太空数据中心距落地仍有关键技术与经济门槛

4 Min Read

SANTA CLARA, Calif. – Orbital data centers are edging closer to reality, but significant technical and economic obstacles remain, according to industry and government leaders who discussed the future of computing in space at the AI Infra Summit, held Sept. 15-17.

Thermal, Bandwidth, and Launch Hurdles

Sumeet Singh, chief data and AI officer at Viasat, a satellite communications company, identified three primary challenges. The first is thermal management – the ability to keep orbital systems cool. In the vacuum of space, heat can only be shed through radiators, which must be large enough to dissipate substantial thermal loads. “With size comes weight. Weight is a huge currency when you think about orbital data centers,” he said during a panel discussion on Thursday, Sept. 17.

Singh also pointed to limited downlink and uplink capacity – the ability to move data to and from orbit – and the cost of launching the data centers into low Earth orbit (LEO). He said today’s costs of $2,500 to $3,000 per kilogram would need to drop to roughly $200 to $500 per kilogram for orbital data centers to make financial sense.

Joe Yaffe, COO and chief legal officer of Cowboy Space Corp., which is building orbital data centers and aims to launch them, agreed that those obstacles are real. He added that reliance on a single dominant satellite launch provider has become a bottleneck in its own right.

Yaffe said Cowboy Space is tackling the bottleneck by developing its own launch vehicle and using the rocket’s second stage as the data center itself. To address bandwidth, the company plans to use optical links. “We’re believers that data transmission needs to take place optically, uplink and downlink, in order to move what will be increasingly larger amounts of data,” Yaffe said.

Power is another constraint. Camille Bergin, chief marketing officer of Star Catcher, an in-space energy utility, said the company is working on laser-based power beaming to augment existing solar arrays, potentially delivering up to 10 times more power to orbiting systems.

Customer Readiness and the Road to Scale

From the customer standpoint, the US Air Force is not yet ready to adopt orbital compute at scale, said Deepak Sachdeva, CIO of the US Air Force. His priorities are operational effectiveness, real-time or almost real-time latency, resilience, and security. By those measures, he said, the technology still falls short. “From a consumer-to-consumer level, I think it’s ready. But from an operational need, which is global in its nature, I think there is still some work to be done,” he said. “The best approach, in my view, is to do some sort of proof of concept.”

Singh, whose company, Viasat, wants to participate in the emerging orbital data center ecosystem, said the industry must prove itself incrementally – both technically and commercially. He cited StarCloud’s November 2025 launch of a satellite featuring an Nvidia H100 GPU as a promising sign, while noting a single-GPU demo is a far cry from megawatt- and gigawatt-class facilities. “We suddenly talk about megawatt- and gigawatt-class data centers. I think we need intermediate proof points … something in between where you have 100- to 500-kW-class compute,” he said.

Other key milestones, Singh added, include ensuring sufficient launch capacity to get these systems into orbit and demonstrating that customers will actually buy compute in space.

Bergin agreed on the need to “crawl, walk, run,” describing how her company first proved its technology on Earth before moving to space-based demonstrations. “There are a lot of people in this industry who have a tendency to just say, ‘This is the thing that we’re going to build, and it’ll just happen,’” she said. “I think from the orbital compute perspective, what are those incremental steps? How do we get from operating at 5% to 10% on a single H100 as a prototype to a megawatt-class, gigawatt-class data center? … We can’t just design something in space because we think it’s cool and it’s the next frontier. There has to be a real need and a real business model.”

Bergin also argued that orbital data centers could address community pushback against terrestrial facilities. “All of us are super-familiar with the fact that people don’t like these data centers. They don’t want them in their communities. They are super-problematic. … So space [is] offering a solution if we can get there.”

Despite the challenges, Singh remains confident. “This isn’t a question of if it would happen. I think this is more of a question of the step function. When would this happen?” he said.

About the Author

Chips and Hardware Writer

Wylie Wong is a journalist and freelance writer specializing in technology, business and sports. He previously worked at CNET, Computerworld and CRN and loves covering and learning about the advances and ever-changing dynamics of the technology industry. On the sports front, Wylie is co-author of Giants: Where Have You Gone, a where-are-they-now book on former San Francisco Giants. He previously launched and wrote a Giants blog for the San Jose Mercury News, and in recent years, has enjoyed writing about the intersection of technology and sports.