Lunar Orbit Compute: Why Starcloud Is Flying a Payload With Firefly

The phrase lunar orbit compute did not describe a product anyone could buy until this autumn. On 30 September 2026 Firefly Aerospace, the Nasdaq-listed launch and lunar lander company, announced a commercial payload agreement with Starcloud, a two-year-old startup that builds data centre satellites. Starcloud will place a processor called SC-1L on Firefly’s Elytra orbital vehicle, which is due to travel to the Moon on Firefly’s third lunar mission. The payload is small and the launch is at least two years away, but the agreement shows how two fashionable ideas, orbital data centres and commercial lunar services, are starting to trade with each other. This article is part of the Kurums Space Economy hub and looks at the deal from the point of view of revenue, customers and risk.

What did Starcloud and Firefly agree?

Firefly will carry Starcloud’s SC-1L compute processor on Elytra, its orbital transfer vehicle, to lunar orbit. The payload will ingest data from Elytra’s Solux vision system, run high-performance computing on it and transmit the results to Earth. Launch is targeted for no earlier than 2028.

According to Firefly’s press release, the flight is part of the company’s third lunar mission. Elytra will first act as a transfer vehicle and communications relay for a Blue Ghost lander heading to the Gruithuisen Domes, then remain in lunar orbit. Firefly says Elytra is designed to operate there for five years, which gives hosted payloads such as SC-1L a long working life after the lander’s surface mission is over.

Ray Allensworth, Firefly’s vice president of spacecraft, said the company was “proud to collaborate with innovative customers like Starcloud.” Ezra Feilden, Starcloud’s co-founder and chief technology officer, said the company had “successfully validated our ability to run and train AI models on enterprise grade GPUs in low Earth orbit” and, in comments reported by GeekWire, that “space is the future of data centers, and the moon is the next frontier for that vision.”

What the two companies did not say is as important. No contract value, payload mass, power allocation or chip model was published in the sources reviewed for this article. The agreement is best understood as a hosted-payload demonstration, not as the start of a data centre around the Moon.

Why would anyone want lunar orbit compute?

Processing data near the Moon reduces the amount that has to be sent across the Earth-Moon link, which is narrow and expensive. Cameras and sensors in lunar orbit can generate far more data than deep-space communications can return, so computing on board turns raw imagery into smaller, more valuable products.

Firefly made this argument a few weeks before the Starcloud deal, when it announced that NVIDIA Jetson modules would process imagery for Ocula, its planned lunar imaging service. In that release Firefly said on-orbit processing addresses deep-space bandwidth limits and can deliver insights for mapping and mineral detection. Ocula uses telescopes from Lawrence Livermore National Laboratory and software from Firefly’s SciTec subsidiary, and chief executive Jason Kim said it “is set to be the first commercial lunar imaging service available.”

The Starcloud payload extends the same idea with data centre class hardware. The first data source is Firefly’s own Solux vision system, so the initial customer for the compute is, in effect, the host spacecraft. The commercial question is whether third parties, such as lander operators, rover companies and space agencies, will later pay to have their data processed in lunar orbit. Nobody has announced such a customer yet.

Who is Starcloud and how does it make money?

Starcloud is a Redmond, Washington startup founded in January 2024 that plans to sell computing capacity from solar-powered satellites. It has flown one small demonstration satellite, has raised a reported US$450 million and intends to launch a larger satellite carrying customer workloads in late 2026.

The company was founded as Lumen Orbit by Philip Johnston, a former McKinsey consultant, Ezra Feilden, formerly of Airbus Defence and Space, and Adi Oltean, who worked at SpaceX and Microsoft Azure. It went through Y Combinator in 2024. Its first satellite, Starcloud-1, launched in November 2025. GeekWire describes it as a 130-pound spacecraft carrying an NVIDIA H100 chip, and the company says it used the satellite to run Google’s Gemma model and to train a small language model, nanoGPT, in orbit.

The business model is to sell GPU capacity in orbit in the way cloud providers sell it on the ground, with sunlight as the power source and radiators replacing water cooling. Starcloud-2, planned for late 2026, is described by GeekWire as generating 100 times the power of the first satellite, carrying NVIDIA’s Blackwell B200 chip and running customer workloads. Unite.AI lists Crusoe, AWS, Google Cloud and NVIDIA as early customers or partners, and Wikipedia records a partnership with Crusoe, announced in October 2025, to deploy Crusoe’s cloud platform with GPU capacity expected in early 2027.

Revenue has not been disclosed. Johnston told GeekWire in March that he expects the economics to favour space-based computing within “three to five years.” Until then Starcloud is a pre-revenue or early-revenue company funding hardware from equity, which is the context for a lunar payload: it is a relatively cheap way to add a second proof point.

What does Firefly get out of the deal?

Firefly gets a paying commercial payload on a spacecraft that is already flying for another purpose, plus a second computing partner alongside NVIDIA. Its shares rose about 6% on the day of the announcement, according to market coverage, although the contract’s value was not disclosed.

Firefly’s lunar business began with NASA delivery contracts for its Blue Ghost landers. Elytra changes the unit economics of those missions. The vehicle carries the lander towards the Moon and relays its communications, then stays in orbit for years. Every additional hosted payload or data service sold on Elytra is incremental revenue against a spacecraft whose primary cost is covered by the lander mission. Firefly has said that Blue Ghost Mission 2 is targeted for no earlier than late 2026 and that it has contracts for two further Elytra vehicles through Missions 3 and 4.

The Starcloud agreement therefore fits a pattern: Ocula imagery, NVIDIA Jetson processing and now a third-party compute payload all turn Elytra from a delivery stage into a platform. For a company that listed on Nasdaq under the ticker FLY, showing recurring, non-NASA revenue from lunar orbit is part of the equity story. Kurums has examined how other listed space companies made a similar move from one-off missions to platforms in the Rocket Lab company story and the Planet Labs company story.

Who else is competing in space-based computing?

Starcloud faces very large and very small rivals. SpaceX has filed with the FCC for up to one million data centre satellites, Google is working on Project Suncatcher, and GeekWire also names Axiom Space, Kepler Communications and Sophia Space as companies pursuing orbital computing.

The competitive picture has two layers. In low Earth orbit the contest is about scale: how many kilowatts of solar power and how many GPUs a company can launch, cool and connect. SpaceX controls launch and a laser-linked network, and Starcloud itself buys Starlink laser terminals for its next 25 satellites, which makes its largest potential competitor a key supplier. Starcloud’s own filing for up to 88,000 satellites, and the 20 gigawatt target cited by Unite.AI, show the scale of ambition, not of deployed hardware.

A second layer consists of companies that avoid building satellites at all. Satlyt, for example, is assembling a virtual network from spare processing capacity on other operators’ spacecraft; Kurums covered Satlyt’s seed round earlier this autumn. Not everyone is convinced by the category. GeekWire quoted Microsoft president Brad Smith saying the company was not “betting on” satellite infrastructure strategies at present.

Around the Moon there is, so far, little direct competition, because there is little demand. That is both the attraction and the weakness of the Firefly agreement. Being first costs relatively little, but the market being addressed does not yet exist at any meaningful size.

What does the deal mean for founders, operators and investors?

For founders, it shows that platforms with spare capacity are a faster route to flight than building a spacecraft. For operators, it signals that processing at the edge will be offered as a service beyond Earth orbit. For investors, it adds narrative more than near-term revenue.

Founders should note the structure. Starcloud is not launching a lunar satellite. It is renting space, power and communications from a company that already has to go to the Moon. The same logic applies to startups building sensors, communications terminals or manufacturing experiments. Our article on Outpost’s Space Factories service describes a similar offer for companies that want to manufacture in orbit without building a spacecraft.

Operators of lunar missions gain an option. If compute is available in lunar orbit as a service, a rover or lander company could design for a lower downlink budget and buy processing instead. That only works if the service is reliable and priced sensibly, and neither has been demonstrated.

Investors should separate the two businesses. Starcloud’s valuation, which doubled from US$1.1 billion in March to a reported US$2.3 billion in August, rests on the low Earth orbit data centre thesis and on Starcloud-2 working. The lunar payload does not change that. For Firefly shareholders the deal is a small positive data point on Elytra utilisation. In both cases the absence of disclosed financial terms means the agreement cannot be modelled, only noted. Readers comparing capital raised across the sector can use the Kurums space funding rounds archive.

What are the main technical and commercial risks?

The main risks are schedule, thermal and radiation performance of commercial GPUs far from Earth, and the lack of identified paying users for lunar processing. All three are unresolved today, and the companies have published few technical details about SC-1L.

Schedule comes first. Blue Ghost Mission 2 is itself targeted for no earlier than late 2026, and the Starcloud payload flies on the mission after that. Lunar programmes have a history of slipping, and a hosted payload has no control over the host’s timeline.

The technical risk is that hardware proven in low Earth orbit meets a harsher environment. Starcloud’s flight experience comes from one small satellite inside the partial protection of Earth’s magnetic field. Lunar orbit offers no such protection and different thermal conditions. Starcloud says it has validated running and training models in low Earth orbit; it has not yet published results from Starcloud-2, the satellite meant to show commercial-grade power and cooling.

The commercial risk is demand. The first workload comes from Firefly’s own vision system. A market exists only if several lunar missions generate more data than they can downlink and are willing to pay a third party to process it. That depends on how many missions actually fly in the next five years.

What should you watch next?

Watch the Starcloud-2 launch and its first customer workloads, Firefly’s Blue Ghost Mission 2 and the debut of Elytra, any disclosure of pricing for hosted compute, and announcements of third-party users for lunar processing.

Starcloud-2 is the nearer and more important milestone. If it flies in late 2026 and runs paid workloads, the company will have a product. If it slips, the lunar plan becomes harder to take seriously. On Firefly’s side, the first Elytra flight will show whether the vehicle can deliver the multi-year orbital life that the hosted-payload model assumes.

It is also worth watching how servicing and logistics companies fit into this picture, because long-lived orbital platforms eventually need maintenance. The failure of Katalyst’s Swift rescue mission this summer is a reminder that first operational flights of new space services often go wrong, and that contract structure decides who bears the cost.

Frequently Asked Questions

Sources

• Firefly Aerospace press release: Firefly signs Starcloud as commercial customer (30 September 2026)
• GeekWire: Starcloud to test AI computing in lunar orbit with Firefly
• Parameter: Firefly Aerospace stock rises on lunar AI computing partnership
• Firefly Aerospace press release via Barchart: on-orbit processing for Ocula with NVIDIA Jetson
• Unite.AI: Starcloud raises $250M Series A extension at $2.3B valuation
• GeekWire: Starcloud hits $1.1B valuation to build space-based data centers
• Wikipedia: Starcloud