Just three months ago, Elon Musk wrote that Anthropic “hates Western civilization.” He called it hypocritical, “misanthropic,” and “evil.” And now he has leased it one of the world’s largest AI supercomputers.
Anthropic and SpaceX announced a partnership that gives the creators of the Claude chatbot full access to the Colossus 1 data center in Memphis, Tennessee. The agreement covers more than 220,000 Nvidia graphics processors, including the H100, H200, and latest GB200 models, with a total power draw of over 300 megawatts. The new capacity is expected to become available within a month of the contract being signed.
What prompted two of the biggest rivals in the American AI industry to strike such a surprising deal? The answer is simple: one desperately needed computing power, while the other had a massive data center that was sitting partly unused.
Anthropic is growing eighty times faster than expected
To understand why the deal happened, we need to go back a few weeks. In early May, Anthropic CEO Dario Amodei publicly admitted that the company had expected its revenue to grow tenfold, but instead it grew eightyfold. Claude Code, an AI tool for developers, became an unexpected hit. Paying users were running into access restrictions, and developers complained about slower response times during peak hours.
Anthropic already has an agreement with Amazon for up to 5 gigawatts of infrastructure, a similarly large contract with Google and Broadcom, a partnership with Microsoft and Nvidia worth $30 billion for Azure capacity, and a $50 billion investment in US AI infrastructure through Fluidstack.
But all these agreements have one major problem. None of this capacity is available now. Amazon’s gigawatt will arrive at the end of 2026, while Google and Broadcom will be ready in 2027. Anthropic needed computing power immediately, not in eighteen months. Meanwhile, Colossus 1 was sitting in Memphis, with xAI using only about 11% of its computing power.
SpaceX absorbed xAI, leaving Colossus 1 empty
At the beginning of the year, Musk merged SpaceX with xAI in a transaction valued at $1.25 trillion. It was the largest corporate merger in the history of the American technology industry. But it soon became clear that things inside xAI were not going smoothly. xAI employees were using competitors’ models internally instead of the company’s own Grok. A wave of departures followed, with almost all of the company’s co-founders other than Musk leaving. Musk announced that xAI would cease to exist as a separate entity and that the entire operation would be renamed SpaceXAI.
Meanwhile, Grok, Musk’s AI model, failed to achieve significant success outside the X platform. It is almost entirely absent from enterprise deployments, where Anthropic and OpenAI compete for every customer. Colossus 1 therefore lost its main user when xAI moved the training of its own models to the new Colossus 2 facility. As a result, one of the world’s largest GPU centers sat largely unused in Tennessee, destroying value instead of creating it.
Musk changed his mind after meeting Anthropic’s leadership
How do you persuade Musk to lease infrastructure to a company he claimed hated Western civilization? An in-person meeting was apparently enough. “I spent a lot of time with senior members of the Anthropic team over the past week. I was deeply impressed,” Musk wrote on the X platform. “No one triggered my evil detector. If they continue to think critically about themselves, Claude will likely be a benefit to humanity.”
His decision came amid court proceedings in which Musk testified for three days as a witness in his lawsuit against OpenAI and Sam Altman. Shortly after the hearings ended, he signed the contract with Anthropic.
For SpaceX, the deal provides a clear business case ahead of its planned IPO. In investors’ eyes, the company is moving from being “just a rocket company” to becoming an AI infrastructure provider with a prestigious reference client. As analyst Ryan Mallory, CEO of data center company Flexential, noted: “The fact that serious companies are even discussing computing capacity in space says everything about how aggressively the market is searching for power and scale.”
Same here.
— Elon Musk (@elonmusk) May 6, 2026
By way of background for those who care, I spent a lot of time last week with senior members of the Anthropic team to understand what they do to ensure Claude is good for humanity and was impressed.
Everyone I met was highly competent and cared a great deal about…
A few hours after the announcement, Anthropic doubled the five-hour access limits for Claude Code on Pro, Max, Team, and enterprise accounts. It removed peak-time restrictions for Pro and Max users. It also significantly increased the limits for developer API calls to Claude Opus models. The computing power from Colossus 1 is therefore genuinely flowing into production.
Memphis, turbines, and an unresolved problem
Colossus 1 is located in Boxtown, a historically African American neighborhood in Memphis. To power the data center, xAI installed dozens of gas turbines, arguing that they were “temporary” and did not require federal permits. According to reports, the turbines worsened air quality in the surrounding area and have prompted protests by organizations such as the NAACP since 2024.
Hours after the partnership was announced, the director of the NAACP’s environmental division issued a sharply worded statement: “Any company that ignores the obvious health impacts on Black communities in order to power a future that is supposed to help all of us is making it clear who it truly intends to serve.”
Anthropic responded by promising to cover any increases in consumer electricity prices caused by its data centers in the US. The company has not yet specified exactly how it will address the turbine issue in Memphis itself.
And then there are the satellites
The contract also includes a provision that, in any other year, would sound like science fiction. The two companies expressed interest in jointly developing several gigawatts of computing capacity in orbit.
SpaceX has applied to the FCC for a license to launch up to one million data satellites. According to earlier reports, Project Terafab, Musk’s chip factory in Austin, is expected to direct 80% of its production to satellite systems in low Earth orbit. Anthropic is the first AI laboratory to publicly express an interest in actually purchasing orbital computing capacity.
If even a single gigawatt of these plans becomes reality, it will reshape AI infrastructure for decades to come. For now, however, it remains merely an “expression of interest.”
Sources: x.ai, techcrunch.com, and reuters.com
