So Anthropic just made a deal to use “all of the capacity” of xAI’s Colossus data center, and honestly, I can’t stop thinking about how messy this is going to get.
On the surface level, it makes sense. Anthropic is compute-starved. Training modern AI models requires absurd amounts of computational resources, and the shortage of available capacity is real. But the specific data center they chose to partner with? That’s where things get complicated.
The Colossus facility has a particularly rough environmental history. The gas turbines powering it initially operated without Clean Air Act permits or proper pollution control equipment. The company got away with this by classifying them as “temporary” installations, which is a creative interpretation of regulations if I’ve ever seen one. Local health data shows credible links between the facility and increased hospital admissions for air quality-related issues. This isn’t theoretical environmental impact. This is people getting sick.
Andy Masley, who’s probably done more rigorous work debunking misleading data center rhetoric than anyone else I follow, was direct about it: “I would simply not run my computing out of this specific data center.” That’s a strong statement from someone who typically avoids hyperbole.
The Political Moment We’re In
Here’s what makes this timing particularly awkward. Data centers have become genuinely contentious political territory right now. Utah just had major news stories about data center developments. States and communities are starting to push back harder on where these facilities get built and how they operate.
Anthropic signing up with the one data center that has active pollution issues and a sketchy compliance history? That’s not just a technical decision anymore. It’s a statement, whether they intended it to be or not. And in an industry where brand trust and public perception increasingly matter, it’s a statement that lands wrong.
I genuinely understand the compute constraints. These models don’t train themselves, and the waiting lists for available capacity are real. But there’s a difference between “we had to make a difficult choice” and “we’re comfortable with this choice.” The messaging around this deal suggests Anthropic is treating it like the former when the facts point toward the latter.
The Colossus Confusion
Most people initially assumed that Anthropic getting access to Colossus meant xAI was giving up on their own Grok models. That’s not what happened. Anthropic is getting Colossus 1. xAI kept Colossus 2 for their own work. This is actually important context that got buried in the first wave of coverage.
But here’s what’s interesting: the same night Anthropic announced their deal, xAI sent out deprecation notices for Grok 4.1 Fast and several other models. Two weeks notice. Two weeks. A developer who migrated to Grok 4.1 Fast back in March had this to say: “I just spent time and money to migrate to grok 4.1 fast, and you’re disabling it with less than two weeks notice, after releasing it in November, with no migration path to a fast/cheap alternative.”
This isn’t directly Anthropic’s fault, but it paints a picture of how unstable the infrastructure situation is. When companies are deprecating models that fast, it suggests they’re making rapid shifts in strategy. It suggests nobody really knows what the long-term compute strategy looks like. And when you’re betting your infrastructure decisions on someone else’s shifting priorities, that’s a risk.
The Elon Factor
Elon Musk tweeted that he was “impressed” with Anthropic’s commitment to ensuring Claude is good for humanity, and that’s why he was okay leasing Colossus 1 to them. Fine. He also said SpaceX provides compute to competing AI companies “with fair terms and pricing,” which is a fair business model.
But then he added this: “We reserve the right to reclaim the compute if their AI engages in actions that harm humanity.”
Let that sink in. Anthropic’s entire compute infrastructure is now subject to a unilateral reclamation clause decided by a single person’s judgment of what constitutes “harm to humanity.” That’s not a business relationship. That’s a supply chain hostage situation with a vague moral coating.
What happens if Anthropic’s models are used in ways Elon disagrees with? What if Claude is deployed for something he considers harmful? What’s the appeal process? Who decides? These aren’t theoretical questions. They’re the foundation of Anthropic’s entire operational resilience.
I get that Elon has been critical of Anthropic in the past (he used to call them “Misanthropic,” which is pretty funny actually). Maybe this represents a genuine warming of relations. But that doesn’t make the structural risk disappear. It just means Anthropic is now betting that Elon’s goodwill will hold steady while they train increasingly capable models.
What This Means for Everyone Else
If you’re building products on top of Claude or relying on Anthropic’s infrastructure in any way, this deal should probably be on your radar. Not because the sky is falling, but because it introduces a dependency layer that’s messier than it was last month.
And if you’re thinking about where to host your own compute infrastructure, the Colossus situation is a useful reminder that having capacity isn’t the same as having good capacity. Sometimes the most available option comes with costs you don’t want to pay.
The real tension here is that Anthropic faces genuine constraints. You can’t build frontier AI systems without compute, and compute is scarce. But scarcity doesn’t absolve you of having to think carefully about where that compute comes from and what it costs, both financially and otherwise.
So what happens next? Anthropic gets the infrastructure they need, xAI gets revenue that helps fund Colossus 2, and Elon gets to maintain influence over one of the major AI labs. Everyone technically wins. But the environmental issues at Colossus don’t go away, the community concerns about data centers don’t get addressed, and AI companies suddenly have one more supplier relationship that carries geopolitical weight.