Starcloud raises $250 million to build AI data centres in space

Key points

• Starcloud added a $250 million funding extension to its March 2025 Series A round, bringing the total raised to $420 million.
• The extension values Starcloud at $2.3 billion and was led by Manhattan West Ventures, with Nvidia contributing $25 million.
• Starcloud is the only known company running an Nvidia H100 GPU, a high-performance computer chip, aboard an orbiting satellite.
• The company plans to launch its next satellites, the Starcloud-2 generation, on shared rocket flights in 2027.
• Starcloud has asked US regulators for permission to eventually operate 88,000 spacecraft.

What if the server powering your AI application was not in a data centre outside Phoenix or Dublin, but circling the Earth at 17,000 miles per hour? That is the business Starcloud is building, and investors just handed it another $250 million to try.

The Seattle-area startup, first reported by TechCrunch AI, announced the funding extension this week. Added to the $170 million it raised in March 2025, Starcloud has now taken in $420 million in total and carries a valuation of $2.3 billion.

The round was led by Manhattan West Ventures. Nvidia, the chipmaker whose graphics processors do most of the heavy lifting inside AI systems worldwide, joined as a notable backer, putting in $25 million. Cisco, Benchmark, EQT and several other funds also participated.

Why would anyone run AI in space?

For now, almost all AI inference, meaning the process of a trained model answering a question or analysing an image, happens in giant land-based data centres. Starcloud argues that satellites can eventually do this work faster for some customers, particularly government and defence agencies that need data processed close to where it is collected, without beaming it back to Earth first.

The company already has one satellite operating in orbit with an Nvidia H100 chip aboard. The H100 is the same high-end processor used in many of the world's most powerful AI systems. Starcloud says it is the only company that has trained an AI model using a chip flying in space, and it is sharing what it learned with Nvidia as the chipmaker designs its first chip built specifically for space, called the Vera Rubin Space-1.

What is the launch problem?

Getting satellites into orbit is expensive and, right now, getting harder. SpaceX's Falcon 9, the workhorse rocket that carries most commercial satellites today, is scheduled to stop flying in 2028. SpaceX's replacement, the much larger Starship, is still being tested and has not yet flown a paying payload.

This matters for Starcloud because its whole growth plan depends on Starship eventually making launch cheap enough to put hundreds of satellites into orbit affordably. CEO Philip Johnston said this week that securing launch slots is now one of the company's biggest costs. The company has already applied to the US Federal Communications Commission for permission to fly 88,000 spacecraft.

Starcloud's next step is launching two Starcloud-2 satellites, each producing 8 kilowatts of computing power, on shared rocket flights in 2027. Its most ambitious design, the Starcloud-3, is intended to fly aboard Starship and is described as a full orbital data centre.

The company employs 25 people and builds its satellites at a 100,000 square-foot facility in Woodinville, Washington, near factories where SpaceX and Amazon also assemble satellites.

Common questions

Does this affect ordinary internet users?

Not yet. Starcloud's first customers are US government agencies. Broader commercial services, if they arrive, are years away.

What happens if Starship keeps being delayed?

Johnston acknowledged the risk plainly: if Starcloud cannot book SpaceX launch capacity by 2029, growth becomes very difficult. The company is also talking to other rocket providers as a backup.

Is the Nvidia chip for space ready to fly?

No. The Vera Rubin Space-1 chip has not been built yet. Starcloud hopes to fly it in late 2028, assuming the design and testing go to plan.