It was a memo that promised strategy and handed out bruises instead. Employees opened Denise Dresser’s note expecting roadmaps; they found a running critique of Anthropic and a reopening of corporate warfare. You could feel the room tilt.
I read the memo so you don’t have to. Here’s what matters: OpenAI’s chief revenue officer publicly questioned Anthropic’s revenue math, resurfacing a rivalry that now runs through cloud deals with Amazon and Google, hardware shipments from Broadcom, and the old Microsoft axis. I’m going to walk you through the claims, the counterclaims, and what they mean for enterprise customers, investors, and anyone who uses models from these firms.
Employees found a memo that accused a rival of inflating revenue
The observation: staff expecting guidance instead received an attack on Anthropic’s accounting choices. Dresser’s memo, reported by The Verge, claims Anthropic has been “grossing up” revenue-sharing deals with Google and Amazon to present a larger top-line number.
That matters because last week Bloomberg reported Anthropic’s annualized run rate north of $30 billion (€28 billion). OpenAI pushes back hard: Dresser says Anthropic’s figure is inflated by about $8 billion (€7 billion), putting it nearer $22 billion (€20 billion). OpenAI’s own cited run rate sits around $24 billion (€22 billion), which gives the company rhetorical—and possibly market—momentum.
How much revenue does Anthropic have?
Numbers differ by definition. If you count grossed-up revenue before revenue shares are removed, Anthropic can show the larger number that Bloomberg cited. If you use net revenue after partner cuts, as Dresser argues OpenAI prefers, that headline drops by roughly $8 billion (€7 billion) in her framing. I read those two positions as competing narratives: one emphasizes scale; the other emphasizes margins and customer economics.
Customers and cloud deals exposed a strategic squeeze
The observation: Amazon has legs in both camps now, and Microsoft isn’t the only gatekeeper.
Amazon announced plans to invest up to $50 billion (€46 billion) in OpenAI, a move that shifts balance in commercial distribution and cloud provisioning. Anthropic, meanwhile, has close ties to Google (and access to Google TPU chips via Broadcom partnerships) and still sells through Amazon and Google channels. When Dresser claims Anthropic is “grossing up” revenue, she’s attacking the optics that make Anthropic look like the enterprise winner—particularly in coding—when, in her view, the economics are different.
Is Anthropic bigger than OpenAI?
Depends on your metric. By gross run rate, Anthropic may appear larger. By net economics and certain enterprise contracts, OpenAI argues it leads. I think the fight is as much about perception as it is about dollars: customers choose vendors based on product fit, cloud integration, and trust. The trust equation now includes public memos and press narratives as much as performance.
Leadership and the politics of control shook the memo’s tone
The observation: the memo didn’t only argue numbers; it attacked culture.
Dresser called Anthropic “built on fear, restriction, and the idea that a small group of elites should control AI.” That’s a heavy charge coming from OpenAI, which itself has publicly courted influence—Sam Altman once admitted he sleeps better knowing he can exert influence—and has spent on lobbying to shape federal oversight. The accusation reads like a claim on moral high ground, but the real contest is over who gets to set rules and who pays for the gatekeepers.
Why is OpenAI attacking Anthropic?
There are practical stakes. Market share matters for cloud pipelines (AWS, Google Cloud, Microsoft Azure), for model training economics, and for enterprise contracts. Dresser frames the attack as corrective: calling out accounting treatment weakens Anthropic’s sales narrative. I read it as competitive strategy dressed in regulatory rhetoric—an attempt to persuade customers and partners that OpenAI’s figures and alignment are more reliable.
You should watch three places: deal structures with Amazon and Google, third-party verification of run rates, and how enterprise customers respond in procurement cycles. For users, the argument affects pricing, support, and integration; for investors, it affects valuations and future rounds; for vendors like Broadcom and Google, it affects where TPU and chip allocations land.
Think of the memo as a courtroom cross-examination meant to unsettle a witness, and a wedge splitting a boardroom conversation over how revenue really flows. The question now is whether this strategy wins customers or just prolongs a public feud—will you bet on numbers or on narrative?