Starcloud Funding Hits $450M as Nvidia Backs 20-GW Orbital AI Plan

A $2.3 billion bet on data centers in orbit

Starcloud, a startup from the Seattle area, has added $250 million to its Series A. The extension values the company at $2.3 billion after the new money, and it brings Nvidia into the cap table directly.20 Manhattan West led the round. Existing investors Benchmark, EQT, Soma, NFX and 776 returned, and the new names include Nvidia, Cisco Investments, Cedar Capital, Goanna Capital and Standard Capital.15 Starcloud was founded in 2024 and has now raised $450 million in total.20

The headline figure is the company's long-term goal: a constellation of 88,000 satellites delivering 20 gigawatts of computing capacity in orbit.18 The money itself is going to more ordinary things. Starcloud says it will pay for more manufacturing capacity, joint engineering work with Nvidia, and launch slots on future rockets.20 Production lines for its next spacecraft, Starcloud-3, are being built at a new 100,000-square-foot plant in Woodinville, Washington.1

The 20-gigawatt number is what made the news. The funding story is the speed at which investors have repriced the company.

Doubling the valuation in five months

In late March, Starcloud raised $170 million at a $1.1 billion valuation in a round co-led by Benchmark and EQT Ventures.11 That made it the fastest company from Y Combinator's accelerator to reach a $1 billion valuation, just 17 months after finishing the program.13 Before that, Starcloud had raised $34 million from backers including Andreessen Horowitz and In-Q-Tel, the CIA's venture capital arm.19

The August extension more than doubled the valuation only five months later.18 Total funding went from about $200 million to $450 million in that time.1118 Even by the standards of AI infrastructure, that is fast repricing for a company that has flown one satellite in orbit.

The rest of the cap table matters too. Benchmark general partner Chetan Puttagunta took a board seat in the March round.13 Manhattan West's Lauren Selig joins as a board observer after the extension.20 Manhattan West had a small stake in the March round and led the August one, so an existing investor chose to raise its bet sharply.1118 Infrastructure money is also involved. EQT runs more than 70 data centers on the ground, and Macquarie Capital's infrastructure arm joined the earlier round.13 Those backers usually care about predictable returns on long-lived assets, and their presence suggests Starcloud is being pitched as future infrastructure as well as a venture bet.

Why Nvidia's check matters more than its size

Nvidia's stake is reportedly about $25 million, a small share of the $250 million round.12 Its importance comes from the history behind it. Starcloud is a graduate of Nvidia's Inception startup program. In November 2025, its first satellite, Starcloud-1, carried an Nvidia H100 GPU, a data-center-grade AI chip.5 On that satellite, Starcloud trained Andrej Karpathy's small NanoGPT model on the complete works of Shakespeare, and ran Google's open Gemma model.6

Nvidia then made space a product line. At its GTC conference in March, the company announced the Space-1 Vera Rubin Module and named Starcloud among its launch partners, alongside Aetherflux, Axiom Space, Kepler Communications, Planet and Sophia Space.4 Nvidia says the module's Rubin GPU offers up to 25 times the AI compute of the H100 for inference in space.4 Nvidia executives have said it will be available in 2027.10 Starcloud's satellites are meant to be an early flight platform for those space-rated chips.15

Read together, the coverage suggests Nvidia is buying a test platform as much as making a financial investment. One analysis says Nvidia is believed to have negotiated access to Starcloud's data on how radiation affects its chips, but no other coverage confirms this.16 The broader point holds without that detail. A chip supplier investing in a customer that serves as its main testbed in orbit has a strategic reason to do so, separate from what the valuation does.

The 20-gigawatt claim, in context

Twenty gigawatts is a very large number. Newsweek compares it to the output of several large power stations.3 The goal has also grown. In late 2025, Starcloud talked about a single 5-gigawatt orbital data center with solar and cooling panels about 4 kilometers on each side.56 The plan has since become a distributed constellation. In February 2026, Starcloud asked the US Federal Communications Commission (FCC) for permission to launch up to 88,000 satellites.12

The company's pitch is the same across the coverage. It says AI demand is running into the limits of power grids on Earth, while orbit offers nearly constant sunlight and lets satellites shed heat by radiating it into space instead of using water-based cooling.115 CEO Philip Johnston has claimed energy costs ten times lower than on the ground, even after paying for launches.6

Reporters treat that claim more cautiously than the company does. Morgan Stanley analysts have pointed to radiation, the difficulty of repairs in orbit, space debris, and regulatory questions.6 Newsweek notes that cooling is still a major obstacle and calls orbital computing unproven.3 One analysis cites an August 2026 Boston Consulting Group estimate that, counting all costs, space data centers cost 2.5 to 3 times as much as data centers on the ground. It also calculates that if launch prices stall around $1,500 per kilogram, Starcloud's projected energy costs would be well above parity with the ground.16 Analysts generally expect near-term revenue in this sector to come from ground-based contracts while orbital deployment waits for cheaper launches.14

Where the reporting diverges

The sources agree on the round size, valuation, investor list and the 88,000-satellite target. They disagree on timing, which matters most to anyone trying to judge the valuation.

Starcloud-2, the commercial follow-up to the first satellite, is the clearest example. In March, Via Satellite reported that it would launch later in 2026.11 CNBC earlier reported an October 2026 launch with Nvidia Blackwell chips and a cloud platform from Crusoe on board.6 SiliconANGLE's coverage of the August round puts the launch in 2027.17 That looks like slippage. If so, the valuation doubled while the next milestone that would justify it was moving further out.

The sources also differ on how much the company is earning. One analysis says Starcloud had disclosed no commercial revenue as of April.16 In March, Johnston told Reuters that the committed customer contracts were mainly for other spacecraft, such as Earth-observation satellites. He said binding energy agreements with hyperscalers, the largest cloud providers, were still being negotiated.19 Some secondary analysis describes deeper partnerships with AWS and Google Cloud, but the main wire coverage is more restrained on that point.1619

The competitive squeeze

Starcloud is not raising money alone. Reuters framed its March round against SpaceX's acquisition of xAI and SpaceX's stated plan for an orbital data center network of a million satellites.19 SpaceX is reportedly developing its own AI satellite, called AI1, with computing capacity similar to Starcloud-3, and hopes to start mass production as early as late 2027.17 SpaceX is also Starcloud's launch partner. In May, Starcloud signed a contract to put Starlink laser links on its next 25 satellites.1712 So Starcloud depends on its biggest potential competitor for both launches and connectivity.

Below Starcloud, the field is filling with smaller companies. Aetherflux has set a target of early 2027 for its first orbital data center satellite.6 Sophia Space closed a $10 million seed round for space computing systems.10 Coverage of the extension also notes that launch availability has become a growing concern for every company in the category.14 That helps explain why Starcloud is putting money aside to secure future launch capacity.20

The reading

This round is best understood as investors paying for a position, not for proof that the economics work. Starcloud has shown that a data-center GPU can run AI workloads in orbit. It has not shown that orbital computing can be cheaper than computing on the ground, and the strongest independent cost analysis cited in the coverage points the other way.16 The 20-gigawatt constellation is a long-range goal, not a plan for the next few years. Even one favorable analysis treats it as something for the mid-2030s or later.16

The investors are buying three things: the company with the clearest record of firsts in the sector, a close engineering partnership with Nvidia, and factory and launch capacity in a market where both are scarce. The next test is Starcloud-2. If it launches on schedule and runs paying workloads, the $2.3 billion valuation will look early rather than reckless. If it slips again while SpaceX ramps its own hardware, this round may be remembered as the peak of the orbital-compute boom.