Space, the Final Server Room

GriftMatrix Tech Oligarchy

Nvidia is putting AI data centers in orbit and nobody blinked

NVIDIA announced yesterday that it’s building data centers in space. To be fair – this has been theorized for a while now – but they finally said it out loud. The company that already powers basically all of artificial intelligence on Earth has decided that Earth is not enough, and they’re taking the whole operation orbital.

Jensen Huang, Nvidia’s CEO, unveiled something called the Vera Rubin Space-1 Module during his keynote at the company’s annual GTC conference in San Jose. It’s a chip system specifically designed to run AI from orbit. Six companies are already signed up to use it, including a startup called Starcloud that’s literally building data centers meant to float around the planet. Last November, Starcloud launched an Nvidia H100 chip on a satellite and ran an AI model in actual outer space. Nvidia says the new Vera Rubin space chip will deliver 25 times more computing power than that.

The reason this is happening is pretty simple: AI data centers use an obscene amount of electricity. Like, “straining the power grid and driving up electricity costs for entire communities” levels of obscene. Terrestrial data centers need massive cooling systems to keep the chips from melting. In space, you’ve got unlimited solar energy and no need for air conditioning.

The trade-off is that there’s also no air, no convection cooling, and you have to figure out how to deal with radiation. Huang acknowledged this on stage, saying “in space, there’s no convection, there’s just radiation,” then casually added that they’ve got “lots of great engineers working on it.” The way you’d mention you’re having someone fix the garage door.

NVIDIA isn’t the only one thinking about this. Google has something called Project Suncatcher, exploring the same concept. And last month, Elon Musk merged SpaceX with his AI company xAI in a $1.25 trillion deal – the largest merger in history – and said the whole point was to build orbital data centers. SpaceX has already filed with the FCC for authorization to launch up to ONE MILLION satellites for AI computing. Musk’s exact words were “space is called space for a reason,” which he punctuated with a cry-laughing emoji, because of course he did.

Jeff Bezos predicted last October that gigawatt-scale data centers in orbit were 10 to 20 years away. Musk thinks it’s two to three. Nvidia is just showing up with the chips and saying “call us when you’re ready.”

All of this was buried under Huang’s other headline, which was that Nvidia expects to make a trillion dollars selling AI chips between now and the end of 2027.

That’s double the $500 billion he projected last year. To put that in context, no company on Earth currently generates a trillion dollars in annual revenue. Walmart, the biggest company by sales, does about $681 billion. Nvidia is projecting more from two product lines over two years than Walmart makes selling literally everything to literally everyone.

The stock barely moved. Up about 1.7% for the day. When your company is already worth $4.45 trillion and you’ve been riding the AI wave for three straight years, a trillion-dollar revenue forecast is apparently just a normal Monday.

There was a pile of other news too. Nvidia and Uber are rolling out self-driving cars in L.A. and San Francisco starting in the first half of 2027, expanding to 28 cities across four continents by 2028. The target is 100,000 autonomous vehicles on Uber’s network. BYD, Nissan, Hyundai, and Geely are all building Level 4 self-driving cars on Nvidia’s platform. Level 4 means the car handles the driving and you can take a nap.

NVIDIA also launched a new CPU called Vera that puts it directly in competition with Intel and AMD in the processor market. Intel currently owns about 60% of server CPUs. AMD has 24%. Nvidia has 6%. But given Nvidia’s track record of devouring markets, those numbers are probably making some people very nervous.

Then there’s the Groq situation. Last December, Nvidia paid $20 billion for a chip startup called Groq – not Grok, Elon’s chatbot, different company. They hired 90% of its employees, took all its technology, and called the whole thing a “licensing agreement” instead of an acquisition so regulators wouldn’t look too hard.

One analyst said the deal was structured to keep “the fiction of competition alive.” Groq makes chips that specialize in making AI respond faster. Nvidia showed off the first product using that technology yesterday, the Groq 3 LPU, shipping later this year.

Competition is building though. Cerebras, a chip company you’ve probably never heard of, just signed deals with both OpenAI and Amazon Web Services. Google keeps building its own chips. AMD isn’t going anywhere. Every major cloud provider – Amazon, Google, Microsoft – is designing custom processors in-house. Nvidia is still the king, but the kingdom is getting crowded.

The question nobody wants to answer out loud is still the same one: all these companies are spending hundreds of billions on AI chips, but where is the money coming back from? Nvidia is selling the shovels during a gold rush. The shovel business is incredible. But the prospectors still need to find gold.

Huang ended his keynote by predicting that software companies will pivot to selling AI agents instead of software. Then he showed off a robot Olaf from Frozen that Disney built with Nvidia’s help.

We’ve gone from “we’re making chips for chatbots” to “we’re putting data centers in orbit and building sentient snowmen” in about three years.

Sources