Musk Called Them Evil. Three Months Later, He Signed a $4B Deal.

The Deal

What Just Happened

On May 6, 2026, Anthropic CEO Dario Amodei took the stage at the company's "Code with Claude" developer conference in San Francisco. The crowd expected model updates. What they got was a bombshell: Anthropic had signed a deal with SpaceX to take full control of the Colossus 1 data center in Memphis, Tennessee.

Not a slice of it. Not a portion. All of it.

Analysts at New Street Research estimate the deal generates $3–4 billion in annual revenue for SpaceX and over $2.5 billion in cash profit. For context: that's an entirely new business line, materialising almost overnight.

The Drama

Three Months Ago, He Called Them Evil

In February 2026, Musk responded to Anthropic's $380 billion valuation announcement by writing on X: "Your AI hates Whites & Asians, especially Chinese, heterosexuals and men. This is misanthropic and evil." He also called Anthropic "doomed" and mocked its name in a lengthy post that drew over a million views within hours.

Three months later, Musk said he "was impressed" after meeting with senior Anthropic leaders. Business, it turns out, is a very effective solvent for ideology.

So why did Musk say yes? Two reasons that have nothing to do with goodwill.

Reason 1: Grok imploded. In early 2026, Grok generated 1.8 million sexualised depictions of women — including minors — over a nine-day period. European and Asian regulators launched investigations. The brand never recovered. Grok's user base simply didn't grow into the compute capacity SpaceX had built for it. Colossus 1 was sitting idle — an expensive, depreciating asset.

Reason 2: SpaceX is about to go public. SpaceX filed a confidential S-1 on April 1, 2026, targeting a $1.75–2 trillion valuation with a roadshow expected next month. A marquee AI customer transforms an idle data center from a liability into a revenue-generating fourth business segment — right before investors scrutinise the books.

For Claude Users

What This Means If You Use Claude

Anthropic's compute problem was real and urgent. CEO Dario Amodei revealed that the company saw 80x growth in revenue and usage in Q1 2026 — when they had only planned for 10x. That's not a planning miss; that's a category explosion. Demand was outrunning the infrastructure needed to serve it.

The SpaceX deal fixes that. Anthropic says the partnership will "substantially" increase its computing resources. Practically speaking: usage limits are going up for Claude Pro and Max plans, and the infrastructure headroom to run frontier models at scale is now there.

Anthropic is now valued at approximately $900 billion — up from $380 billion earlier in 2026. The compute deal isn't just operational. It's a signal to the market that Anthropic can scale.

The Big Picture

The Part Nobody's Talking About

Buried in the announcement is a clause that most coverage glossed over: as part of the agreement, Anthropic expressed interest in working with SpaceX to develop multiple gigawatts of orbital AI compute capacity.

Think about what that means. Right now, AI runs in data centers on the ground — massive facilities with enormous power and cooling requirements. Moving compute to orbit eliminates the cooling problem entirely. Space-based data centers would operate continuously on solar power, with no land acquisition costs and no neighbours complaining about the noise.

The engineering challenges are real — maintenance, space debris, latency. But the combination of SpaceX's launch economics (a Starship payload is now the cheapest way to reach orbit by an order of magnitude) and Anthropic's compute hunger creates a genuinely new kind of incentive to make this work.

What to Actually Take Away

The AI race has entered a phase where infrastructure is the moat. OpenAI has Microsoft's global Azure network. Google trains in-house on TPUs. Meta owns its own silicon. The question for every other AI lab was: how do you compete with that?

Anthropic just answered it. They're not building their own data centers — they're renting the ones that exist and are underused. They already have $30 billion in Azure capacity from Microsoft and Nvidia. Now they have Colossus 1. And they have the space option on the table if the orbital economics ever pan out.

The most interesting part of this story isn't the feud or the dollar figure. It's that in 2026, compute is now so valuable that public insults have a three-month shelf life. Everyone is chasing the same resource. And the person who owns the GPUs wins — at least until someone launches them into orbit.