On July 9, an X user posting under the handle @yacineMTB warned Elon Musk, the CEO of Tesla and SpaceX, that he held dangerous leverage over a rival. SpaceXAI, a day earlier, had shipped Grok 4.5, a model the user argued could finally compete with Anthropic's Opus line. Anthropic, meanwhile, was completely dependent on compute it rented from Musk's own company. If Musk wanted to kill Anthropic, the post said, he could.
Musk's reply did not defend that leverage. It surrendered the argument he had spent the better part of a year making.
"I was clearly wrong about Anthropic. They are obviously currently the leader in AI. No company has released a model as good as Mythos/Fable and they will undoubtedly have Mythos 2 ready soon. And I would never cut them off in a way that hurt them badly, even as a competitor. That's not my style." — Elon Musk, on X, July 9, 2026
The admission landed roughly five months after Musk called Anthropic's Claude models "misanthropic and evil." It followed a September 2025 post dismissing the company's chances of winning at all. And it arrived while Anthropic pays Musk's SpaceXAI unit, the company SpaceX formed after absorbing xAI in February, $1.25 billion a month to rent the supercomputer that trains the very models he was now praising.
For Amazon and Alphabet, which have together committed tens of billions of dollars to Anthropic, the reversal is more than gossip about a billionaire's mood. It is a public data point about the asset their money is riding on, delivered by one of its loudest former critics.
What Musk Actually Said
Musk named Anthropic's two current flagship models directly. Claude Fable 5 is the version available to the public. Claude Mythos 5 is the more capable version Anthropic restricts to vetted partners through a program called Project Glasswing. Both launched in June.
His post contained one small, telling error. Anthropic's restricted model is already Mythos 5, not the "Mythos 2" Musk referenced. Whether that was loose shorthand or an honest mixup, Anthropic did not publicly correct him, or respond to his post at all, in the days that followed.
Musk then addressed the sharper part of the exchange: the suggestion that he could weaponize Anthropic's dependence on his compute.
He pointed to Tesla's 2014 decision to open its patents and its Supercharger network sharing with rival automakers. He also cited SpaceX's practice of launching competing satellite constellations "with no increase in price or use of unfair terms." All three, he said, prove that hurting a rival with market power is not his style.
From "Evil" to Landlord in Five Months
The tone shift did not happen overnight. It tracked closely with how much money started moving between the two companies.
| Date | What Musk Said | Context |
|---|---|---|
| September 2025 | "Winning was never in the set of possible outcomes for Anthropic." | Dismissing Anthropic's competitive chances |
| February 12, 2026 | "Your AI hates Whites & Asians... This is misanthropic and evil. Fix it." | Reacting to Anthropic's $30 billion funding round |
| May 6, 2026 | "Nobody set off my evil detector." | After touring Anthropic's operations, the same week as the Colossus deal |
| July 9, 2026 | "They are obviously currently the leader in AI." | Replying to a warning that he could cut off Anthropic's compute |
Between the second and third rows, something concrete changed. SpaceXAI agreed to lease Anthropic the entire 300-megawatt output of Colossus 1, the Memphis, Tennessee supercomputer Musk's team built to train Grok, a deal LDS covered when it was first reported in May.
The Price of the Compute He Once Called Evil
The financial terms of that lease surfaced weeks later, buried in SpaceX's own IPO filing. Anthropic agreed to pay $1.25 billion every month through May 2029, with either side able to exit on 90 days' notice.
That rate adds up to more than 40 billion dollars over the contract's full term.
SpaceXAI has since signed a similar arrangement with Google, which pays roughly $920 million a month for its own slice of SpaceX's infrastructure through June 2029. Musk's compute business now counts two of Anthropic's biggest financial backers among its customers too.
Those backers have their own reasons to want Anthropic to keep winning.
| Investor | Total Committed | What Anthropic Owes Back |
|---|---|---|
| Amazon | Up to $33 billion cumulative since 2023, including new money added in April 2026 | Over 100 billion dollars in AWS chip spending over the next decade |
| Alphabet | Up to $40 billion across its funding rounds, at an April 2026 valuation of 350 billion dollars | Roughly 5 gigawatts of Google TPU capacity over five years |
All three companies, Amazon, Alphabet, and now SpaceXAI, profit when Claude grows, regardless of what Musk says about it in any given week.
Not Everyone Reads This as Sincere
TechCrunch, which first reported the exchange, put the underlying tension in blunt terms: whether Anthropic should trust a competitor to host its models at all.
The outlet's own reporting complicates Musk's pledge. In May, when the Colossus lease was first announced, Musk wrote separately that SpaceX "reserves the right to reclaim the compute" if Anthropic's AI "engages in actions that harm humanity." That clause did not appear in the formal press release for the deal, and it remains unclear whether it appears in the signed contract at all.
The reversal also landed the same week independent benchmarks from Artificial Analysis ranked SpaceXAI's newly launched Grok 4.5 fourth on a leading intelligence index, well behind the "Opus-class" performance Musk claimed for it at launch.
Musk's record with rivals adds further reason for caution. He sued OpenAI and its CEO, Sam Altman, over the company's drift from its nonprofit mission and lost in May, after a jury found he had waited too long to file. During that trial, Musk acknowledged under oath that "generally AI companies distill other AI companies," a reference to training a cheaper model on a rival's outputs, the same practice Anthropic has accused Chinese AI labs of using against Claude.
None of that proves the praise is hollow. It means Anthropic is trusting its infrastructure to a competitor whose history includes both open patents and a federal lawsuit against a different rival, and whose promise not to retaliate is a public statement, not a contract term.
The Bottom Line
Strip away the individual quotes and the sequence tells its own story. A man who called a company evil in February was thanking it in July.
In between, that company became one of the two biggest paying tenants of his supercomputer. Musk's business and his rhetoric moved in the same direction at the same time, and nothing in the public record separates how much of one caused the other.
Anthropic did not need Musk's blessing to be a leading AI lab. Its restricted models were good enough that Washington spent nineteen days this summer deciding whether to let the rest of the world use them, and its investor list already runs to tens of billions of dollars from Amazon and Alphabet alone.
What Musk's post adds is not new information about Claude. It is a live demonstration of how fast a rival's insults turn into a customer's testimonial once the invoices start arriving.
"That's not my style," Musk wrote, about a company he had called evil roughly five months earlier. The style, evidently, depends on the deal.
Sources
- Elon Musk praises Mythos/Fable, promises not to 'cut off' Anthropic — TechCrunch, July 9, 2026
- Elon Musk on X — X, July 9, 2026
- Musk calls rival Anthropic the current frontrunner in AI — Yahoo Finance (Investing.com), July 9, 2026
- Elon Musk Admits He Was 'Clearly Wrong' About Anthropic, and Promises Not to Weaponize SpaceXAI's Compute — Benzinga, July 10, 2026
- Elon Musk admits he was 'clearly wrong' about Anthropic — Teslarati, July 10, 2026
- Elon Musk Says He Was "Clearly Wrong" About Anthropic's AI Models — The Motley Fool, July 13, 2026
- Elon Musk slams Anthropic AI models as 'misanthropic and evil' in scathing social media post — Fox Business, February 12, 2026
- Elon Musk called Anthropic 'evil' 3 months ago. Now he's taking $4 billion to become its data landlord — Fortune, May 7, 2026
- Musk Called Grok 4.5 Opus-Class. Independent Testers Ranked It Fourth. — LDS, July 10, 2026
- Anthropic Will Pay Musk $1.25 Billion a Month for a Supercomputer He Couldn't Fill — LDS, May 22, 2026
- Elon Musk Called Anthropic "Evil." Yesterday He Handed Them 220,000 GPUs. — LDS, May 7, 2026
- Amazon Just Put Another $25 Billion Into Anthropic — LDS, April 21, 2026
- Google Will Pay Anthropic Up to $40 Billion — LDS, April 27, 2026
- A Jury Took Less Than Two Hours to End Elon Musk's War on OpenAI — LDS, May 20, 2026