이미지: TechCrunch AI
Summary
- Starcloud, which develops satellites for orbital AI inference, raised an additional $250 million following its $170 million Series A in March, pushing its valuation up to $2.3 billion.
- CEO Philip Johnston said securing launch capacity is becoming increasingly difficult as SpaceX's Falcon 9 is slated for retirement in 2028.
- NVIDIA invested $25 million in this round, and Starcloud is providing NVIDIA with data from training models on H100 GPUs in orbit to inform NVIDIA's development of space-grade chips.
- 추가 유치액
- 2억5000만달러(3월 시리즈A 1억7000만달러 확장)
- 기업가치
- 23억달러
- CEO
- 필립 존스턴(Philip Johnston)
- 주도 투자자
- Manhattan West Ventures, 엔비디아(2500만달러), 시스코, 벤치마크 등
- FCC 신청 위성 수
- 8만8000기
- 차기 발사 계획
- 8kW급 컴퓨트 위성 'Starcloud-2' 2기, 2027년 라이드셰어 발사
- 제조시설
- 워싱턴주 우드인빌, 약 9300평(10만 제곱피트) 규모, 직원 25명
Servers in orbit, valuation jumps to $2.3 billion
Starcloud, which develops satellites that perform AI inference in orbit, announced it has secured an additional $250 million. This is an extension to the $170 million Series A round it closed in March, and the new funding brings the company's valuation to $2.3 billion. Starcloud plans to use the capital to open a larger manufacturing facility and advance development of "Starcloud-3," a large orbital data center satellite set to launch aboard SpaceX's next-generation rocket, Starship.
Orbital data centers are a business segment built on the idea of processing AI computation at lower cost than on the ground, drawing power from solar energy and using the vacuum of space for cooling. But building a satellite and actually getting that satellite into orbit are entirely different problems. No matter how good a satellite is, it stays grounded without a rocket seat.
Rocket seats are scarce
CEO Johnston said that securing launch capacity has become the biggest variable for the business, as SpaceX's flagship rocket Falcon 9 is set to be retired from its program in 2028. As he put it, "launch is pretty constrained right now" — SpaceX is gradually phasing out Falcon 9 and transitioning to the still-unproven large rocket Starship. Around the same time, Blue Origin's New Glenn and ULA's Vulcan have yet to reach regular flight cadence, and Rocket Lab's Neutron has yet to make it to the launch pad. This is why Starcloud is also considering purchasing dedicated Falcon 9 launches or contracting with other launch providers in addition to SpaceX.
Starcloud has already filed with the Federal Communications Commission (FCC) for authorization to operate 88,000 satellites. In the near term, it plans to launch two 8kW-class compute satellites, "Starcloud-2," as a rideshare (launched alongside other satellites) in 2027, to perform orbital inference work for customers including US government agencies. According to this report, SpaceX's Elon Musk said this week that Starship recovery attempts would be pushed back several months, with the first attempt at re-flying the vehicle now planned for late this year or early 2027.
| Category | Timing | Amount | Valuation |
|---|---|---|---|
| Series A | March 2026 | $170 million | - |
| Additional round (extension) | August 2026 | $250 million | $2.3 billion |
Why NVIDIA put in $25 million
The round was led by Manhattan West Ventures, with participation from NVIDIA and Cisco. According to sources familiar with the deal, NVIDIA's investment amounted to $25 million, with Benchmark, EQT, Soma, NFX, 776, Cedar Capital, Goanna Capital, and Standard Capital also participating.
Johnston pointed to NVIDIA's investment as a signal of Starcloud's competitive strength in the space computing market. Starcloud is known as the first to actually operate NVIDIA's H100 — a data center GPU designed for ground use — in orbit and use it to train a model. This sets it apart from most other space-grade GPUs, which are designed for low-power edge processing. Johnston explained that Starcloud is sharing data from its first satellite, "Starcloud One," with NVIDIA, which is using it to develop its first dedicated space GPU, "Vera Rubin Space-1." The chip has not yet been built, and Starcloud plans to launch it into orbit for testing around the end of 2028. According to Johnston, engineers are focused on key variables such as the relationship between the chip's operating temperature and the size of the radiator needed to dissipate heat, radiation shielding placement, and durability design to withstand rocket launch shock.
A manufacturing plant in Woodinville, Washington
Starcloud currently has a staff of 25 and is building a roughly 100,000-square-foot production line in Woodinville, Washington. The area is reportedly near facilities where SpaceX and Amazon each build their own communications satellites.
Editor's take
This round underscores once again that the real bottleneck for orbital data center startups isn't satellite design but rocket seats. With Falcon 9 facing retirement in 2028 and Starship still unproven in operational use, companies like Starcloud find themselves lining up in advance for launch slots on rockets that don't even exist yet. The fact that one space startup has decided to build its own rocket entirely shows just how severe this bottleneck has become.
Anyone who has watched the ground-based data center industry will see a familiar shadow in today's orbital computing market. Just as securing power became the key constraint a few years ago — more so than the GPUs themselves — securing rockets to launch GPUs into space has now become the key constraint. This investment also reaffirms that NVIDIA is staying in the business of selling picks and shovels rather than jumping into the chatbot race itself. NVIDIA wanted real-world data on how the H100 holds up in orbit more than it wanted equity in Starcloud, and it paid $25 million for that.
For domestic cloud and satellite communications companies, this news may still sound like a distant story. But if space-grade GPUs enter testing in 2028 and Starship truly shortens its reuse cycle, orbital computing could emerge within five years not merely as a complement to ground-based data centers but as an alternative for specific workloads. Government agencies and satellite communications operators would do well to start examining data sovereignty and latency issues around orbital inference services now. Over the coming months, whether Starship successfully re-flies and whether the Vera Rubin Space-1 design is finalized will be the next milestones to watch in this industry.
Comments