資金調達が軌道上のAIデータセンターの基盤を築く

2 Min ReadIllustration of space-based data centerOrbitalLos Angeles-based startup Orbital said it has received funding from A16z Speedrun, an accelerator program run by venture capital firm Andreessen Horowitz, that will move it closer to its vision of operating AI data centers in space.It is not clear what the extent of A16z Speedrun's investment is, but the group typically invests about $1 million in startups. The funding will enable Orbital to launch its first test mission in April of next year, according to the company, which is also opening an R&D center.Orbital's business plan is to launch data centers in space because the power required to run them on Earth is excessive and may ultimately prove prohibitive.Key to Orbital's plan is a constellation of satellites it is developing to operate in low Earth orbit, each of which will house a cluster of Nvidia-powered servers and feature solar panels. Solar power is available  24 hours a day -- because there is no weather, nighttime or dependence on the power grid -- while cooling is achieved by radiating heat into space, where the temperature is near absolute zero.Related:The Real AI Shift Isn’t New Models. It’s Control.“AI progress is being constrained by the grid. Data center economics are dominated by electricity and cooling, and both are getting harder," Orbital CEO and founder Euwyn Poon said in a statement. "In orbit, solar power is continuous, and cooling is fundamentally different. Orbital is building compute infrastructure that removes the energy ceiling and scales with AI's potential.” Moving into space would address growing concerns about the extensive energy consumption required by AI. This week, the Stanford Index Report found that AI systems now consume the same amount of power as a smaller European nation such as Austria or Switzerland, while in the U.S., pushback is growing against the data centers required for AI processing, with the state of Maine, for example, even considering banning them.Orbital's proposition is a refrigerator-sized satellite with a small number of GPUs, a solar array and a separate structure to diffuse heat. As Orbital points out, while training large AI models requires thousands of closely connected GPUs communicating at near-zero latency, this is not the case for inference, in which requests are handled separately. Inference is what Orbital is focused on, and why it is pursuing space as a viable option.A16z’s backing will allow Orbital to assess its vision within a year, the company said.About the AuthorContributing WriterGraham Hope has worked in automotive journalism in the U.K. for 26 years, including spells as editor of leading consumer news website and weekly Auto Express and respected buying guide CarBuyer.