Anthropic just bought up all the compute capacity at xAI’s Colossus 1 data center in Memphis, Tennessee, and the AI press isn’t buying the official spin. According to TechCrunch AI’s latest Equity podcast, hosts Kirsten Korosec, Sean O’Kane, and Anthony Ha picked the deal apart and landed on a far less flattering read: xAI looks like it’s quietly pivoting away from frontier model training right before SpaceX’s mega-IPO.
Here’s what actually happened. Anthropic, hungry for compute to power its enterprise-focused products like Claude, locked down the entire Colossus 1 facility. xAI, the AI arm Elon Musk folded into SpaceX after a reported $250 billion acquisition, becomes the landlord. TechCrunch AI reports that Musk is now planning to dissolve xAI as a standalone entity and rebrand the combined operation “SpaceXAI.”
Why this matters
This isn’t a routine cloud contract. It’s a tell.
Frontier AI labs hoard compute. They don’t lease it out, because every spare GPU is supposed to be training the next model. When a company that’s positioned itself as a frontier lab suddenly hands its biggest data center to a competitor, the market reads that as a signal: the training ambitions aren’t real, or they aren’t working.
Korosec laid out the term on the podcast: xAI is becoming a “neocloud,” the industry shorthand for outfits that buy Nvidia GPUs and rent them out instead of training their own models. As TechCrunch AI’s Russell Brandom pointed out, most data center builders still prioritize internal training over rentals. Choosing the rental path says something.
The cynical read
O’Kane didn’t sugarcoat it. “Why be positive when you can be cynical?” he asked, calling the move “a major heat check before the IPO.”
The context backs him up:
- Grok isn’t winning the consumer chatbot race outside of X, and it’s picked up a reputation for unpleasant and arguably illegal outputs.
- Reporting cited on the podcast says xAI employees were using competitor models internally instead of Grok, triggering a shakeup that pushed out every co-founder except Musk.
- Musk has reportedly told the team he’s “starting from scratch” on xAI, despite SpaceX’s $250 billion price tag.
- Colossus 1 is already facing an environmental lawsuit.
Korosec tried the optimistic angle: hey, at least they found a way to make money. True, but renting GPUs is a fundamentally different pitch than “we’re building the next frontier model.” One gets boring infrastructure multiples. The other gets the kind of valuations OpenAI and Anthropic are commanding.
What it means for Anthropic
For Anthropic, the deal is a clean win. The company has been openly hunting for more compute, and Colossus 1 is an escape valve that lets it scale Claude’s enterprise footprint without waiting on new builds. Anthropic gets capacity now, when capacity is the binding constraint for every serious lab.
The asymmetry is the story. One side of the deal is buying fuel for its frontier ambitions. The other side is admitting it doesn’t need that fuel.
What to watch
A few things worth tracking as SpaceX moves toward going public:
- IPO positioning. Will SpaceX market the AI piece as a frontier lab or as infrastructure? The pitch deck will tell us which story Musk is committing to.
- Grok’s roadmap. If xAI is really starting over, expect a quiet period on model releases. Or a flashy launch designed to muddy the neocloud narrative.
- More compute deals. If Anthropic is taking all of Colossus 1, who’s next? Microsoft, Google, and Meta are all looking. Other neoclouds will follow this template.
The tension O’Kane flagged is real. A reliable GPU rental business is more believable than a moonshot lab in the near term, but it’s not what draws frontier-lab money. SpaceX is about to ask the public markets to pick a story, and this deal makes the choice harder, not easier.
More details and the full podcast conversation are available at the original source.