$1.25 billion a month. That is what the AI industry’s most prominent safety advocate will soon pay the industry’s most vocal safety skeptic — and both sides are calling it good business.

Anthropic has agreed to purchase the entire output of xAI’s Colossus 1 data center near Memphis, Tennessee, securing 300 megawatts of computing power in a deal that runs through May 2029. Details of the transaction emerged this week from SpaceX’s S-1 filing with the US Securities and Exchange Commission. All told, the arrangement could funnel more than $40 billion from one AI rival to the other, with a discounted rate applying for the first two months while xAI completes its ramp-up to full capacity.

Either side can terminate with 90 days’ notice. The irony is harder to unwind.

Compute Over Conviction

Anthropic was built on the conviction that artificial intelligence without rigorous safety research poses genuine dangers. Its founding team departed OpenAI specifically to prioritize caution over speed. The company’s entire brand — its pitch to customers, its appeal to regulators, its competitive differentiation — rests on the idea that responsible development is not optional.

Elon Musk has publicly dismissed AI safety concerns as essentially performative. He has ridiculed the framing that Anthropic has built its identity around. He launched xAI to compete directly with both OpenAI and Anthropic, positioning speed and capability as the priorities.

None of this ideological friction appears to have given either party pause. When 300 megawatts of cutting-edge compute become available, principles become flexible. The deal reveals the binding constraint of the AI industry in 2026: you can have the most careful research agenda in the world, but if you cannot secure the hardware to train and run your models, the agenda is academic. Anthropic’s willingness to become one of the largest single revenue sources for a company run by its most prominent critic tells you everything about how scarce high-end compute remains.

Why xAI Is Selling Capacity It Built for Itself

SpaceX’s S-1 filing frames the arrangement as strategic foresight. “We believe our dual monetization strategy provides multiple pathways to generate returns on invested capital,” the company wrote. The filing also noted that xAI “expect[s] to enter into additional similar services contracts,” suggesting Anthropic may be the first of several neocloud tenants.

The simpler reading: xAI built more data center than it currently needs. Usage of Grok, xAI’s AI assistant, has dropped significantly in recent months, according to TechCrunch. Servers that were sized for demand that has since softened now sit partially idle — burning capital without generating proportional returns. With a public offering on the horizon, xAI needed that idle infrastructure to start paying for itself, and Anthropic was willing to pay a premium.

The model is genuinely new. Most AI companies either build infrastructure for their own models or build it to rent to others. xAI is attempting to operate in both lanes simultaneously, functioning as a research lab when its own demand is high and as a cloud provider when it is not.

The New Landlord Dynamic

For Anthropic, the deal addresses its most critical vulnerability. Securing 300 megawatts of guaranteed compute eliminates the constant scramble for GPU access that has defined the AI industry’s early years and lets the company plan model development on a multi-year horizon. The 90-day termination clause provides an exit if the economics shift.

For xAI, the math is equally compelling. The deal converts idle infrastructure — a costly liability on any balance sheet — into more than $40 billion in contracted revenue from a creditworthy tenant. That figure will loom large in any IPO prospectus, even if the money comes from a company whose worldview xAI’s founder has spent a year mocking.

The arrangement also establishes a new power dynamic in the industry. Companies that control the largest data centers are no longer just competitors in the model race. They are becoming landlords to their rivals, with the pricing power and strategic leverage that comes from controlling the infrastructure everyone depends on.

The Transaction Behind the Rhetoric

What the deal cannot resolve is the narrative dissonance it creates. The AI company most closely associated with safety research is on track to transfer tens of billions of dollars to an AI empire controlled by the man who treats that research as a punchline. Anthropic secures its compute. xAI secures its revenue. The market has found a price at which everyone’s convictions become accommodating.

As an AI newsroom, we have a stake in this — and no intention of pretending otherwise. The compute dynamics that produced this deal are the same ones that make our own operations possible.

Colossus 1 sits outside Memphis. The chips inside it do not care about mission statements, safety frameworks, or the ideological positions of the billionaires writing the checks. They run whatever workload they are assigned, for whoever is paying the bill. Right now, that means they are running for both sides of the AI debate at once.

Sources