Elon Musk Rented His Supercomputer to His Competitor. Why?
Yesterday, Anthropic announced it's taking over all the compute at xAI's Colossus 1 data center in Memphis. Over 220,000 NVIDIA GPUs and 300 megawatts of power, now running Claude instead of Grok.
Three months ago, Musk posted on X that Anthropic "hates Western Civilization." Yesterday he posted that the Anthropic team impressed him and "no one set off my evil detector."
I don't know Elon Musk. I can't tell you what's really going on in his head. But I can read the numbers and the timing, and I think everything happening here makes sense if you look at it from a few directions at once.
What happened at xAI
The short version: xAI lost its entire founding research team and couldn't use its own hardware.
SpaceX acquired xAI in an all-stock deal in February 2026. Within days, co-founders started leaving. Tony Wu left on February 10. Jimmy Ba followed within 24 hours. By late March, all 11 original co-founders had departed. Musk posted that xAI "was not built right first time around, so is being rebuilt from the foundations up."
Then, separately, The Information reported that an internal xAI memo showed GPU utilization at around 11%. Business Insider confirmed the memo. xAI's president, Michael Nicolls, used the phrase "embarrassingly low." For context, Meta and Google run their GPU clusters at 43% and 46% respectively. The industry floor is around 35%. xAI was operating at roughly a third of the worst performers.
So the picture is: 550,000 GPUs in Memphis, all 11 founding researchers gone, and the cluster running at a fraction of what it should. That's billions of dollars in hardware generating almost nothing.
The IPO math
SpaceX filed its IPO paperwork with the SEC on April 1, targeting a $1.75 trillion valuation and a $75 billion raise. Nasdaq listing is planned for June.
Reuters reviewed parts of the filing. What they found: xAI lost $6.4 billion in 2025 and ate up 61% of SpaceX's total spending that year. Meanwhile, Starlink made $4.42 billion in profit. So Starlink earns the money, and xAI burns it. The whole company swung from $791 million in profit in 2024 to a $4.9 billion loss in 2025, mostly because of xAI.
Now imagine you're taking this company public next month. Your biggest capital investment is a GPU cluster running at 11%. Your AI division is bleeding $6.4 billion a year. Your founding research team is gone.
You need a customer. Fast.
Enter Anthropic
On the other side of this deal, Anthropic is drowning in demand. CEO Dario Amodei said at a developer conference yesterday that the company saw 80x growth in revenue and usage in Q1 2026 when they'd planned for 10x. They don't have enough compute. Claude users have been hitting rate limits for weeks.
So Musk has GPUs he can't use. Anthropic has demand it can't serve. Not a hard negotiation.
Axios reported that by renting Colossus 1 to Anthropic, Musk avoids having to explain billions of dollars in unused hardware to investors right before the IPO. Instead of going public with an idle cluster on the books, SpaceX goes public with Anthropic as a paying customer. The deal is reportedly worth billions.
The trial timing
And then there's the court case.
This deal was announced while Musk is in the middle of his trial against OpenAI. Last week he spent three days on the stand in Oakland. Under oath, he admitted there was no written agreement governing his $38 million donation to OpenAI. The judge struck one of his repeated phrases from the record.
Two days before the trial started, according to a CNN report on an OpenAI court filing, Musk texted Greg Brockman to test the waters on a settlement. Brockman suggested both sides drop their claims. Musk replied: "By the end of this week, you and Sam will be the most hated men in America."
Anthropic was founded by people who left OpenAI. By partnering with them now, during this specific week, Musk is sending a message whether he means to or not. The people who walked away from Sam are the reasonable ones. The people who stayed are the problem.
I have no idea if that factored into the deal. But it looks like what it looks like.
What I actually think
I think all of these things are probably true at the same time.
The financial pressure is real. You don't go public with your biggest asset sitting idle. The Anthropic deal turns a money pit into a paying customer. That alone is reason enough to do it.
The step back from competing on AI models is real too. When Google had excess compute, they kept it and built their own models. Sundar Pichai admitted last month that Google Cloud revenue was lower than it could have been because they were "capacity constrained" and chose to use the GPUs internally. Musk made the opposite choice. He rented the GPUs to a competitor. That's not what you do if you think your own models are about to break through.
And the trial timing is convenient at minimum. Going from "Anthropic hates Western Civilization" to "I was impressed, no one set off my evil detector" in three months, right as you're in court arguing that OpenAI betrayed its mission? That kind of flip serves a purpose even if it wasn't planned.
TechCrunch asked the right question: is xAI just a GPU rental company now? CoreWeave and Lambda built their whole businesses renting out chips to AI companies. xAI looks like it's heading the same direction, except it was supposed to be building the next frontier AI model. The $230 billion valuation from January was for a company building AGI. The company renting its data center to Anthropic in May looks more like a landlord.
Musk is many things but he's not stupid. This deal solves several problems at once. The IPO gets a cleaner balance sheet. The idle GPUs start making money. The trial gets a convenient headline. And Anthropic gets the compute it desperately needs. I don't think you have to pick one motive. People do things for multiple reasons all the time, especially when billions of dollars and a public trial happen to land in the same week.