Starcloud Inc.
Orbital AI data center pioneer, headquartered in Redmond, Washington — the first operator of an Nvidia H100 in orbit
Starcloud Inc. was founded in January 2024 in El Segundo, California under the name Lumen Orbit, and was rebranded to Starcloud around February–March 2025 in what appears to have been a trademark-related move. The company is now headquartered in Redmond, Washington, with a dedicated manufacturing facility located near the satellite production lines of SpaceX and Amazon. In our assessment, the company’s core investment thesis is that it circumvents the power, land, and permitting constraints that increasingly bind terrestrial data centers by locating AI inference infrastructure in orbit, where solar energy is effectively unconstrained and radiative cooling operates without atmospheric interference.
Born 1986 in the United Kingdom. Holds a First Class BSc in Applied Mathematics and Theoretical Physics from the University of Nottingham, an MA from Columbia University, an MBA from Wharton, and an MPA from Harvard Kennedy School, where he served as Student Body President representing roughly 1,200 graduate students and later taught as a Teaching Fellow in Entrepreneurial Finance. Per company disclosure, he previously worked at McKinsey & Company and has a background on projects with national space agencies. This is his second venture: he co-founded Opontia, a digital brand aggregator that ranked 12th on Forbes Middle East’s list of most-funded startups in 2021 before being acquired by Perfection in 2023.
Brings space hardware engineering experience from Oxford Space Systems and Airbus Defence and Space. Leads Starcloud’s satellite platform architecture, drawing on prior work in structural design and deployment mechanisms for orbital environments.
Spent more than two decades as a software engineer at Microsoft before joining SpaceX’s Starlink division, where he worked on GPU clusters and satellite tracking-beam systems. His combined background in hyperscale software infrastructure and satellite communications informs Starcloud’s compute-and-connectivity architecture, in our view a relatively uncommon skill overlap in the sector.
Prior mechanical engineering experience at Helion Energy and SpaceX. Applies high-reliability hardware development experience from fusion and launch-vehicle programs to thermal management and structural design for Starcloud’s satellites.
Benchmark partner Chetan Puttagunta joined the board of directors alongside the March 2026 Series A, introducing formal venture-governance oversight at a relatively early stage in the company’s lifecycle.
Approximately 21 months after founding, Starcloud launched its first satellite, Starcloud-1, in November 2025 aboard a SpaceX Falcon 9 rideshare mission. The spacecraft carries an Nvidia H100 GPU, which per company disclosure is the most powerful GPU operated in orbit to date. Starcloud has stated it is the first to train an LLM (nanoGPT) in orbit and the first to run Google’s Gemini/Gemma models on an orbital GPU; we note these are self-reported technical firsts that we have not independently verified beyond company and press disclosures.
Headcount figures are inconsistent across sources: 12 employees per Starcloud’s YC profile (as of late 2025), 15 per a 2026 Wikipedia entry, and approximately 25 per TechCrunch’s August 21, 2026 report. This may reflect genuine hiring velocity, but as self-reported figures with no fixed reference date, they should not be treated as directly comparable across time.
In our assessment, Starcloud’s operating model rests on three pillars.
Provides GPU compute in orbit and transmits only processed results to the ground, reducing bandwidth costs and data latency versus terrestrial alternatives. The 8kW-class Starcloud-2 compute satellite is slated for a 2027 rideshare launch and is intended to serve customers including US government agencies. The company has referenced a partnership with Crusoe Cloud to offer orbital GPU capacity from early 2027, though this detail derives from third-party sourcing and warrants cross-verification against company disclosure.
Designs and builds its own satellites out of a 100,000-square-foot facility in Woodinville, Washington. The company has opted for automotive-grade commercial off-the-shelf components rather than space-rated hardware to reduce cost, offsetting unproven radiation-hardness risk with its own flight data. Starcloud-3, the company’s largest planned orbital data center spacecraft, is designed around a future SpaceX Starship launch and is reported at roughly 200kW and three tons.
Starcloud is sharing orbital flight data — covering the relationship between chip operating temperature and radiator sizing, radiation-shielding placement, and launch-survivability ruggedization — with Nvidia as it develops its first purpose-built space GPU, the Vera Rubin Space-1. Target flight timing is late 2028; the chip has not yet been built, and this remains an early-stage program.
Target unit economics: CEO Johnston has stated that if launch costs fall to roughly $500 per kilogram, a Starcloud-3-class spacecraft could achieve costs on the order of $0.05 per kWh, positioning it as the first orbital data center to be cost-competitive with terrestrial facilities. We flag that this target is contingent on launch-cost declines tied to Starship, which has not yet flown commercially, and should be treated as a forward-looking management estimate rather than a demonstrated outcome.
Regulatory posture: On February 3, 2026, Starcloud filed with the US Federal Communications Commission for authorization to operate a constellation of up to 88,000 spacecraft. This represents a substantial scale-up relative to the single operational satellite (Starcloud-1) currently in orbit, and both the timing and probability of approval remain uncertain, in our view.
Since its January 2024 founding, Starcloud has raised approximately $440 million in aggregate disclosed external capital across roughly two years and seven months, from pre-seed through the Series A extension. Notably, the company’s valuation roughly doubled — from $1.1 billion to $2.3 billion — in the less-than-five-month span between its March 2026 Series A and its August 2026 extension round.
Co-founded by Philip Johnston, Ezra Feilden, and Adi Oltean in El Segundo, California. The founders booked a SpaceX rideshare launch slot the day after incorporation, an early signal of the execution pace that would characterize the company’s subsequent development.
Led by Nebular, with participation from Caffeinated Capital, Plug & Play, Everywhere Ventures, Tiny.vc, Sterling Road, and Foreword Ventures, alongside more than 20 angel investors including four affiliated with the Sequoia Scout Fund.
Graduated from YC’s Summer 2024 batch, establishing an early investor network that fed subsequent funding rounds.
An $11 million close in December 2024 was followed by an additional $10 million SAFE close in February–March 2025, bringing aggregate seed funding to $21 million — reported at the time as one of the largest seed rounds ever raised by a YC company. The company rebranded from Lumen Orbit to Starcloud around this period.
Launched via SpaceX Falcon 9 rideshare, approximately 21 months after founding. Per company and press reporting, this marked the first operation of a data-center-grade Nvidia H100 GPU in orbit and the first reported case of training an LLM in space.
Formal regulatory step supporting the company’s longer-term ambition to reach gigawatt-scale orbital compute capacity. Approval timing and probability remain undetermined.
Starcloud reached unicorn status at a $1.1 billion post-money valuation, achieved 17 months after its YC demo day — reported as the fastest unicorn in YC history. Benchmark partner Chetan Puttagunta joined the board. Proceeds were earmarked for Starcloud-2 (carrying a Blackwell chip, an AWS server blade, and a bitcoin-mining computer) and for development of Starcloud-3, the Starship-class spacecraft.
Deal structure: An extension of the March 2026 Series A, led by Manhattan West Ventures, with new participation from Nvidia and Cisco. A person familiar with the deal indicated Nvidia’s contribution was approximately $25 million; we note this figure is anonymously sourced and has not been confirmed by the company. Existing backers Benchmark, EQT, Soma, and NFX were joined by new or follow-on participants including 776, Cedar Capital, Goanna Capital, and Standard Capital.
Use of proceeds: Earmarked for a larger manufacturing facility, continued development of Starcloud-3, and securing forward launch capacity. CEO Johnston explicitly cited the scheduled 2028 retirement of SpaceX’s Falcon 9 program as a driver of the raise, framing launch-slot access as an increasingly binding constraint.
Strategic significance: We view Nvidia’s participation as reflecting the value of Starcloud-1’s flight data to the co-development of the Vera Rubin Space-1 chip; per Johnston, Nvidia conducted more extensive technical diligence ahead of the investment than any other participant.
• Round type: Extension of the March 2026 Series A ($170M)
• Total raised: $250,000,000
• Post-money valuation: $2,300,000,000 (roughly 2x the March 2026 mark)
• Lead investor: Manhattan West Ventures
• Key new participants: Nvidia, Cisco
• Follow-on participants: Benchmark, EQT, Soma, NFX, 776, Cedar Capital, Goanna Capital, Standard Capital
Round-size discrepancy: TechCrunch’s URL slug and initial page metadata reference “$200 million,” while the published headline and body text state “$250 million.” We have used the $250 million figure from the final published body text as our base case, but flag the discrepancy as indicative of a possible mid-publication revision that warrants confirmation against subsequent company disclosure.
Valuation-path ambiguity: In April 2026, The Information reported Starcloud was in talks for a new round at a $2.2 billion valuation; this was an unconfirmed, in-negotiation figure at the time. The round that ultimately closed in August 2026 was valued at $2.3 billion. The two figures are close but represent distinct events — a reported negotiation versus a confirmed closing — and should not be conflated.
Self-reported revenue commentary: Per the company’s YC profile, Starcloud-2 is expected to “generate more cash than it costs to build and launch.” This is an unaudited, forward-looking company estimate; actual revenue and profitability metrics remain undisclosed.
In our view, Starcloud’s competitive position rests on a demonstrated first-mover status backed by actual flight heritage, a strategic co-development relationship with Nvidia, and an execution cadence that took the company from founding to orbital demonstration in 21 months. Where much of the competitive field has been confined to low-power, edge-processing space GPUs, Starcloud is, to our knowledge, the only company to have operated a data-center-grade GPU in orbit — a differentiator we see as meaningful.
Starcloud-1 carries an Nvidia H100 and is reported to have processed real AI workloads in orbit, with the company claiming firsts in orbital LLM training and running Google’s Gemini/Gemma models in space. Unlike competing orbital-compute players largely confined to low-power edge chips, Starcloud has flight-proven hyperscale-class workload processing, which we view as a genuine technical moat.
Nvidia’s participation in the August 2026 round followed extensive technical diligence and is tied to a co-development arrangement in which Starcloud’s flight data informs Vera Rubin Space-1, Nvidia’s first purpose-built space GPU. We see this as extending beyond capital participation to preferential access to Nvidia’s future space-silicon roadmap — a relationship-based barrier to entry that would be difficult for competitors to replicate quickly.
Starcloud moved from founding to orbital flight demonstration in 21 months, and from YC demo day to unicorn status in 17 months — reported as the fastest such progression in YC history. The company’s investor base includes top-tier institutional names (Benchmark, EQT, Manhattan West Ventures) alongside In-Q-Tel, whose participation we read as a signal of credibility on national-security-adjacent diligence.
CEO Johnston’s McKinsey background and reported national space agency project experience, CTO Feilden’s satellite hardware career at Airbus Defence and Space and Oxford Space Systems, and Chief Engineer Oltean’s Microsoft and SpaceX Starlink GPU-cluster experience together span both satellite hardware engineering and hyperscale compute infrastructure — a combination we view as a differentiator versus comparable orbital-compute startups.
Commercial-off-the-shelf component strategy: Starcloud has opted for automotive-grade commercial components rather than space-rated, radiation-certified hardware, validating reliability through its own flight data rather than pre-qualification. This lowers development cost and accelerates iteration speed, but in our assessment introduces a trade-off around long-term reliability risk in the radiation environment that has not yet been tested over extended mission durations.
In our assessment, Starcloud’s business model is fundamentally premised on the successful commercialization of SpaceX’s Starship, which has not yet flown commercially. As a result, notwithstanding the company’s demonstrated technical execution, we see it as exposed to meaningful exogenous risk that sits outside management’s direct control.
The business model’s central premise — Starship — remains unproven, while SpaceX’s Falcon 9 program is scheduled to be retired in 2028, tightening competition for launch slots. CEO Johnston himself acknowledged that an inability to secure SpaceX launch capacity by 2029 would present a material challenge to the business.
Valuation roughly doubled from $1.1 billion in March 2026 to $2.3 billion by August 2026 — a five-month span. We cannot rule out that this reflects late-stage private-market momentum and strategic-investor premium rather than validated cash-flow fundamentals, and would flag valuation durability as a point requiring further diligence.
As a private company, Starcloud has not disclosed audited financial statements, and revenue and profitability commentary rests on company estimates. Even basic metrics such as headcount are reported inconsistently across sources.
Competitor Cowboy Space has moved to address the launch bottleneck by building its own rockets, while approval timing and probability for Starcloud’s 88,000-satellite FCC filing remain unresolved.
As power, land, and permitting constraints increasingly bottleneck terrestrial data center buildout across the AI industry, orbital compute has the potential to emerge as a structural long-term alternative.
Starcloud-2’s customer base is reported to include US government agencies, and the company’s relationship with In-Q-Tel points to a potential demand pipeline in national-security-adjacent markets.
Should the Vera Rubin Space-1 co-development succeed, Starcloud would be positioned to secure an early lead within the emerging space-native GPU ecosystem.
If Starship demonstrates reusability and launch costs fall toward the targeted ~$500/kg benchmark, Starcloud’s unit economics would have room for structural improvement.