Nvidia Backs Starcloud's $2.3B Orbital Data Bet

The deal

Starcloud, the Woodinville, Washington startup building orbital AI data centers, said on August 21, 2026 that it raised a $250 million extension to the $170 million Series A it closed in March 2026, taking its post-money valuation to $2.3 billion and its total capital raised since the company's 2024 founding to $450 million. The round was led by Manhattan West, with new investors Nvidia, Cisco Investments, Cedar Capital, Goanna Capital and Standard Capital joining existing backers Benchmark, EQT, Soma, NFX and 776.

TechCrunch's own headline on the news put the new money at $200 million, a discrepancy with the $250 million figure Starcloud and Manhattan West gave BusinessWire; this article uses the company's own numbers as authoritative. Starcloud said the proceeds will fund manufacturing capacity, an engineering collaboration with Nvidia, and future launch procurement for its satellites.

Who is betting on it and why

Nvidia's presence on the investor list is the detail worth pausing on: the company whose GPUs are the reason data-center power demand has become a political and regulatory problem across the UK and EU is now putting its own capital behind a company that wants to build compute where the grid does not reach at all. Manhattan West led the round, and the roster of new and returning backers spans generalist venture funds, chipmaker corporate venture arms and space-focused investors, a mix that reads as a bet on infrastructure timing as much as on any single technology.

The numbers show how quickly the round grew on top of itself.

The grid problem this is meant to solve

Servola has tracked a run of stories this month showing AI data-center demand outrunning the power grid that is supposed to feed it: Ofgem has proposed a Data Centre Commitment Fee after contracted UK grid-connection demand jumped from 41 gigawatts to 125 gigawatts in seven months, Amazon's AI cluster in Shreveport, Louisiana is straining local grid, water and labor capacity, and Ohio regulators ruled that ratepayers, not OpenAI, must cover the grid upgrades an 8-gigawatt campus requires. The EU faces its own version of the same queue: grid-connection waiting lists are stretching years in several member states as data-center demand competes with industrial electrification for the same substations.

Starcloud's answer is to skip the grid entirely: its stated vision is an 88,000-satellite constellation delivering 20 gigawatts of orbital compute capacity, powered by solar panels that, in orbit, see near-continuous sunlight rather than the day-night cycle a ground-based solar farm has to work around. The company is building a new 100,000-square-foot manufacturing facility in Woodinville for its next satellite generation, Starcloud-3, to begin scaling toward that constellation.

What is still unproven

None of this makes orbital compute an imminent substitute for terrestrial data centers. Launch costs still dominate the economics of putting hardware into orbit, thermal management is harder in a vacuum than in an air-cooled hall, network latency limits which workloads even make sense off-planet, and a chip failure in orbit cannot be fixed with a technician and a replacement part the way a ground failure can. Starcloud itself is still pre-revenue at any meaningful scale, and 20 gigawatts of orbital capacity remains a target, not a shipped product.

What the round does prove is narrower and still worth noting: Nvidia, the company whose chips are driving the terrestrial grid-connection crunch, is now pricing a no-grid-required architecture as a hedge worth funding at a $2.3 billion valuation. For any UK or EU operator watching its own grid-connection queue stretch into years, that is the first real signal of where the largest AI infrastructure investors think the ceiling on ground-based compute actually sits.

Read next: Brazil Will Not Pick a Side in the AI Chip War  |  Domyn's $1.1bn AI Buildout Is 90% Debt