OpenAI is scared of Anthropic. That is the only honest reading of a shareholder memo that CNBC obtained this week, in which OpenAI explicitly trash-talked its biggest rival to investors.

The memo claims Anthropic is "operating on a meaningfully smaller curve" and positions OpenAI's compute strategy as the winner. OpenAI says it will have 30 gigawatts of compute by 2030, while Anthropic will top out at 7 to 8 gigawatts by the end of 2027. "Even at the high end of that range, our ramp is materially ahead and widening," OpenAI wrote.

Here is the problem with that argument: Anthropic just tripled its annual revenue run rate to $30 billion. It signed a massive long-term deal with Google and Broadcom for TPU access. It launched Project Glasswing, a cybersecurity initiative so powerful that the Treasury Secretary and the Fed Chair called an emergency meeting about it. And investors are falling over each other to get in on the IPO.

OpenAI, meanwhile, is burning $85 billion a year, has lost three top executives in a single week, and just bought a media company run by the man who coined the phrase "vast right-wing conspiracy." Its CFO and CEO cannot agree on the path to profitability.

The memo reads like insecurity dressed as strategy. When you are actually winning, you do not send your investors a letter explaining why your competitor is losing. You just ship the numbers. Anthropic shipped the numbers. OpenAI shipped the letter.

The deeper story is about compute philosophy. OpenAI is betting that raw scale wins. More gigawatts, more tokens, more everything. Dario Amodei has publicly described Anthropic's approach as "deliberately conservative" on compute. That sounds like a weakness until you realize Anthropic's models are matching or beating OpenAI's on most benchmarks with a fraction of the infrastructure.

The compute war is the new arms race, and both companies are now gearing up for IPOs that could happen this year. They are collectively valued at over $1 trillion. The pitch to Wall Street is essentially: "We can burn money faster and smarter than the other one."

But here is what the memo really reveals: OpenAI knows the narrative is shifting. A year ago, nobody questioned who was winning the AI race. Now Anthropic is closing the revenue gap, winning enterprise contracts, and building capabilities that have the federal government calling emergency meetings. That does not happen when you are "operating on a meaningfully smaller curve."

First reported by Bloomberg. Memo viewed by CNBC.