Spin to Stars
Euwyn Poon sold 250,000 e-scooters to cities around the globe before walking away from Ford with a tidy exit. Then he ran eight Nvidia A100s out of a Santa Clara data center, serving open-weight AI models — and something clicked.
Here’s the real pivot: when Poon looked at what those chips were truly hungry for — raw, uninterrupted compute — he saw rooftops weren’t the answer. He needed orbit.
That’s how Orbital was born. Not from aerospace experience, but from a founder who noticed space offers exactly what AI can’t live without: infinite uptime and near-zero thermal constraints. Today, Orbital is a $5 million seed-stage startup, quietly building the infrastructure layer for space-based AI inference.
The timing couldn’t be starker. SpaceX is about to go public, and the IPO window opens what venture firms now call “the capital-intensive phase” of commercial space — a phrase that used to sound like snake oil. Now? It sounds inevitable.
Orbital’s Step-by-Step Ascent
Orbital didn’t leap straight to 10,000 satellites. It came via a16z’s Speedrun accelerator in May 2026, where Poon pitched three ideas before landing on orbital compute. Investors包括 Basis Set, Human Element, Wayfinder, Antler, and a dozen others leaned in. They didn’t ask for space credentials — they asked for scaling muscle.
Poon’s resume answered that. Spin, his first startup, didn’t just deploy scooters — it scaled them across 100 cities before Ford bought the company. That kind of operational grit matters when your deployment timeline spans years, not quarters.
The near-term plan? Fly a Blackwell GPU on a partner’s satellite this year, testing radiation shielding and thermal management in the wild. If it holds up, Orbital bets its first production craft — loaded with Nvidia’s Space-1 Vera Rubin-class GPUs — launches in 2028.
That demo flight is the critical hinge. Until it happens, Orbital exists as a theory with good math and strong investors. Once it’s proven, the real work starts: deploying enough satellites to make 100 kW each, cumulatively reaching a distributed gigawatt of on-orbit compute.
Why SpaceX Is Orbital’s Lifeline
Here’s the hard truth: Falcon 9 pricing still makes Orbital’s business case look like a PowerPoint joke. Poon told TechCrunch bluntly: “We will get to full scale when Starship comes online.”
Until then, he’s building around the assumption that orbital access must become dramatically cheaper — and that’ll only happen when SpaceX stops treating Starship as a prototype and starts offering it to commercial customers at scale.
Other startups aren’t waiting. Cowboy Space decided to build its own rockets, while Blue Origin plans to use New Glenn for data-center payloads. Starcloud is already flying GPUs and aiming to generate revenue before Starship arrives.
Orbital’s stance is simpler: let the best rocket win — and when Starship does, Orbital will be ready to launch thousands of satellites in rapid succession.
The Competitive Cleverness of Space Data Centers
Poon insists there’s room for many players. “There’s so many lanes for companies in our space to pursue,” he said, listing workload specialization, thermal design trade-offs, and satellite architecture diversity as differentiators.
Starcloud already has a GPU in orbit. Cowboy Space is engineering its own launch. Orbital bets on scale — 10,000 satellites at 100 kW apiece — and long-duration inference workloads. That’s the edge: not just compute, but persistent compute that never sleeps and rarely adjusts its orbit.
Poon compares Orbital’s potential role to the data-center colocation industry — platforms that lease space, power, and cooling to hyperscalers. Orbital’s satellites could host multiple customers’ inference tasks simultaneously, sharing the power budget and thermal load across a constellation.
The first revenue stream? Piece-wise inference jobs on early satellites — small enough to run before the full constellation launches. That’s how you bridge from demo flight to sustainable business.
AI Demand Is the Real Launcher
What really changed in the last five years? Not the rockets — not yet. It’s the demand signal.
Andrew Chen of a16z put it simply: “This kind of thing would have sounded crazy 10 years ago when we were all building mobile apps.”
Today, AI compute demand is so insatiable that investors are willing to back decade-long projects with $5 billion price tags. Venture capital has shifted from optimizing for speed of iteration to optimizing for scale of compute.
That shift is why Orbital’s founding story starts with a tech founder, not an aerospace engineer. The skills that mattered — scaling operations, negotiating with chipmakers, managing hardware deployments — were the same ones Poon used on Santa Clara’s rooftops and later Ford’s assembly lines.
Orbital’s real product isn’t satellites. It’s distributed inference infrastructure. The hardware just happens to live in low Earth orbit.
What Comes Next — and Why It Matters
The IPO timing is no accident. SpaceX’s market debut signals to the world that space is becoming a tradable, predictable business — not just a science project or a geopolitical showpiece.
Orbital’s next milestone isn’t launch day. It’s proving that space-based inference can be cheaper, faster, and more reliable than terrestrial equivalents for specific workloads. That means benchmarking against the largest cloud AI services — and beating them where it counts: latency, energy efficiency, and uptime.
If Orbital wins this race, it won’t just deliver compute from orbit. It’ll redefine where the boundary of AI infrastructure lies — and open a new category: orbital cloud services.
For now, the team’s tiny Los Angeles office (dozen-ish engineers with backgrounds at Amazon LEO, SpaceX, and Northrop Grumman) is building the first flight hardware. They know Starship isn’t ready yet. But they’re building for the day after it is.
That’s Orbital’s real bet: not on SpaceX alone, but on the ecosystem it enabled — a future where space isn’t an exception, but a rack of servers at 500 km altitude.
Orbital’s Step-by-Step Ascent
Orbital didn’t leap straight to 10,000 satellites. It came via a16z’s Speedrun accelerator in May 2026, where Poon pitched three ideas before landing on orbital compute. Investors包括 Basis Set, Human Element, Wayfinder, Antler, and a dozen others leaned in. They didn’t ask for space credentials — they asked for scaling muscle.
Poon’s resume answered that. Spin, his first startup, didn’t just deploy scooters — it scaled them across 100 cities before Ford bought the company. That kind of operational grit matters when your deployment timeline spans years, not quarters.
The near-term plan? Fly a Blackwell GPU on a partner’s satellite this year, testing radiation shielding and thermal management in the wild. If it holds up, Orbital bets its first production craft — loaded with Nvidia’s Space-1 Vera Rubin-class GPUs — launches in 2028.
That demo flight is the critical hinge. Until it happens, Orbital exists as a theory with good math and strong investors. Once it’s proven, the real work starts: deploying enough satellites to make 100 kW each, cumulatively reaching a distributed gigawatt of on-orbit compute.
Why SpaceX Is Orbital’s Lifeline
Here’s the hard truth: Falcon 9 pricing still makes Orbital’s business case look like a PowerPoint joke. Poon told TechCrunch bluntly: “We will get to full scale when Starship comes online.”
Until then, he’s building around the assumption that orbital access must become dramatically cheaper — and that’ll only happen when SpaceX stops treating Starship as a prototype and starts offering it to commercial customers at scale.
Other startups aren’t waiting. Cowboy Space decided to build its own rockets, while Blue Origin plans to use New Glenn for data-center payloads. Starcloud is already flying GPUs and aiming to generate revenue before Starship arrives.
Orbital’s stance is simpler: let the best rocket win — and when Starship does, Orbital will be ready to launch thousands of satellites in rapid succession.
The Competitive Cleverness of Space Data Centers
Poon insists there’s room for many players. “There’s so many lanes for companies in our space to pursue,” he said, listing workload specialization, thermal design trade-offs, and satellite architecture diversity as differentiators.
Starcloud already has a GPU in orbit. Cowboy Space is engineering its own launch. Orbital bets on scale — 10,000 satellites at 100 kW apiece — and long-duration inference workloads. That’s the edge: not just compute, but persistent compute that never sleeps and rarely adjusts its orbit.
The first revenue stream? Piece-wise inference jobs on early satellites — small enough to run before the full constellation launches. That’s how you bridge from demo flight to sustainable business.
AI Demand Is the Real Launcher
What really changed in the last five years? Not the rockets — not yet. It’s the demand signal.
Andrew Chen of a16z put it simply: “This kind of thing would have sounded crazy 10 years ago when we were all building mobile apps.”
Today, AI compute demand is so insatiable that investors are willing to back decade-long projects with $5 billion price tags. Venture capital has shifted from optimizing for speed of iteration to optimizing for scale of compute.
That shift is why Orbital’s founding story starts with a tech founder, not an aerospace engineer. The skills that mattered — scaling operations, negotiating with chipmakers, managing hardware deployments — were the same ones Poon used on Santa Clara’s rooftops and later Ford’s assembly lines.
Orbital’s real product isn’t satellites. It’s distributed inference infrastructure. The hardware just happens to live in low Earth orbit.
What Comes Next — and Why It Matters
The IPO timing is no accident. SpaceX’s market debut signals to the world that space is becoming a tradable, predictable business — not just a science project or a geopolitical showpiece.
Orbital’s next milestone isn’t launch day. It’s proving that space-based inference can be cheaper, faster, and more reliable than terrestrial equivalents for specific workloads. That means benchmarking against the largest cloud AI services — and beating them where it counts: latency, energy efficiency, and uptime.
If Orbital wins this race, it won’t just deliver compute from orbit. It’ll redefine where the boundary of AI infrastructure lies — and open a new category: orbital cloud services.
For now, the team’s tiny Los Angeles office (dozen-ish engineers with backgrounds at Amazon LEO, SpaceX, and Northrop Grumman) is building the first flight hardware. They know Starship isn’t ready yet. But they’re building for the day after it is.
That’s Orbital’s real bet: not on SpaceX alone, but on the ecosystem it enabled — a future where space isn’t an exception, but a rack of servers at 500 km altitude.