Here’s one indicator to track SpaceX’s IPO later this week. The company has revolutionized the way the venture industry views space as a long-term, capital-intensive space, making it possible for talented founders with no space experience to fund space data center companies.
Orbital, a new company launched in May with a $5 million seed round from a16z’s startup accelerator program Speedrun, is the latest company to commit to doing inference in space, similar to how Starship regularly flies. Other investors include Basis Set, Human Element, Wayfinder, Antler, Anti Fund, Ascent, Rubik, Zero Knowledge Ventures, LYVC, Feld Ventures, New Legacy, FNDR, UpHonest, and Asterisk.
Founder and CEO Ewin Poon previously founded electric scooter company Spin in 2017, which he sold to Ford a year later to become part of the auto giant. When the new company was ready to launch, a16z’s Speedlan was an active participant, said partner Andrew Chen. Poon told TechCrunch that he considered several ideas before landing at the space data center.
You’re familiar with the pitch. The demand for AI computing is insatiable, but global adoption has been slow. Why not go to space in search of endless sunlight and an overhaul of the limited environment? The main problem is the brutal economics of launching objects into orbit, which is why we are currently unable to complete the business case.
Orbital, like many of its competitors, is betting that SpaceX will develop the Starship rocket and deliver it to commercial customers. “Once Starship comes online, we’ll really start working on it,” Poon explained. Given the price of the current state-of-the-art Falcon 9, “this is not economically viable.”
So far, Poon and his company, which includes about a dozen Los Angeles residents with experience at Amazon LEO, SpaceX and Northrop Grumman, are working toward demonstration flights that will fly Nvidia Blackwell chips on partner satellites and test Orbital’s radiation shielding and thermal management technology. The company hopes to launch its first data processing spacecraft powered by Nvidia’s Space-1 Vera Rubin-class GPUs in 2028.
At that point, the company hopes to begin segment-by-segment inference work, which will allow it to generate revenue every time a satellite is launched. This is a similar path to that of rival data center startup Starcloud. Starcloud already has GPUs in orbit and plans to launch several more to generate revenue until Starship can deploy a complete constellation.
Orbital’s goal is to deploy 10,000 satellites providing distributed gigawatts of computing power, with each satellite providing 100 kW of power. For comparison, Elon Musk said SpaceX expects its AI satellites to generate up to 150 kilowatts of power, and StarCloud expects to deploy larger spacecraft rated at 200 kilowatts to run the chips.
Some companies can’t wait for Starship. Cowboy Space Company, another space data center startup backed by a16z, recently decided to start building its own rockets. Jeff Bezos’ space company Blue Origin also announced plans to launch a data center into space using a New Glenn rocket.
Poon believes that the breadth of demand for AI will enable many companies to succeed. “There are many avenues in our universe for companies to pursue,” he told TechCrunch. He then listed a variety of options, including different AI workloads, designs, and companies pursuing what a space data center might look like.
Chen said Poon’s experience in scaling up a company with 250,000 scooters in 100 cities shows he can handle the difficult task of building an aerospace company. In the long run, such a project could take 10 years and more than $5 billion, but venture companies are accustomed to such schedules, Chen said.
“Ten years ago, when we were developing mobile apps, this would have sounded crazy,” he says. “Starting in 2026 will allow us to take advantage of all the energy and excitement that is happening in the capital markets.”
Poon found his way into the space data center business through a circuitous route. After leaving Ford, he purchased an Nvidia A100 at Lark and co-located it in his Santa Clara data center, offering an open weight model. This first-hand experience convinced him of the value of delivering computing in the age of AI.
All that’s left is to put thousands of GPUs in space.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
