Our Take
A founder with zero space experience can now fund a space startup because venture capital has shifted its risk tolerance on long-term, capital-intensive bets—not because space data centers are close to economical.
Why it matters
The shift reflects how completely SpaceX's progress has reset venture expectations for aerospace timelines and costs. For practitioners building compute infrastructure, this signals that orbital alternatives are moving from thought experiment to multi-year deployment planning.
Do this week
Infrastructure teams: stress-test your 2028+ capacity plans against a scenario where space-based inference becomes a marginal cost option, so you can decide whether to reserve Starship allocation now.
A founder with no space experience just raised $5 million for space data centers
Euwyn Poon founded Spin, an e-scooter company, in 2017 and sold it to Ford a year later after deploying 250,000 scooters across 100 cities. He then spent time at Ford and independently ran GPUs in a Santa Clara colocation. In May 2026, he emerged from a16z's Speedrun accelerator with Orbital, a startup that plans to deploy 10,000 satellites providing distributed gigawatt-scale AI inference compute. The seed round totaled $5 million (per TechCrunch), with 15 named investors including a16z, Basis Set, Human Element, Wayfinder, and Feld Ventures.
Orbital is one of at least three well-funded entrants in the space data center category. Starcloud already has a GPU in orbit and plans incremental launches to generate revenue before larger constellation deployment. Cowboy Space Company, also backed by a16z, recently decided to build its own rockets rather than wait for Starship. Blue Origin announced plans to launch data centers using its New Glenn vehicle.
Poon's immediate plan is modest: a demo flight in the near term to test radiation shielding and thermal management, flying an Nvidia Blackwell chip on a partner's satellite. In 2028, Orbital aims to launch its first data-processing spacecraft using Nvidia's Space-1 Vera Rubin-class GPUs, then generate revenue through piece-wise inference work (small-scale inference jobs per satellite launch).
The long-term target is 10,000 satellites, each delivering 100 kilowatts of power. For context, SpaceX expects its own AI satellites to produce up to 150 kW (per Elon Musk's statements), and Starcloud targets 200 kW-rated spacecraft.
The economics don't work yet, but venture thinks SpaceX will fix them
Poon was explicit about the constraint: Falcon 9 launch costs make the business case "not economically feasible." Orbital's entire timeline depends on Starship reaching routine commercial operation and driving launch costs low enough that on-orbit compute becomes marginal-cost competitive with ground-based data centers.
a16z partner Andrew Chen framed this as a shift in venture risk tolerance, not a technical breakthrough. "This kind of thing would have sounded crazy 10 years ago," Chen told TechCrunch. "Starting it in 2026 just lets you tap into all the energy and excitement that's happening in the capital markets." He also noted that venture firms are now comfortable with aerospace timelines spanning a decade and budgets exceeding $5 billion.
The story is not that space data centers work today. It is that a founder with proven ability to scale hardware operations (250,000 units across 100 cities) can now raise capital for a venture that requires Starship to succeed and won't generate material revenue until 2028 at the earliest. That threshold shift is real. It reflects SpaceX's credibility, not Orbital's.
Treat 2028 as a fork in your capacity plan, not a certainty
If Starship reaches $10 million per launch (an industry estimate, not a guarantee), space-based inference becomes worth modeling. If it doesn't, 10,000 orbital data centers stay theoretical. Infrastructure teams should begin war-gaming both scenarios now. That means understanding which workloads benefit most from latency-free satellite proximity (edge inference, real-time processing) versus those that don't (batch model training). It also means tracking Starship's actual flight cadence and payload costs as they emerge, not industry speculation.
The other risk: regulatory and orbital debris policy. Deploying 10,000 satellites introduces meaningful crowding. If governments move to restrict commercial megaconstellations or require active deorbiting, the entire math changes. Monitor FCC filings and international space policy committees through 2027.