OpenAI's Memo Reveal It Is Losing the AI War to Anthropic

OpenAI's chief revenue officer sat down last Monday to write a strategy memo for her sales team. It was not supposed to leave the building. But it was the fourth internal document leak in six months.

The memo, obtained by The Verge, is one of the most direct financial attacks one AI lab has publicly lodged against another. CRO Denise Dresser accused rival Anthropic of inflating its reported $30 billion revenue run rate by roughly $8 billion — claiming the company counts partnership payments from Amazon and Google before splitting them, not after, which flatters the headline number considerably. 

Anthropic did not respond to a request for comment. OpenAI did not either. Neither company is publicly traded, so nobody outside their boardrooms can say for certain who is right. 

How OpenAI Got Here

Eighteen months ago, OpenAI owned half the enterprise AI market. Today it owns a quarter. Anthropic has gone from 12 percent to 32 in the same stretch — and among companies signing new AI contracts as of March 2026, 65 percent went with Anthropic. OpenAI drew 32. That kind of slide, sustained over more than a year, is not a bad quarter. It is a structural problem. 

Also read: Why Claude Mythos May Be Anthropic’s Most Dangerous Model Yet  

CEO Sam Altman responded in December by declaring a company-wide Code Red, scrapping a $1 billion advertising push mid-build and redirecting everything back to ChatGPT. A separate Altman memo, obtained by the Wall Street Journal, conceded that Google had been "doing excellent work in every single aspect" while OpenAI had lost ground on the infrastructure that determines how capable its models can actually become.

The numbers were bad enough. Then March happened.

When It Turned Personal

After the Pentagon designated Anthropic a supply-chain risk, OpenAI moved within hours to sign the military contract Anthropic had just refused. Anthropic CEO Dario Amodei responded with a 1,600-word memo to his staff, leaked to The Information, calling OpenAI's safety record "safety theater" and its public statements "straight up lies." The reason the administration came after Anthropic, he wrote, was not complicated: "We haven't donated to Trump, while OpenAI/Greg have donated a lot." 

Also read: Anthropic wins first court fight over Trump and the Pentagon blacklist.    

OpenAI president Greg Brockman and his wife had given $25 million to the MAGA Inc super PAC. Altman gave $1 million to Trump's inauguration. Amodei later walked back the tone, though not the substance.

What OpenAI Is Promising Investors Instead

Monday's memo went beyond the swipe at Anthropic's books. Dresser laid out a new model codenamed Spud — described internally as making all of OpenAI's products "significantly better" and an agent platform called "Frontier," built to weave OpenAI into enterprise operations so thoroughly that pulling it out becomes more trouble than it's worth. The goal, Dresser wrote, is to go "from product vendor to operating infrastructure." That is a compelling pitch. It is also exactly the kind of pitch that needs clean, defensible revenue figures behind it before a company rings the opening bell.

For a company that wants to run the world's infrastructure, it keeps leaving its own doors unlocked.