SpaceX Is Renting 300MW of Compute to Anthropic Despite the Pentagon Dispute

In one of the most unexpected business arrangements of 2026, SpaceX is reportedly renting more than 300 megawatts of compute capacity at its massive Tennessee data centre to Anthropic — the AI company that Elon Musk has repeatedly criticised and that the Pentagon recently designated a supply-chain risk. The deal perfectly captures the strange, contradictory dynamics of the 2026 AI infrastructure race: ideology and business incentives are pulling in completely opposite directions, and business is winning.

Why Does Anthropic Need 300 Megawatts?

To put 300 megawatts in context: a typical large data centre consumes between 20 and 100 megawatts. Anthropic is renting the equivalent of multiple hyperscale data centres worth of compute from SpaceX alone — and this is on top of its existing infrastructure agreements with Amazon Web Services, Google Cloud, and other providers. The compute demand is driven by training and running Anthropic's most advanced models, including the recently released Mythos, which requires extraordinary computational resources both to train and to run at scale.

The AI industry's hunger for compute has made the bottleneck shift from talent and algorithms to electricity and physical infrastructure. Anthropic, OpenAI, Google DeepMind, and xAI are all in an arms race not just to build the smartest AI — but to secure the physical infrastructure to run it. Whoever controls the compute controls the future of the industry.

The Musk-Anthropic Contradiction

The deal is extraordinary given Elon Musk's public position on Anthropic and AI safety. Musk has been among the most vocal critics of AI safety-focused labs, calling their approach to AI governance paternalistic and anti-competitive. He is also currently suing OpenAI in a separate legal dispute. And yet SpaceX — one of Musk's companies — is providing critical infrastructure to Anthropic, an AI safety lab, allowing it to run the very models that have triggered the most significant AI governance crisis of the year.

The explanation is straightforward: data centre leasing is an enormous revenue opportunity, and SpaceX's Tennessee facility has capacity that generates revenue when rented out. Business logic has overridden ideological positioning, and the result is one of the most ironic infrastructure arrangements in tech history.

The Broader AI Infrastructure Battle

The SpaceX-Anthropic deal is a single data point in a much larger infrastructure war. KKR has committed more than $10 billion to launch Helix Digital Infrastructure, a company focused exclusively on building AI data centres and power generation. SoftBank is manufacturing large-scale battery cells at its Osaka plant specifically to meet AI power demand. Private equity firms are pouring billions into purpose-built AI infrastructure globally.

The constraint is no longer GPUs or even money — it is electricity. AI training runs at scales that require the output of small power plants. The companies that can bring reliable, large-scale power online fastest will determine who wins the next phase of the AI race. SpaceX's Tennessee facility, with its significant power capacity, is simply a resource that the AI industry needed — regardless of who owns it or what its owner thinks about AI safety.

What It Means for the AI Race

The deal signals that Anthropic — despite the Pentagon dispute, the EU access controversy over Mythos, and the White House tensions — is not slowing down. It is aggressively scaling compute at a rate that requires renting infrastructure from a company whose CEO is publicly hostile to its mission. In the AI industry of 2026, the imperative to scale compute overrides almost everything else. The SpaceX-Anthropic arrangement is proof of that.