The chip does not exist. Starcloud is building a satellite constellation around a graphics processing unit that will not ship until late 2028, and on Friday, investors decided that gap was worth $2.3 billion.
The Redmond, Washington startup added $250 million to the $170 million Series A it closed in March, when it was valued at $1.1 billion. Five months later, the valuation more than doubled. Nvidia and Cisco Investments joined as new backers, alongside Cedar Capital, Goanna Capital and Standard Capital. Manhattan West led the round. Existing investors Benchmark, EQT, Soma, NFX and 776 all returned. Total capital raised since Starcloud's 2024 founding now stands at $450 million (Businesswire, 2026).
The valuation is a bet on flight data, not a working chip
Starcloud put the first data center-grade GPU into orbit last November, an Nvidia H100 riding a satellite called Starcloud-1. That satellite trained the first AI model in space and ran a version of Google's Gemini above the atmosphere, according to the company. Surviving long enough to run those workloads is a real result. It says nothing about whether Nvidia's purpose-built space chip works, because that chip does not exist yet.
That distinction is the whole thesis behind Nvidia's participation. The company is using Starcloud's flight data to design the Vera Rubin Space-1 module, a GPU built from the ground up for radiation exposure and orbital thermal swings. Shashi.co covered the Vera Rubin Space-1 announcement in March, when Starcloud was one of five named launch partners for a chip that existed only as a roadmap slide ("When the Chip Race Leaves Earth," shashi.co, March 2026). Five months later, the roadmap slide has a customer willing to put $25 million behind it, a figure a person familiar with the deal gave to TechCrunch and Dealroom.
Cisco's presence in the round reads differently. Cisco Investments is buying a seat at the table for whatever networking standard ends up connecting satellites to each other and to ground stations at gigawatt scale. That bet costs less to be wrong about.
Manufacturing capacity was never the constraint
Starcloud says the new capital funds a 100,000 square foot facility in Woodinville, built to mass-produce the 200-kilowatt Starcloud-3 satellite. The factory can already build satellites faster than the company can launch them. Launch capacity, not manufacturing, is the real constraint on Starcloud's plan.
SpaceX's Falcon 9 program is scheduled to wind down around 2028, and rideshare reservations on the rocket are already booked past late 2028 into 2029, according to SpaceX partners cited by SpaceNews. Starship, the vehicle Starcloud needs for Starcloud-3, has not reached the operational cadence that would let a startup plan a manufacturing ramp against it. CEO Philip Johnston told TechCrunch the company will need "an enormous amount of launch" in the years ahead. That phrase explains more about this raise than the valuation does.
Starcloud is not alone in reading the launch market this way. Cowboy Space raised $275 million at a $2 billion valuation earlier this year with a different answer to the same problem: build its own rockets, with upper stages designed to double as orbital compute platforms once the payload reaches orbit. Starcloud is choosing to depend on someone else's rocket and hedge with capital instead. Its near-term plan routes around the Starship bottleneck entirely: two 8-kilowatt Starcloud-2 satellites are booked on rideshare missions in 2027 to serve inference workloads for customers that include U.S. government agencies, a smaller and more immediate business than the 88,000-satellite constellation named in Starcloud's FCC filing.
The orbital compute thesis is bigger than one startup
Shashi.co has tracked the orbital side of this buildout since SpaceX began renting compute capacity to Google and Anthropic ahead of its own IPO ("SpaceX Is Not a Cloud Company. It Is Becoming One Before the IPO," shashi.co, June 2026). Starcloud's round confirms that the constraint SpaceX faces on the ground, enough skilled labor and enough power to keep building terrestrial data centers, has a mirror image in orbit: enough launch slots to get anything up there at all.
Google's Project Suncatcher is exploring the same orbital territory. So is SpaceX itself, with a filing for up to a million data center satellites under a project called Starmind. Flight hours are Starcloud's advantage right now. A satellite that has already trained a model in orbit beats a filing every time.
What happens next
Starcloud-2, the pair of 8-kilowatt satellites headed for rideshare launch in 2027, is the more immediate test. If those satellites run the orbital inference workloads Starcloud has promised its early government customers, the company has a real argument that its flight data, not its funding round, is the asset worth $2.3 billion. If Starship still has not reached a reliable commercial cadence by the time Starcloud-3 is ready to fly, the manufacturing facility in Woodinville becomes expensive storage.
Two questions carry more weight than the valuation. Does the Vera Rubin Space-1 module ship on the timeline Nvidia has set, and does Starcloud have a contracted launch slot when it does.
Business Wire. "Starcloud Raises $250 Million at $2.3 Billion Valuation to Scale AI with Orbital Data Centers." Business Wire, 21 Aug. 2026, businesswire.com.
Fernholz, Tim. "Starcloud Raises $250 Million for Orbital Data Centers as Launch Options Dry Up." TechCrunch, 21 Aug. 2026, techcrunch.com.
Rainbow, Jason. "Nvidia Joins Starcloud's $250 Million Orbital Data Center Funding Round." SpaceNews, 21 Aug. 2026, spacenews.com.
Dealroom. "Starcloud Raises $250M to Put AI Data Centers in Orbit." Dealroom.co, 21 Aug. 2026, dealroom.co.
Reuters. "Orbital Data Center Startup Starcloud Valued at US$2.3 Billion in Latest Funding." Reuters, 21 Aug. 2026, reuters.com.
Bellamkonda, Shashi. "When the Chip Race Leaves Earth." shashi.co, Mar. 2026, shashi.co.
Bellamkonda, Shashi. "SpaceX Is Not a Cloud Company. It Is Becoming One Before the IPO." shashi.co, Jun. 2026, shashi.co.
