The news hit the tech world like a bombshell last week. Elon Musk, the man who built rockets that land themselves and electric cars that outsell Toyotas, quietly announced on X that his AI baby xAI was being dissolved and folded into SpaceX. Just like that. No grand farewell, no all-hands meeting, no tearful goodbye from the billionaire who once swore he’d save humanity from the evils of woke AI. Just a three-sentence post, and a $250 billion unicorn ceased to exist.
Three years ago, Musk stood on a stage and declared war on OpenAI. Today, his AI army has surrendered. And the most brutal part? He just rented his prized supercomputer to his sworn enemy.
What the hell happened?
The Empire That Never Was
Let’s rewind to 2023. OpenAI had just dropped GPT-4 and the world was losing its mind. Musk, who co-founded OpenAI back in 2015 and watched Sam Altman turn it into a for-profit juggernaut, was furious. So furious that he signed an open letter calling for a pause on advanced AI development while quietly registering a new company behind everyone’s backs. That company was xAI.
He poached 11 of the brightest minds from DeepMind, OpenAI, Google, and Microsoft. He called them his “dream team.” Their mission? Nothing less than “understanding the nature of the universe.” In Musk’s private conversations, it was much simpler: build a model so good it would make Altman look like a fraud.
And for a while, it worked. xAI raised over $42 billion across multiple funding rounds. Its valuation hit $250 billion in early 2026, making it the highest-valued AI unicorn on the planet. It built Colossus 1, a supercomputer with 220,000 Nvidia GPUs, in just 122 days in an abandoned appliance factory in Memphis. The scale was staggering. The ambition was pure Musk.
But underneath the glossy numbers, something was rotting.
The $250 Billion Flop
Here’s the uncomfortable truth that nobody in Musk’s inner circle wanted to admit: Grok wasn’t that good.
Yes, Grok 3 briefly topped the leaderboards in early 2025. Yes, the chatbot had a cult following on X, where its “rebellious” personality and willingness to say things other AIs wouldn’t made it a favorite for edgy memes. But memes don’t pay the bills. And Grok was bleeding cash at an apocalyptic rate.
By late 2025, xAI was burning about $1 billion a month. That’s $33 million a day. $1.4 million an hour. You just lost another $23,000 reading this sentence.
Where did it all go? Mostly into Colossus. The supercomputer was a marvel of engineering, but it was also a financial black hole. xAI was spending fortunes on H100s, H200s, GB200s, and enough natural gas turbines to power a small city, all just to keep the lights on in Memphis. And for what?
Grok’s market share peaked at around 14% in the U.S. last December. Sounds decent, until you remember that ChatGPT had 53% and Gemini had 29%. Even worse, the gap wasn’t closing, it was widening. By April 2026, Grok’s global monthly active users had plummeted to 12.2 million, dropping from second place to fifth. Meanwhile, Claude grew to 23 million users, and Anthropic overtook OpenAI in enterprise API revenue.
The killer blow came from the developer community. Grok’s coding abilities were a joke. Industry insiders had a crude saying: no one uses Grok for programming, not even xAI’s own engineers. They were quietly using Claude to do their jobs. Musk himself eventually admitted that xAI’s coding models were far inferior to the competition.
A large language model that can’t code, in 2026, is like a car that can’t drive. It’s a toy. And nobody pays serious money for toys.
The Exodus
If the technology was a failure, the management was a catastrophe.
Musk runs his companies the same way he runs his life: at 100 miles per hour, fueled by chaos, caffeine, and sheer force of will. That worked for Tesla and SpaceX. It didn’t work for xAI. Scientists don’t respond well to being told to work 80-hour weeks. They don’t appreciate having their research direction flipped every quarter because the CEO had a new idea at 3 AM.
The result was a mass exodus that Silicon Valley has rarely seen. Between February 2025 and March 2026, all 11 of xAI’s original co-founders quit. Every single one. The last to go was Ross Nordeen, Musk’s most trusted lieutenant. By the end, it was just Elon, alone in a room full of empty desks.
More than 80 other employees fled during the same period. Some left and immediately posted on X that their “number one priority is finally getting eight hours of sleep.” You can’t make this stuff up.
Musk, in a rare moment of candor, finally admitted in March 2026 that “xAI was not built right the first time around.” He said the company would be rebuilt from the ground up. But by then, it was already too late. The IPO clock was ticking, and SpaceX needed a story to sell.
The Real Reason xAI Died
Here’s what the headlines aren’t telling you. xAI didn’t just fail because Grok wasn’t good enough. It failed because Musk never really wanted to win the AI model war in the first place.
Think about it. In 2025, before xAI collapsed, Musk had already orchestrated a series of moves that look, in retrospect, like a master class in corporate jiu-jitsu. First, xAI bought X in an all-stock deal valued at $113 billion, giving Grok access to the real-time firehose of 600 million active users. Then Tesla poured $2 billion into xAI, ostensibly to power its self-driving and Optimus robot training. Then SpaceX bought xAI outright for a combined valuation of $1.25 trillion.
Each step moved xAI’s assets closer to the crown jewel: SpaceX.
And on May 6, 2026, the final piece clicked into place. xAI was no more. Its 220,000 GPUs? Rented to Anthropic in a deal reportedly worth $3 to $6 billion per year. Its Colossus 2 supercomputer? Now the backbone of SpaceXAI, a division dedicated to putting data centers in orbit. Its 11 founders? Gone. Its mission? Absorbed.
Musk didn’t lose the AI race. He pivoted. He realized that the real money isn’t in making the best AI model, it’s in being the landlord of the infrastructure that every AI model depends on. And who builds better infrastructure than SpaceX?
The Irony of Renting to the Enemy
The most shocking part of the announcement wasn’t the dissolution of xAI. It was the partnership with Anthropic.
Just three months earlier, Musk had called Anthropic “misanthropic and evil.” He accused them of “hating white and Asian people.” He said winning was “never in their possibility set.” Now, he was handing them the keys to Colossus 1, the supercomputer that xAI had spent billions building.
Musk’s explanation was classic Elon. He said he spent a week talking to Anthropic’s core team and “no one set off my evil detector.” He said they were “very capable” and “really care about doing the right thing.” He added that SpaceXAI reserved the right to take the compute back if Claude ever became a threat to humanity.
It sounded like a man making excuses for sleeping with the enemy. But the business logic is undeniable. Colossus 1 was sitting idle while xAI’s training workloads moved to Colossus 2. Renting it to Anthropic turns a depreciating asset into a cash cow right before SpaceX’s IPO. And it positions SpaceX not as an AI wannabe, but as the AWS of space-based computing.
The math is simple. Train AI models on Earth? You compete with Amazon, Google, and Microsoft. Own the rockets that put the servers in orbit? You have no competitors. SpaceX is the only company on the planet with the launch cadence, orbital insertion economics, and constellation experience to make space-based data centers a reality.
The Bigger Picture
So was xAI a failure? Yes and no.
As a standalone AI company, it was a disaster. A quarter-trillion dollars in valuation, vaporized in three months. The founders all gone. The product a punchline. Musk’s dream of beating OpenAI at its own game ended not with a bang, but with a three-sentence tweet.
But as a strategic asset in Musk’s larger empire, xAI served its purpose. It built the infrastructure. It trained the models. It burned the cash. And now, it’s been absorbed into the vehicle that will carry Musk’s entire empire into its next phase: the IPO of SpaceX, the most valuable private company in the world.
SpaceX is reportedly targeting a June IPO with a valuation of $1 trillion or more. The offering could raise $50 billion, breaking every record in history. And the story Musk will tell investors is not “we build rockets,” but “we are the infrastructure backbone of the AI age, and we’re the only company that can put data centers in space.”
Whether that story holds up is another question. Analysts are skeptical. Building orbital data centers is orders of magnitude harder than Musk’s timelines suggest. His predictions for the future are almost always wrong. But one thing is certain: when Musk loses, he loses in a way that sets up his next win.
xAI is dead. Long live SpaceXAI.
The Lesson
Here’s what the rest of the tech world can learn from this.
Building a great AI model is hard. Building a great AI company is harder. But building an AI company that can compete with OpenAI, Google, and Anthropic while reporting to a CEO who thinks scientists should work like factory workers is nearly impossible. Musk’s management style works when the goal is to land a rocket on a drone ship. It doesn’t work when the goal is to nurture a research culture that produces breakthroughs.
The other lesson is about knowing when to fold. Musk could have kept xAI alive, burning a billion dollars a month, bleeding talent, watching Grok’s market share dwindle to nothing. Instead, he pulled the plug, repurposed the assets, and moved on. It’s brutal. It’s cold. And it’s exactly the kind of decision that has made him one of the most successful entrepreneurs in history.
xAI’s story is not a tragedy. It’s a pivot. The AI model game was always going to be won by someone else. But the AI infrastructure game? That’s just getting started. And Musk just bet the entire farm on being the one who wins it.
Will it work? Nobody knows. But if history is any guide, betting against Elon Musk has been a reliably expensive mistake.
Just ask Sam Altman.