboard months before both companies are expected to file for dual IPOs, with OpenAI targeting a window as early as the fourth quarter.
"Their stated run rate is inflated," Dresser wrote, per the memo text published by The Verge. "They use accounting treatment that makes revenue look bigger than it is, including grossing up rev share with Amazon and Google." OpenAI's own analysis, she added, puts Anthropic's true run rate closer to $22 billion.
Key Takeaways
- OpenAI CRO Denise Dresser's four-page memo to employees accuses Anthropic of inflating its $30B run rate by roughly $8B through gross revenue accounting on cloud-partner sales.
- OpenAI reports Microsoft share net, putting it closer to public-company standards, while Anthropic books full AWS, Azure and Google Cloud billings as top-line revenue.
- A parallel investor memo projects OpenAI at 30 gigawatts of compute by 2030 versus 7–8 gigawatts for Anthropic by end-2027, an argument Anthropic disputes.
- Ramp data shows Anthropic at 30.6% of enterprise AI-paying customers versus OpenAI's 35.2%, with a projected crossover within two months ahead of dual IPOs.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
An accounting fight dressed as strategy
Anthropic pushed back to CNBC, arguing it recognizes gross revenue because it is the principal in the transaction while cloud partners serve as the distribution channel. That position is defensible under GAAP. It is also the kind of gross-versus-net question a public-company auditor would probe line by line during an S-1 review.
Dresser knows that. "We report Microsoft revshare net, which is more inline with standards we would be held to as a public company," she wrote. The subtext: when Anthropic files its prospectus, the $30 billion headline it began touting earlier this month may get restated. Anthropic recently told reporters its run rate had jumped from $9 billion at the end of 2025 to more than $30 billion by early April, a trajectory that anchored its $380 billion February valuation.
The memo dropped the same week Ramp's corporate-card data showed Anthropic at 30.6 percent of AI-paying business customers versus OpenAI's 35.2 percent, with Anthropic projected to pass OpenAI within two months. Dresser is not writing to steady employees who feel comfortable. She is writing to employees who can read the same charts she does.
Thirty gigawatts versus eight
The compute section reads like a second front. OpenAI told its investors four days earlier that Anthropic is "operating on a meaningfully smaller curve," projecting 30 gigawatts of OpenAI capacity by 2030 against 7 to 8 gigawatts for Anthropic by end-2027. Today OpenAI runs roughly 1.9 gigawatts. Anthropic runs 1.4. "Even at the high end of that range, our ramp is materially ahead and widening," the investor memo read.
Futurism's Victor Tangermann framed the brag more sharply: half the US data centers slated to open are delayed or canceled, and OpenAI's $600 billion infrastructure pledge is already less than half of what it originally promised. The 30-gigawatt line is projection, not installed base. It also sets a capital wall rivals would need hundreds of billions to clear.
Get Implicator.ai in your inbox
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
Anthropic's CFO Krishna Rao answered with a compute deal through Broadcom and Google: five gigawatts of next-generation TPU capacity starting 2027. "We are making our most significant compute commitment to date," Rao said.
The enterprise door is already ajar
Dresser's memo frames five priorities, and four of them circle the same problem: OpenAI's enterprise moat is shallower than its valuation implies. She concedes the Microsoft partnership "has also limited our ability to meet enterprises where they are." For many, she wrote, "that's Bedrock." The fix is the February Amazon tie-up, worth up to $50 billion, which Dresser says has generated "frankly staggering" inbound demand.
That is an admission, not a victory lap. Anthropic's Claude drew the biggest crowds at last week's HumanX conference in San Francisco, where Glean CEO Arvind Jain called it "Claude mania." Implicator's earlier reporting documented Anthropic winning roughly 70 percent of new enterprise deals in late Q1, a slide that triggered the "side quest" cleanup Dresser now oversees. Codex has grown to 3 million users in a quarter, from near zero. Claude Code still draws the developer reverence the memo cannot replicate by press release.
What Dresser is really asking for
The memo closes with staff instructions more than strategy. Stay focused on customers. Treat compute as a moat. Treat Anthropic's accounting as a tell. Dresser came from Slack and Salesforce, and she writes like someone who has been inside a public-company earnings cycle. Her employees have not.
That gap matters. OpenAI is pitching a retail IPO slice, a $122 billion private round it already closed at an $852 billion valuation, and a $600 billion compute spend through 2030. Anthropic is pitching Claude Code, Project Glasswing and a disciplined compute narrative. The gross-versus-net fight is the first public argument over how either story will survive an auditor.
The market is noisy, Dresser wrote. It is also counting.
Frequently Asked Questions
What exactly does OpenAI's memo claim about Anthropic's revenue?
OpenAI CRO Denise Dresser wrote that Anthropic's stated $30 billion run rate is overstated by roughly $8 billion because it books gross revenue from AWS, Azure and Google Cloud sales, including the cloud providers' cut. OpenAI's analysis puts the comparable figure closer to $22 billion.
Is Anthropic's accounting actually wrong?
No. Both gross and net approaches are permitted under US GAAP. Anthropic says it recognizes gross revenue because it is the principal in the transaction while cloud partners serve as distribution. The dispute is which treatment auditors and public markets will accept during an S-1 review.
How big is the compute gap between OpenAI and Anthropic?
OpenAI projects 30 gigawatts of capacity by 2030, against 7 to 8 gigawatts for Anthropic by end-2027. Current capacity is roughly 1.9 gigawatts for OpenAI and 1.4 gigawatts for Anthropic. Anthropic counters with a Broadcom and Google TPU deal worth 5 gigawatts starting 2027.
Why does this memo matter for the IPO timeline?
Both companies are preparing for public offerings potentially this year, with OpenAI targeting a window as early as Q4. Public markets value growth companies on run rate and growth rate, so any restatement of Anthropic's $30 billion figure would directly affect its valuation narrative against OpenAI's $852 billion mark.
Who is Denise Dresser and why is she writing this now?
Dresser is OpenAI's chief revenue officer, hired in December 2025 after serving as Slack CEO and holding executive roles at Salesforce. She recently absorbed responsibilities from former COO Brad Lightcap. She is writing now because Ramp data shows Anthropic is on track to pass OpenAI in enterprise customer share within two months.
AI-generated summary, reviewed by an editor. More on our AI guidelines.


IMPLICATOR