Elon Musk’s SpaceX (NASDAQ:SPCX) pitched orbital data centers as a centerpiece of the biggest IPO in Wall Street history, but critics say the physics and the costs don’t add up.

A chip founder backed by Bill Gates disagrees.

Patrick Bowen is the co-founder and CEO of Neurophos, a photonic chip startup that raised a $110 million Series A in January led by Gates Frontier, with participation from Microsoft’s venture fund M12.

“There is no single engineering obstacle that can’t be overcome for this application,” Bowen said, though he acknowledged that maintenance, serviceability, reliability, and lifetime costs remain open questions compared to running data centers on Earth.

Geopolitical analyst Peter Zeihan recently called the orbital data center concept “so many levels of dumb,” arguing that cooling chips in a vacuum would require impossibly large radiators, which he estimated could cost $250 billion.

Where The Critics Are Right

Bowen conceded the thermal problem is real. Space is a vacuum, so waste heat can only leave through radiators, and at gigawatt scale those radiators become enormous.

But he told Benzinga that feasibility comes down to engineering and economics rather than any hard physical limit.

The variables that matter, he said, are radiator efficiency, operating temperature and launch costs.

Looking further out, Bowen floated the idea that the launch bottleneck itself could one day be bypassed by manufacturing giant radiators and solar arrays directly on the Moon rather than lifting them from Earth.

Zeihan has also argued that high-end chips can’t survive radiation in orbit, estimating the necessary shielding would add some 20 million pounds to any facility.

Google’s (NASDAQ:GOOGL) testing points the other way: under Project Suncatcher, the company reported that its Trillium AI chips survived radiation doses equivalent to a five-year low Earth orbit mission without significent damage.

Why Chip Efficiency Changes The Math

Bowen’s argument rests on a chain of consequences.

More efficient chips draw less power, less power means less waste heat, less heat means smaller radiators, and smaller radiators mean less mass to launch.

Photonic processors may also run at higher temperatures than conventional GPUs, he said, and radiators shed heat more efficiently the hotter they run. That attacks Zeihan’s radiator math from both ends.

Neurophos says its first chip, the Tulkas T100, could deliver roughly 10 times the compute of Nvidia’s (NASDAQ:NVDA) latest Rubin GPUs at similar power, though the numbers remain theoretical and the chip isn’t expected to ship before 2028.

The company is already exploring orbital applications through a partnership with Eclipse Space, a startup staffed largely by veterans of SpaceX’s Starlink division.

Everything Hinges On Starship

Bowen said the timeline becomes practical only when launch costs fall dramatically, which he expects once Starship reaches full operational capacity.

“We believe that Elon Musk is in a unique position to make compute in orbit viable,” he said.

Traders are less convinced. Kalshi traders give a 1-megawatt orbital data center just a 14% chance of going live before January 2029, rising to 38% before 2035.

SPCX priced at $135 in June in the largest IPO in Wall Street history, but the stock has since slid to around $115.

Image: Shutterstock

Kalshi and Benzinga have an existing data collaboration agreement.