Musk publicly changes his tone on Anthropic
Elon Musk said on X that he had “obviously misjudged Anthropic,” a sharp break from his earlier attacks on the company. Months ago, he had described Anthropic as “hypocritical and evil” and said it “could not win.” In his latest comments, he called Anthropic a leader in AI, said no company had produced models as strong as Mythos/Fable, and predicted Anthropic would soon release Mythos 2.
The exchange was triggered by an X user who argued that SpaceXAI, the name used after xAI was merged into SpaceX, already had a frontier model that could compete with Claude Opus 4.8. The same post said Anthropic relied on compute rented from SpaceXAI, meaning Musk could bring the company down whenever he wanted.
The timing made the response stand out. On July 8, Musk had just released Grok 4.5 to the public and pitched it as faster and cheaper. The article says that by late June he had also claimed Grok 4.5 was getting close to, and might even surpass, Claude Opus in internal testing, while promising a new from-scratch model every month this year.
Still, the piece notes that on the Artificial Analysis intelligence ranking, Grok 4.5 sits behind Fable 5 and GPT-5.5. This time, Musk did not keep insisting otherwise and instead acknowledged that his earlier judgment had been wrong.
Months of criticism before the reversal
The report says Musk had offered almost no public praise for Anthropic since September last year. At that time, he said the company had no path to winning. In January, he mocked the name Anthropic and said it was destined to become “misanthropic.”
In February, after Anthropic announced a $30 billion fundraise at a $380 billion valuation, Musk attacked again, accusing the company of large-scale theft of training data. The article says he also called it self-righteous, sanctimonious, and hypocritical, and said it hated Western civilization. In March, he asked whether there was any company more hypocritical than Anthropic.
Even so, he had at times conceded Claude’s technical strength. The article points to a January complaint in which Musk said Anthropic had cut xAI engineers off from Claude’s latest coding features, while also admitting the company had built something special in programming.
The SpaceX compute contract
The reversal came against the backdrop of a major compute agreement between Anthropic and SpaceX. According to the article, Anthropic agreed in May to take the full capacity of SpaceX’s Colossus 1 data center in Memphis. That amounted to more than 300 MW of power, or more than 220,000 Nvidia GPUs.
On May 6, 2026, Anthropic announced the deal. Citing SpaceX’s S-1 filing, the report says Anthropic committed to pay $1.25 billion per month through May 2029, with either party able to end the agreement on 90 days’ notice.
TechCrunch was cited as saying that by July 2026, Anthropic had already become one of SpaceX’s largest customers. The article frames the arrangement as a win for both sides: SpaceX gained a major customer it could highlight in its filing, while Anthropic locked in multi-year compute supply.
On the day the partnership was announced, Musk said he had spent a lot of time with Anthropic’s leadership to understand how they worked to ensure Claude was beneficial to humanity. He said he was impressed and that no one had triggered his “evil detector.”
Musk says he would not cut service
Responding to the claim that he could shut Anthropic down by cutting off compute, Musk said he would not. Even as a competitor, he said, he would never cut them off in a way that seriously harms them, adding that it was “not my style.”
To support that point, he cited three examples: Tesla open-sourced its patents, the Supercharger network was opened to rivals, and SpaceX launches satellites for competitors without charging more or imposing unfair terms.
The article also notes that Anthropic’s position does not depend only on Musk’s personal style. The agreement includes a 90-day termination notice period, and cutting the deal would also mean giving up $1.25 billion in monthly revenue.
Competition and dependence at the same time
The piece closes by noting another remark Musk made in late May: SpaceXAI was only three years old, half the age of Anthropic and one-quarter the age of OpenAI, so the market should be judged again in three years. In that framing, recognizing a rival’s lead is not the same as conceding defeat.
The broader point in the article is that the AI model race no longer runs only on direct competition. Companies are also tied together by compute, chips, and data centers. Musk’s shift in tone has put that relationship into plain view.