Orbital Plans First Space-Based Data Center by 2027

Orbital, a downtown Los Angeles space company, plans to deploy the first-ever space-based data center by April 2027, founder Euwyn Poon announced Tuesday.

The announcement lands at a moment when the AI industry’s energy appetite has gotten impossible to ignore. Southern California energy firms raised roughly $100 billion in 2026 to find cheaper, sustainable ways to power data centers straining the grid. That’s a staggering number. Communities from California to the Southeast have organized against ground-based data centers, arguing the facilities offer thin economic benefits while driving up energy costs and posing environmental risks for surrounding neighborhoods. Poon’s argument is straightforward: if the problem is Earth’s physical constraints, get off Earth.

Orbital is developing a constellation of satellites that house servers powered entirely by solar energy. Each refrigerator-sized unit will carry Nvidia Corp. graphics processing units and feature solar panels roughly half the size of a tennis court. Specialized radiators handle cooling, which on the ground requires enormous volumes of water and electricity. In orbit, the cold vacuum of space does most of that work for free.

It’s a genuinely different approach.

The company’s design confronts one of the thorniest technical problems immediately. Large language models on the ground run across hundreds of GPUs packed so tightly that data moves between them with essentially zero delay. You can’t replicate that in orbit, at least not yet. Orbital isn’t trying to. Instead, the company is focusing on AI inference, the process of delivering results based on information a model has already processed. Inference workloads are more tolerant of the slight delays that come with satellites orbiting a few hundred miles up, and they don’t require the same dense interconnections as training a model from scratch.

Franco Granda, a senior research analyst at PitchBook, said the approach is well-matched to current technical realities. “Small-scale orbital compute is realistic this decade, especially for defense and remote sensing workloads where latency is acceptable and data originates in orbit,” Granda said in an analyst note.

That framing matters for understanding who Orbital’s early customers probably are. Defense contractors and remote sensing operators already generate data in space and need to process it quickly without bouncing it back to a ground station first. That’s a real market, not a speculative one.

The broader aerospace context also helps explain why this idea is getting traction now. According to LA Business Journal, Los Angeles’ aerospace hub raised $7.13 billion in 2025, a record high for the industry according to PitchBook data. Launch costs have dropped sharply over the past decade as reusable rockets became standard. Putting hardware in low-Earth orbit doesn’t carry the price tag it once did, which changes the math on projects like Orbital’s significantly.

Poon doesn’t frame the venture as a niche defense play, though. He sees demand for computing as essentially unlimited, and he’s betting that orbital infrastructure will eventually serve commercial AI workloads as those workloads grow and diversify.

“I think that our demand for computing is just going to be insatiable,” Poon told reporters. “We tend to find a way to use up all the bandwidth. Humans are very clever at that.”

That’s not an unreasonable read of recent history. Every expansion of computing capacity, from personal computers to smartphones to cloud platforms, got absorbed faster than most analysts predicted. If Orbital can bring inference compute to orbit at competitive cost and acceptable latency, there’s a credible case that demand finds it.

The harder questions involve execution. Getting a constellation of satellites operational by April 2027 is an aggressive timeline. The National Aeronautics and Space Administration sent its first crew past low-Earth orbit in decades earlier this month, which has renewed public attention on what’s actually achievable in space, but institutional space programs and private satellite constellations operate on very different schedules and face different failure modes.

Thermal management in orbit, while cheaper than ground cooling in some respects, introduces its own engineering challenges. Satellites cycle from intense solar radiation to deep shadow every 90 minutes or so in low-Earth orbit, and that thermal cycling stresses hardware in ways that ground-based data centers don’t face. Nvidia’s GPUs weren’t designed with that environment in mind, and radiation hardening adds cost and complexity.

For Burbank and the broader San Fernando Valley, the Orbital announcement is worth watching because data center siting battles have become a recurring local political fight. Several proposed facilities in Los Angeles County have drawn opposition from neighborhood groups concerned about noise, water use, and the strain on already-stretched power infrastructure. An orbital alternative doesn’t eliminate ground-based facilities, but it represents a credible pressure valve if the technology proves out at scale.