It’s plain to see that Elon Musk’s ambition of putting data centers in space is a daring and risky undertaking. 

Further underscoring the challenges, experts tell Reuters that a previous failed attempt at taking data centers off solid ground has alarming parallels that could spell doom for Musk’s plan for SpaceX.

In 2015, Microsoft deployed “Project Natick,” a cutting edge underwater data center off the coast of Scotland. Resembling the size and shape of a semi truck’s fuel tanker, it was designed to use seawater to cool itself and be largely self-sufficient once anchored to the seabed. The idea was full of promise: cooling a data center is one of its most costly aspects; now it was accomplishing it for free. It was also supported by wind power, providing an aspect of sustainability.

Flash forward to the present, however, and the data centers that are popping up everywhere are amid the AI boom are most decidedly not being built in the ocean. Sources told Reuters that the project was figuratively sunk by lack of client demand and unviable economics for reasons that could also plague Musk’s orbital facilities.

“These problems are likely to be more ​severe in space than under the sea,” Roy Chua, founder of industry research firm AvidThink, told Reuters. 

Critically, both projects rely on modular units that are expensive to deploy, and once operational, can’t be upgraded or even repaired. Potential customers favored sticking to terrestrial facilities because they could be brought online quicker and be upgraded with the latest hardware — a more crucial capability than ever, because AI chips are constantly improving.

Once, or if, Musk deploys his orbital data centers, they’ll be “locked-for-life.” A new generation of AI hardware — perhaps one optimized for another type of AI architecture that becomes the cutting edge, as many in the industry believe large language models are an eventual dead end — could obviate Musk’s expensive satellites.

Experts have also been incredulous at Musk’s proposed size for each of these data center satellites, which according to company graphics will dwarf the International Space Station.

All that’s before we even begin to look at the exorbitant costs of getting these data centers into space at scale. Reminder: Musk wants to deploy one million of the satellites. Ars Technica editor Eric Berger estimated the barebones cost of doing that to be at least $1 trillion. Analysts at equity research group Moffett Nathanson, in a note cited by Reuters, said the cost would be trillions-plural.

But what are mere trillions in this day and age? If SpaceX somehow gets the money for all this — perhaps with a little help from its forthcoming IPO — it would quickly be overwhelmed by the sheer number of space launches needed to pull this off. According to Moffett Nathanson’s estimates, SpaceX would have to launch its Starship rocket 3,000 times per year, or eight times per day. (Last year, the company launched 167 rockets total.)

Starship is designed to be reusable and carry far more massive payloads to orbit than existing rockets, making it the most cost-efficient vehicle for the job — in theory. It’s years behind schedule and has exploded in many of its 11 flight tests, none of which have reached Earth’s orbit yet.

If space data centers have a future, it’ll be as a niche complement to conventional ones, perhaps for military applications or providing computing power to space stations. That’s nice, but a far cry from Musk’s promise that space data centers will be the future.

“I strongly believe that there’ll be no ​way in the foreseeable future that ⁠space‑based data centers can replace ground data centers,” Rousseau, a research director at consulting firm Analysys Mason, told Reuters.

More on data centers: OpenAI’s Obsession With Data Centers Is Running Into Trouble