How an e-scooter founder raised $5 million to build space data centers

Overview

Orbital, a Los Angeles startup founded by former Spin e-scooter CEO Euwyn Poon, raised a $5 million seed round to build AI data centers in space. The round was led by venture firm a16z through its Speedrun accelerator. Orbital wants to deploy 10,000 satellites that together deliver a gigawatt of computing power for AI inference, with each satellite supplying about 100 kW. The company aims to reach full scale once SpaceX Starship flies on a regular schedule.

Key Takeaways

Orbital raised $5 million in an oversubscribed seed round led by a16z through its Speedrun accelerator program, with a long list of additional backers.
The company plans a constellation of 10,000 satellites delivering a combined gigawatt of compute for AI inference, with each satellite carrying about 100 kW of power.
Founder Euwyn Poon previously scaled e-scooter company Spin to roughly 250,000 scooters across 100 cities before selling it to Ford in 2018.
Orbital aims to launch its first data-processing spacecraft in 2028 using Nvidia GPUs, after an earlier demo flight to test radiation shielding and thermal control.
The business model depends on SpaceX Starship flying regularly, because current launch costs make space-based compute too expensive to scale.
Orbital joins rivals Starcloud and a planned SpaceX effort in the race to move AI inference off the planet.

Stats & Key Facts

#$5 million seed round raised, led by a16z's Speedrun accelerator
#10,000 satellites planned for the full constellation
#1 gigawatt of distributed computing power targeted across the fleet
#100 kW of power capacity per individual satellite
#250,000 scooters built at the founder's prior company Spin, across 100 cities
#About a dozen employees based in Los Angeles

Orbital's $5 Million Seed Round Led by a16z Speedrun

The funding came from a wide group of investors centered on a well-known venture firm.

›The round totaled $5 million and was led by a16z through its Speedrun accelerator program.
›Reports describe the round as oversubscribed, meaning investors wanted to put in more than the company accepted.
›Additional backers included Basis Set, Human Element, Wayfinder, Antler, Anti Fund, Ascent, Rubik, Zero Knowledge Ventures, LYVC, Feld Ventures, New Legacy, FNDR, UpHonest, and Asterisk.

From 250,000 Scooters at Spin to Satellites in Orbit

The founder built his reputation in a very different kind of hardware business.

Euwyn Poon founded the e-scooter company Spin in 2017 and sold it to Ford a year later, then joined the automaker. At Spin he scaled the operation to about 250,000 scooters across 100 cities, which gave him experience running a capital-heavy hardware business.

After leaving Ford, Poon got into AI computing in a hands-on way. He bought an Nvidia A100 chip, placed it in a Santa Clara data center, and used it to serve open-weight AI models. That experience pushed him toward the idea of running compute in orbit.

A 10,000-Satellite Constellation for AI Inference

The core plan is to spread computing power across a large fleet of small spacecraft.

›The target is 10,000 satellites that together provide one gigawatt of distributed computing power.
›Each satellite would carry about 100 kW of power capacity.
›The fleet is designed for AI inference, the step where a trained model answers requests rather than the heavier training stage.

Why Put AI Compute in Space at All

The appeal comes down to free power and free cooling.

In orbit, satellites get near-constant solar power and natural cooling from the cold of space. That sidesteps the land, water, and electricity strain that large ground-based data centers place on local power grids.

Demand for AI computing has put pressure on terrestrial sites, where utilities and communities increasingly push back on water use and electricity draw. Moving inference off the planet is one proposed way around those limits, though it remains unproven at scale.

Launch Timeline, Nvidia GPUs, and the Demo Flight

The company has a staged plan that starts small and builds toward full deployment.

›Orbital is working toward a demo flight that puts an Nvidia Blackwell chip on a partner's satellite to test its radiation shielding and thermal management.
›In 2028 it hopes to launch its first data-processing spacecraft using Nvidia's Space-1 Vera Rubin-class GPUs.
›The team plans to do paid inference work along the way to earn revenue with each launch.
›The roughly dozen-person Los Angeles team includes people who worked at Amazon's LEO satellite program, SpaceX, and Northrop Grumman.

The Starship Dependency and the Competition

The economics hinge on cheaper launches and a crowded field of rivals.

Orbital says it will reach full scale only once SpaceX Starship flies on a regular schedule. Current Falcon 9 launch costs make the full business model too expensive, so the company is betting on a future drop in the price of reaching orbit.

Orbital is not alone. Rival Starcloud already has a GPU in orbit and plans to launch more, and SpaceX itself has signaled interest in AI satellites that Elon Musk said could produce up to 150 kW. Poon framed the field as wide open, noting there are many lanes for companies in the space to pursue.

Frequently Asked Questions

What is Orbital trying to build?

Orbital wants to put AI computing in orbit instead of in ground-based buildings. It plans a constellation of 10,000 satellites that together deliver about a gigawatt of computing power for AI inference.

Who founded Orbital?

Orbital was founded by Euwyn Poon, who earlier started the e-scooter company Spin in 2017 and sold it to Ford in 2018. He later worked hands-on with AI hardware before starting Orbital.

How much money did Orbital raise and from whom?

Orbital raised a $5 million seed round led by venture firm a16z through its Speedrun accelerator program. A long list of additional investors took part, including Basis Set, Human Element, Antler, and Anti Fund.

Why would running AI compute in space make sense?

Satellites in orbit get near-constant solar power and natural cooling from space. That avoids the land, water, and electricity demands that large ground data centers place on local power grids.

When will Orbital launch its first spacecraft?

Orbital aims to launch its first data-processing spacecraft in 2028 using Nvidia GPUs. Before that it plans a demo flight on a partner's satellite to test radiation shielding and thermal management.

Orbital is an early-stage bet that AI inference belongs in orbit, where solar power and cold space replace strained grids and cooling water. Whether the idea works depends heavily on SpaceX Starship lowering launch costs enough to make a 10,000-satellite fleet affordable.