Let me tell you a story that sounds like it was written by a screenwriter who had one too many espressos.
Just a few months ago, Elon Musk was on his own platform, X, calling Anthropic “misanthropic” and “evil.” He declared the company “doomed to become the opposite of its name” and accused it of hating Western civilization. The language was characteristically Musk — theatrical, unvarnished, designed to detonate in the timeline.
Fast forward to May 2026. That same Elon Musk just handed Anthropic the keys to one of the most powerful supercomputers ever built by human hands. Twenty-two thousand would be a headline. This is two hundred and twenty thousand Nvidia GPUs — H100s, H200s, and the latest GB200s — all packed into a single facility in Memphis, Tennessee, delivering over 300 megawatts of raw compute power. The entire capacity of Colossus 1, the supercomputer xAI built from a patch of dirt in 122 days flat, now belongs to Claude.
Silicon Valley has seen some strange bedfellows, but this one deserves a frame on the wall.
If you are an investor, a developer, or just someone trying to understand where the AI industry is actually heading, this deal is not a curiosity. It is a map. And the map tells you everything you need to know about compute, capital, and the brutal realpolitik of the frontier AI race.
The Hunger That Drove Anthropic to Musk’s Doorstep
Let us start with the obvious question. Why would Anthropic — a company backed by Amazon and Google, valued at nearly $400 billion, with multi-gigawatt compute deals already on the books — go knocking on Elon Musk’s door?
Because their users were furious, and their infrastructure was buckling.
Claude had become one of the most popular AI coding tools on the planet, and that success was eating the company alive from the inside. Developers were slamming into five-hour rate limits. Peak-hour restrictions made Claude Code unusable right when people needed it most. The company’s core services had been hovering around 99.1% uptime — which sounds fine until you realize that translates to nearly 80 hours of downtime per year, miles away from the “five nines” gold standard of 99.999%.
Anthropic had lined up enormous capacity from Amazon, Google, and Microsoft, but those deals were not going to deliver meaningful power until late 2026 or even 2027. The company did not have a year to wait. It needed atoms moved now.
Enter Colossus 1. The deal gives Anthropic more than 220,000 GPUs within a month — immediate, dedicated, and at a scale that no cloud provider could offer on short notice. Within days of the announcement, Claude Code’s five-hour rate limits doubled across paid tiers, peak-hour restrictions vanished for Pro and Max users, and API rate limits for Claude Opus shot upward. For developers who had been gritting their teeth through months of capacity rationing, this was like someone finally unclogging the pipe.
The lesson here is sobering. In today’s AI market, even a company with Amazon and Google in its corner can still find itself desperately short of compute. The gap between signing a multi-gigawatt deal and actually plugging into live power can be measured in years — and in this industry, years might as well be centuries.
The Landlord Play: Why Musk Walked Away from His Own Creation
Now for the more interesting question. Why would Musk give away the compute crown jewel he spent over $20 billion building?
Because he had already moved on to something bigger, and an idle supercomputer is just an expensive space heater.
Here is the timeline that matters. In February 2026, SpaceX acquired xAI in an all-stock deal that valued the combined entity at $1.25 trillion — the largest merger by valuation in history. By May, xAI was formally dissolved and absorbed into SpaceX as a sub-unit called SpaceXAI. Grok, xAI’s flagship model, was still alive, but its training had already migrated to Colossus 2 — the world’s first gigawatt-class AI training cluster, powered by next-generation GB200 GPUs, with a scale several times larger than Colossus 1.
So Colossus 1 was just sitting there. Twenty-two months of frantic construction, billions of dollars in hardware, 300 megawatts of power — and nobody to use it. Internal utilization at xAI had reportedly fallen to around 11%, compared to the 40% efficiency rates seen at rival labs. All twelve of xAI’s original co-founders had departed between February 2025 and March 2026. The talent exodus left behind a mountain of silicon with no one left to climb it.
Musk looked at this situation and did something characteristically unsentimental. He turned Colossus 1 into a rental property. Industry estimates put the annual lease value between $3 billion and $6 billion. That is real money, even by Musk’s standards.
And then there is the IPO. SpaceX is preparing for what analysts project could be a $1.75 trillion public listing. A company seeking that kind of valuation cannot afford to have its most expensive asset sitting idle, burning electricity and investor confidence in equal measure. Turning Colossus 1 into a revenue-generating compute platform — with Anthropic as the anchor tenant — transforms a liability into a proof point. It tells the market that SpaceXAI is not just a model builder; it is an infrastructure provider, a compute landlord, potentially a direct competitor to CoreWeave and the hyperscalers themselves.
That is the cold calculus behind the warm handshake.
The Enemy of My Enemy
There is a third layer to this deal, and it is the one that makes the whole thing feel almost Shakespearean.
Musk is currently locked in a courtroom battle with OpenAI and Sam Altman. He co-founded OpenAI in 2015 as a nonprofit counterweight to Google’s DeepMind, then left the board in 2018, then watched as the company transformed into a for-profit juggernaut valued in the hundreds of billions. His lawsuit alleges betrayal of the founding mission. The trial is playing out in an Oakland courtroom this very month.
Anthropic, meanwhile, was founded by Dario and Daniela Amodei — former OpenAI executives who left the company precisely because they disagreed with its direction. Anthropic is OpenAI’s most credible direct competitor. Claude versus ChatGPT is the defining rivalry of the consumer AI market.
So when Musk leases Colossus 1 to Anthropic, he is not just renting out GPUs. He is arming OpenAI’s most dangerous rival. He is feeding the enemy of his enemy, and he is doing it at a scale that genuinely shifts the competitive balance. Every GPU that Anthropic can use to improve Claude is a GPU that makes life harder for Sam Altman.
Musk himself provided the narrative cover with a carefully crafted post on X: “I spent a lot of time last week with senior members of the Anthropic team to understand what they do to ensure Claude is good for humanity and was impressed. Everyone I met was highly competent and cared a great deal about doing the right thing. No one set off my evil detector.”
This is vintage Musk. The “evil detector” line is disarming, funny, and impossible to fact-check. It reframes a hard-nosed business decision as a moral endorsement. It gives him an exit ramp from months of vitriol. And it leaves just enough ambiguity about whether he actually trusts Anthropic or simply found a convenient way to hurt the company he is suing.
But the contract reportedly contains what some are calling a “Humanity Clause.” SpaceXAI reserves the right to pull the plug and reclaim the capacity if Claude “engages in actions that harm humanity.” Musk, in other words, has not just rented out his supercomputer — he has kept a kill switch in his pocket, just in case his new tenants misbehave.
The Wildcard: Data Centers in Orbit
As if 220,000 GPUs and a kill switch were not enough drama for one announcement, there is a coda that pushes this deal from “fascinating” into “science fiction.”
Anthropic has expressed interest in partnering with SpaceX to develop “multiple gigawatts of orbital AI compute capacity.” In plain English: they are exploring the idea of putting data centers in space.
This is not entirely a fever dream. SpaceX is the only organization on Earth with the launch cadence, mass-to-orbit economics, and constellation operations experience to make orbital compute a near-term engineering program rather than a research concept. Space-based data centers would have access to near-limitless solar power without the land constraints, cooling challenges, and community opposition that plague terrestrial facilities. If the engineering obstacles can be overcome — and that is a massive “if” — the economics could eventually become compelling.
The NAACP has already filed lawsuits over the environmental impact of xAI’s Memphis facilities, particularly their use of gas turbines and the pollution burden on historically Black communities. Space-based data centers sidestep those problems entirely, or at least relocate them somewhere no one can protest.
Whether this orbital vision materializes in five years or fifty, it signals something important about how Musk sees the compute market. He is not just building the biggest supercomputer on Earth. He is positioning SpaceX as the infrastructure layer for an industry that will eventually outgrow the planet.
What This Deal Really Tells Us
Strip away the theatrics, the feuds, and the space ambitions, and you are left with a deal that reveals three hard truths about the AI industry in 2026.
First, compute is now more valuable than talent. xAI assembled one of the most accomplished AI research teams in the world, and then watched every single co-founder walk out the door. The GPUs stayed. And those GPUs turned out to be a more durable competitive asset than the people who were supposed to use them. That is a strange and slightly dystopian lesson, but it is the one the market just delivered.
Second, the infrastructure bottleneck is reshaping alliances faster than ideology ever could. Musk and Anthropic have diametrically opposed views on AI safety, corporate structure, and probably the weather. None of that mattered when Anthropic needed compute and SpaceX had idle capacity. In a world where GPU supply is severely constrained — cloud providers are prioritizing internal teams, rental prices have spiked over 30% in six months, and wait times stretch into 2027 — pragmatism beats principle every single time.
Third, the frontier AI market is consolidating around a handful of hyper-capitalized players who are vertically integrating in ways that make traditional cloud relationships look almost quaint. Anthropic now has compute agreements with Amazon, Google, Microsoft, and SpaceX — stitching together capacity from across the entire industry because no single provider can meet the demand alone. The companies that control the chips and the data centers will increasingly dictate the terms of the AI race, regardless of who writes the best algorithms.
Musk’s big gamble, stripped to its essence, is this: he is betting that being the landlord is more profitable than being one of the tenants. He is betting that the real money in AI is not in building the smartest chatbot, but in owning the power plants, the chips, and eventually the orbital infrastructure that makes all chatbots possible.
Whether that bet pays off depends on whether SpaceX’s IPO delivers the astronomical valuation it is chasing, whether Anthropic can convert all those GPUs into a lasting competitive advantage over OpenAI, and whether the “Humanity Clause” ever actually gets invoked.
For now, though, the image that will linger is this one: Elon Musk, the man who once called Anthropic a threat to civilization, now cashing rent checks from the very company he promised to oppose. In Silicon Valley, the saying goes, there are no permanent enemies — only permanent interests. Somewhere, a screenwriter is already taking notes.