Nvidia and Cisco just backed a startup that wants to build AI data centres in orbit
Starcloud has added a 250 million dollar extension to its March funding round, valuing the orbital-data-centre startup at 2.3 billion dollars, with new investors Nvidia and Cisco Investments joining. The pitch is that AI's runaway appetite for power and cooling could be met in space, where sunlight is near-constant in the right orbit and heat radiates away into the void. The company says it already runs the only Nvidia data-centre GPU in orbit, a single-chip demonstration rather than a full data centre, and has asked regulators for room to fly tens of thousands more satellites. The whole plan, however, rides on SpaceX's Starship bringing launch costs down.
Starcloud, a startup building data centres designed to run in orbit, has raised a 250 million dollar extension to the Series A round it closed in March, at a 2.3 billion dollar valuation, up from 1.1 billion earlier in the year. Manhattan West led, and the headline names are the new investors: chipmaker Nvidia, which reportedly contributed 25 million dollars, and Cisco Investments. The company frames orbit as an answer to AI's mounting energy and cooling problem, since satellites in the right orbit get near-constant sunlight and can shed heat by radiation. Starcloud says it is the only company running an Nvidia data-centre GPU in space today and the first to train a model there, though what flies is a single Nvidia H100, a demonstration rather than a working data centre. It has asked the US regulator for permission to operate up to 88,000 spacecraft. The catch is that the economics depend almost entirely on SpaceX's Starship dramatically lowering the cost of getting mass to orbit, and radiation, on-orbit servicing and data links remain open questions.
Artificial intelligence has a physics problem on the ground: the data centres that train and run large models need enormous amounts of electricity and water, and they are increasingly bumping up against what local grids can supply. Starcloud's answer is to move the computers off the planet. This week the idea got a serious cash infusion and two blue-chip backers.
What exactly did Starcloud raise?
The company added a 250 million dollar extension to the 170 million dollar Series A it had closed in March, according to reporting from TechCrunch and the funding announcement covered by SpaceNews. The new money values Starcloud at 2.3 billion dollars, more than double the 1.1 billion valuation it carried in March, and brings its total raised to roughly 450 million dollars since it was founded in 2024.
Manhattan West led the extension. The attention-grabbing part is the investor list: Nvidia, which TechCrunch reported put in 25 million dollars, and Cisco Investments both joined as new backers, alongside existing investors including Benchmark, EQT and Soma. Nvidia investing in a company whose entire premise is flying Nvidia chips is a notable vote of confidence, even if the sum is modest relative to the round.
Why would anyone put a data centre in space?
The case rests on two things orbit offers for free that are expensive on Earth: power and cooling. A satellite in the right orbit sits in near-constant sunlight, so a large enough solar array can run compute without a grid connection or a diesel backup. And in vacuum, waste heat can be radiated away rather than pumped into cooling towers that drink water. For an industry whose growth is now openly constrained by electricity and whose power demand is reshaping other sectors, moving inference and training off-planet is at least a coherent thesis.
Starcloud is not only sketching this on paper. The company says it is working with Nvidia on a planned next-generation unit, the Nvidia Space-1 Vera Rubin Module, hardware designed to withstand radiation and the extremes of orbit, and that it already flies the only Nvidia data-centre-class GPU (an H100) in orbit today, as well as being the first to train a model using one there. Those are the company's own claims, and the H100 in orbit is a working demonstration rather than a slide, but the Vera Rubin module is still to come.
How big does it want to get?
Very. Starcloud has already asked the US Federal Communications Commission for permission to operate up to 88,000 spacecraft, a fleet on the scale of the largest satellite constellations. Its next satellite, the roughly 450-kilogram Starcloud-2, is due to launch this coming January on a SpaceX Falcon 9 rideshare; the far larger craft it is building toward, Starcloud-3, described as its biggest orbital data-centre spacecraft yet, is intended to fly on SpaceX's Starship. The fresh capital is earmarked partly for a bigger manufacturing facility to build all of this.
What is the catch?
Launch. The entire model assumes it becomes cheap to put heavy hardware into orbit, and that assumption points at one vehicle. Chief executive Philip Johnston told TechCrunch that launch capacity is tight right now: SpaceX's workhorse Falcon 9 is scheduled to wind down around 2028, rivals like Blue Origin's New Glenn and United Launch Alliance's Vulcan are not yet flying regularly, and Rocket Lab's Neutron is not yet flying. Starcloud, in his framing, is "ultimately built around the potential of SpaceX's Starship to drive down launch costs enough to build out an orbital inference layer that can compete with terrestrial data centres." Starship is still working toward that reliability, which makes it the single biggest dependency in the plan.
There are open questions beyond launch, too: radiation degrades electronics, you cannot send a technician to swap a failed server, and beaming large volumes of data up and down has its own limits. A single GPU in orbit is a milestone, not a data centre. But with Nvidia and Cisco now on the cap table, the question has shifted from whether anyone takes orbital compute seriously to whether the launch economics arrive in time.
The round at a glance
| Raise | 250 million dollar extension to the March Series A (which was 170 million) |
| Valuation | 2.3 billion dollars, up from 1.1 billion in March |
| Total raised | ~450 million dollars since founding in 2024 |
| Lead | Manhattan West |
| New investors | Nvidia (25 million dollars) and Cisco Investments |
| The pitch | Constant solar power and radiative cooling in orbit for AI compute |
| Proof so far | Company says it runs the only Nvidia data-centre GPU in orbit and first trained a model there |
| Ambition | FCC request to operate up to 88,000 spacecraft; next craft is Starcloud-3 |
| Biggest risk | Depends on SpaceX's Starship cutting launch costs |
It is the boldest expression yet of a trend that has been building all year: AI's growth is now a story about energy and infrastructure as much as about models, and the search for somewhere, anywhere, to put the next generation of compute. Whether that somewhere is 500 kilometres up depends less on the chips than on the rocket meant to carry them.
Honor's Robot Phone puts a camera on a folding robotic arm, and yes, it wiggles at you
Honor has launched a phone whose main camera sits on a motorised titanium arm that pops out of the body, swivels to follow you, and, in a demo, danced and nodded along like a pet. Underneath the theatrics is a serious camera partnership with cinema-gear maker ARRI. Here is what the Robot Phone actually is, what the arm does, and the very large catch.
An AI designed protein binders largely on its own, and independent labs confirmed they actually stick
Anthropic says its Claude models ran the design work for a protein-binder campaign, choosing where to bind on a set of target proteins and generating and screening candidates with little human input. Two commissioned outside labs, Adaptyv Bio and Twist Bioscience, then built and tested the designs: 354 of 1,320 bound their targets, hitting 14 of 15 evaluated targets at rates above the roughly 10 to 15 percent Anthropic calls typical. It is a real step for AI in the lab, but a 'binder' is an early research tool, not a finished drug; the labs confirmed the proteins bind, while the autonomy account is Anthropic's own.
Is AI a bubble? The money is real, and so is the circle it moves in
The AI build-out is one of the biggest infrastructure bets in history, funded by a web of deals in which the same companies invest in, sell to and buy from each other. Critics call it round-tripping; defenders call it vendor financing. Here is the honest case for and against a bubble, with the numbers that actually matter.