$250 million. That is the Series A extension Starcloud closed on 21 August, taking its post-money valuation to $2.3 billion and total capital raised since 2024 to $450 million. Nvidia and Cisco Investments joined the round led by Manhattan West. Existing backers Benchmark, EQT, Soma, NFX and 776 returned.
Starcloud now holds the clearest hardware-to-orbit path among pure-play orbital data-centre startups: an H100 already flown, a Blackwell-class Starcloud-2 scheduled for 2027 rideshare, and a 200 kW Starcloud-3 designed for Starship.
Launch capacity has become the binding constraint. Falcon 9 rideshare slots are unavailable beyond late 2028; the company is therefore using part of the new capital to lock Starship and alternative manifests early.
Nvidia’s equity check converts a technology partnership into a shared roadmap. The Space-1 Vera Rubin Module is intended to deliver up to 25× the AI compute of the H100 already demonstrated on orbit.
The same week SpaceNews and Data Center Dynamics reported the raise, the broader orbital-compute race tightened. Star Catcher had closed its own $65 million round on may earlier for an in-space power grid—as we wrote in August. Both deals point to the same structural shift: once launch cost falls far enough, the scarce resource is no longer mass to orbit but continuous power and radiation-tolerant silicon.
March 2026 Series A: $170 M at $1.1 B
August extension: $250 M at $2.3 B
Total raised: $450 M
Lead: Manhattan West · New strategic: Nvidia, Cisco
What the money actually buys
Starcloud’s press release and the SpaceNews account are unusually specific. The capital is earmarked for three items: a 100 000-square-foot production facility already under fit-out in Woodinville, Washington; engineering collaboration on the Space-1 Vera Rubin Module; and launch procurement. The company employs roughly 25 people today. Scaling the factory is therefore the near-term operational risk.
Starcloud-1 (60 kg, launched November 2025 on a Falcon 9 rideshare) carried the first terrestrial-grade Nvidia H100 into low Earth orbit and used it to train NanoGPT, a minimalist educational model written by Andrej Karpathy. That demonstration is the only public proof-point that a data-centre GPU can survive the thermal and radiation environment long enough to perform useful work. Starcloud-2 is planned as a 450 kg, 8 kW platform for early 2027; Starcloud-3 is the three-ton, 200 kW vehicle sized for Starship’s PEZ dispenser.
Technical path versus pure software claims
Launch bottleneck as the real bottleneck
CEO Philip Johnston told TechCrunch that Falcon 9 rideshare reservations beyond late 2028 or early 2029 are effectively closed. Starship remains the only vehicle that can lift the 200 kW class at acceptable cost per kilogram. Until Starship cadence is proven, Starcloud can only tread water with the smaller Starcloud-2 vehicles. Part of the $250 million is therefore option money on future launch manifests—an unusual but rational use of growth capital when the constraint is physical rather than financial.
The same constraint is visible across the sector. Impulse Space’s Caravan GEO rideshare programme is already fully booked into 2028; small-satellite operators report multi-year waits for Transporter slots. Orbital data centres that require frequent, high-mass launches will feel the squeeze first.
An 88 000-satellite constellation delivering 20 GW remains an FCC application, not an authorisation. Regulatory and spectrum coordination risk is real; the near-term valuation is supported by the flown hardware and the Nvidia relationship rather than the ultimate constellation size.
Where the competitive map sits
Cowboy Space, Axiom Space, and several a16z-backed entrants have announced orbital-compute programmes. None has yet flown a terrestrial data-centre GPU or trained a model on orbit. Starcloud’s first-mover hardware position is therefore durable for at least the next 18–24 months, provided the Woodinville line can deliver Starcloud-2 on the stated 2027 rideshare.
Nvidia’s participation is the strategic differentiator. The company is also an investor in SpaceX; placing its next-generation space-hardened module on Starcloud satellites creates a dual-path distribution for space AI silicon. That alignment is difficult for pure-play software or edge-AI competitors to replicate quickly.
Industrial logic behind the valuation
At $2.3 billion the company is priced as if the first 200 kW vehicles will reach orbit and generate revenue before the next major capital raise. That assumption rests on three engineering facts already demonstrated: the H100 survived and performed useful work; the radiator and power subsystem on Starcloud-1 scaled cleanly enough to justify an 8 kW follow-on; and Nvidia is willing to co-develop a radiation-tolerant module rather than simply sell commercial chips. None of those facts existed twelve months ago.
The factory in Woodinville is the operational hinge. A 25-person team cannot simultaneously finish Starcloud-2, qualify Starcloud-3 structures, and negotiate multi-year launch manifests. The new capital therefore buys parallel workstreams. If the production line slips, the launch-procurement money becomes stranded option value. That is the primary near-term execution risk investors should model.
Compared with pure software orbital-compute concepts, Starcloud’s capital intensity looks high. Compared with terrestrial AI data-centre projects that now face multi-year grid interconnection queues and multi-billion-dollar power-purchase agreements, the space path begins to look capital-efficient once Starship pricing is real. The $250 million extension is the bridge capital that keeps the option open until that pricing arrives.
One further data point from the SpaceNews report is worth isolating: early customer conversations already include Crusoe, an AI infrastructure provider that understands both terrestrial power constraints and the value of low-latency inference near the sensor. That customer profile is more realistic than generic cloud-in-space marketing. It also explains why Nvidia’s equity participation matters—the same company that supplies the silicon can introduce the first buyers.
What happens if Starship cadence slips?
Signals to watch
FCC processing of the 88 000-satellite application and any spectrum coordination filings.
First public announcement of a Starship launch contract or multi-manifest reservation.
Delivery schedule and power telemetry from the first Starcloud-2 vehicle.
Whether Nvidia’s Space-1 Vera Rubin Module appears on any competing orbital platform in 2027–28.
A — 45 % Starship reaches commercial cadence by late 2027; Starcloud-3 flies 2028; early customers include Crusoe and defence imagery providers; valuation holds or expands.
B — 35 % Starship slips 12–18 months; company continues Starcloud-2 cadence on Falcon 9 / alternatives; revenue grows but capital intensity rises; secondary raise or strategic partnership required.
C — 20 % Regulatory or technical setback on radiation-hardened modules; competitor flies comparable silicon first; Starcloud loses first-mover narrative and must compete on price and power alone.