A Startup Wants 88,000 Data Centers in Orbit… NVIDIA Just Wrote the Check
Every AI lab on Earth needs three things. Power. Land. Cooling.
Starcloud says it found all three — 400 miles up.
The startup raised $250 million last month at a $2.3 billion valuation. NVIDIA kicked in a stake alongside Cisco, and existing backers Benchmark and EQT joined too. That follows a $170 million first close in March — bringing the company's total capital raised since founding to $450 million.
That's a lot of cash for a small, young company.
But here's the part worth your time.
Starcloud is the only company running an NVIDIA H100 chip in orbit right now. Not on a slide deck. Not in a lab. In space, circling Earth every 90 minutes.
Let me explain…
In November 2025, Starcloud launched a small satellite — 60 kilograms, about the size of a mini fridge — with a single H100 GPU inside. That chip trained a language model in orbit. No one had done it before.
And NVIDIA took notes.
The two companies are now co-building the first GPU made for space. NVIDIA calls it the Space-1 Vera Rubin Module. It doesn't exist yet. But Starcloud's flight data — heat, radiation, vibration — gave NVIDIA the specs it needs to start the design.
In other words, Starcloud is NVIDIA's test lab in orbit. That's part of why NVIDIA wrote the check.
The grand vision? Launch 88,000 satellites. Each one a small data center. Solar panels soak in free power. Deep space runs brutally cold, so radiative cooling costs no electricity bill. No land. No permits. No power grid.
Sound too good?
There's a catch. You still need a rocket. And rockets are tight.
Starship can't haul cargo commercially yet. Blue Origin's New Glenn blew up on the pad in May. Rocket Lab's Neutron hasn't flown. Launch capacity across the industry is getting harder to lock in, not easier.
Starcloud itself has flagged securing enough launch allocation as one of its biggest looming costs.
Its future hangs on cheap, fast rockets that don't quite exist yet at the scale it needs.
So why is NVIDIA betting now?
Because the math works — if launch costs keep falling. An H100 draws about 700 watts on the ground. In orbit, the sun hands you that for free. Strip out the power bill, the land, the cooling towers… and orbital compute starts to pencil out.
Not today. Maybe not next year. But if launch flies often and cheap in the years ahead, this bet could look like a steal.
Big "if." Worth watching.
Elon Will Shock the World Again By September 30th
WSJ: “The last chance to amass generational wealth.”
If you missed Nvidia when I first recommended it back in 2016, before shares jumped as high as 36,000%...
I have good news.
Elon Musk is creating a second and perhaps last chance for you to profit from this AI boom.
You see, I believe by the end of this month…
With a powerful market prophecy that’s been unbroken for generations…
One that has correctly predicted some of the biggest market booms going back to 1950.
And the collision of these two economic forces…
Will give Americans a rare and perhaps last chance to turn a small stake into potentially…
An entire six-figure nest egg in the next 12-18 months.
If that sounds too good to be true…
You should know the last time these two rare economic forces collided…
Investors had a chance to turn a small stake of $10,000 into as much as $366,000 in just 14 months.
But this new retirement window won’t remain open for much longer.
The Wall Street Journal even recently warned Americans that AI advancements like this could be…
“The last chance to amass generational wealth.”
You’ll probably never see an explosive opportunity like this again in your lifetime.
The Name Behind the Round You Haven't Heard Yet
Not NVIDIA. Manhattan West actually led this $250 million round, with NVIDIA, Cisco, and several other new names joining alongside existing backers. That matters because it signals a financial investor, not just strategic tech partners, is willing to underwrite the bet at a $2.3 billion price. NVIDIA's stake gets the headlines because of the chip relationship, but the lead investor writing the biggest check is a specialist fund betting on the return, not the technology roadmap. Watch which other financial-first investors follow Manhattan West's lead.
This Company Didn't Start as "Starcloud"
A fast climb from a pivot. Starcloud was founded in January 2024 as Lumen Orbit, a Y Combinator graduate, and became one of the fastest companies in YC history to reach unicorn status after its March raise. In the eighteen months since, it's gone from a seed-stage idea to training AI models in orbit, running a version of Google's Gemini in space, and now sitting at a $2.3 billion valuation with NVIDIA on the cap table. That's an unusually compressed timeline even by space-startup standards. Watch how the company handles scaling manufacturing at that same pace.
SpaceX Just Announced It's Chasing the Same Idea
The competition isn't hypothetical. On the same August earnings call where SpaceX revealed its AI segment's explosive growth, Elon Musk said the company had decided to build "exclusively on NVIDIA" for its own AI computing expansion, and unveiled plans for "Starmind" satellites, described as an optimized version of NVIDIA's own data-center-class computing system. In other words, the world's biggest rocket company and a 25-employee startup are both racing to build the same thing, using chips from the same supplier. Watch whether Starcloud's head start on flight data outweighs SpaceX's launch advantage.
The Next Satellite Is the Real Test, Not the Funding Round
Watch for Starcloud-3, the company's next spacecraft, built at its new manufacturing facility in Woodinville, Washington.
Here's why that matters more than today's headline number. Starcloud-1 proved a single H100 could survive launch, orbit, and radiation exposure while training a model. That's a proof of concept. It doesn't prove the company can build dozens, let alone thousands, of these satellites on a repeatable production line, which is the entire premise behind an 88,000-satellite vision.
Starcloud-3 is the step that tests manufacturing scale rather than just engineering feasibility. A dedicated production facility only matters if it can actually turn out satellites faster and cheaper than building them one at a time in a lab.
The other clock running in parallel is launch. This new capital is explicitly earmarked partly for procuring future launch allocation, an admission that reserving rocket rides is now a genuine bottleneck, not an afterthought. Every fresh delay at Starship, New Glenn, or Neutron tightens that bottleneck further.
So don't judge this story by the valuation. Judge it by whether Starcloud-3 flies, and whether the company can show it's actually locked in enough launch capacity to matter at scale.
Why a Data Center's Biggest Costs Aren't the Computers
Let's keep this simple.
When people picture a data center, they picture racks of computer chips. But if you actually opened the utility bill for a large data center, the chips themselves are just one line item among several enormous ones.
Land is expensive, especially near enough to cities and power lines to be useful. Electricity is expensive, and a data center full of AI chips can draw as much power as a small town. And cooling is expensive too, because all that electricity turns into heat, and heat has to go somewhere or the chips fail.
Add those three up, and they can rival or exceed the cost of the computing hardware itself.
Now picture stripping all three of those costs away at once. Orbit has no land to buy or lease, so that line item vanishes. Sunlight hits solar panels for free, so the electricity bill shrinks dramatically for the power that would otherwise be purchased from a grid. And the environment around a satellite is so cold that heat can be radiated away without the pumps, chillers, and cooling towers a ground facility needs.
That's the actual pitch behind orbital data centers. It's not that space computers are better computers. It's that three of the biggest costs in running any data center largely disappear, at least on paper, once you move the whole thing off the ground.
The catch, of course, is that you traded three known costs for one giant new one: getting there in the first place, and the specialized hardware needed to survive once you arrive. Whether that trade pencils out depends entirely on how cheap launch gets.
Follow the cost of the ride, and you'll know whether this idea works.
Remember: a data center's biggest expenses are often land, power, and cooling, not the chips themselves. Orbit erases those three costs but adds one huge new one: the cost of the launch. Watch launch prices, because that single number decides whether this whole thesis actually works.
