Elon Musk's trillion-dollar gamble: Sending AI data centers to the sky, is it a genius idea or a sci-fi bubble?

Elon Musk is plotting a grand scheme worth $12.5 trillion.

Just last week, the tech mogul revealed an astonishing plan: to merge SpaceX with his loss-making AI startup xAI. This is not just a capital-level operation; its core logic lies in a seemingly sci-fi business bet—the future of AI is not on Earth, but in orbit.

Musk firmly believes that within the next three years, a massive satellite constellation powered by solar energy and cooled by the vacuum of space will become the cheapest way to generate AI computing power.

Is this yet another captivating story Musk is telling for an IPO, or is it an inevitable path for the AI industry?

The "Interstellar Computing Power" Race Among Giants

Musk is not the only tech giant gazing at the stars.

Although former xAI employees have criticized this idea as "completely unproven," Amazon founder Jeff Bezos has already entered the fray. He not only co-manages the mysterious AI startup "Project Prometheus," but also predicted last October that a "gigawatt-level space data center" will be built in the future.

Meanwhile, Google is also busy. They plan to launch the "Suncatcher Project" in early 2027 in collaboration with Planet, sending two prototype satellites equipped with TPU chips into low Earth orbit for "learning tasks."

Baiju Bhatt, founder of Aetherflux, pointed out succinctly: the AI boom has created a "once-in-a-century inflection point in energy usage." While terrestrial data centers are constrained by power supply and approval processes, space theoretically offers "unlimited energy."

The logic seems perfect: space has free and constant solar energy, and an environment close to absolute zero. As long as high-energy GPUs are sent up and data is transmitted back via Starlink, everything seems to fall into place.

Deutsche Bank's Ledger: The Cost Singularity of 2030

Capital markets are always the first to sniff out the scent of money. Deutsche Bank, in its latest report, provides economic support for this concept.

The current reality is harsh: deploying a 1-gigawatt capacity data center in space costs seven times that on the ground (approximately $114 billion vs. $16 billion).

However, Deutsche Bank analysts predict that this cost gap will narrow at an astonishing rate. The turning point may occur in 2032.

According to model calculations, if the following two conditions are met, the cost of space data centers will drop to a level "roughly comparable" to that on the ground:

1. Dramatic drop in launch costs: Relying on SpaceX's "Starship" and other reusable rocket technologies, the launch cost per kilogram needs to plummet from the current approximately $1,600 to $67.

2. Extreme optimization of hardware: The cost of a single satellite must drop below $2 million, with chips customized for the space environment. Once realized, this would mean that the energy and heat dissipation bottlenecks plaguing the ground computing power industry would be completely broken in space.

Engineer's Cold Water: "Thermodynamic Prison" and Radiation Nightmare

However, compared to the optimistic curves of financial analysts, frontline aerospace engineers and physicists have given a completely different view. They pointed out the most fatal physical flaws behind Musk's grand vision.

First, heat dissipation is a false proposition.

The public often mistakenly believes that the cold of space is suitable for heat dissipation. Wrong! Space is a vacuum environment, with no air convection, making fans useless.

"Space is a thermodynamic prison," one industry commentator vividly metaphorized. The enormous heat generated by millions of processors can only dissipate through one method: thermal radiation.

This means that a 5000-megawatt facility may require about 16 square kilometers of heat dissipation panels. This is equivalent to maintaining an extremely fragile heat dissipation structure the size of a small city. It's like trying to protect a huge sheet of aluminum foil on a battlefield filled with shrapnel.

Second, space is filled with radiation and debris, requiring expensive protection. Space is full of cosmic radiation and solar particles that continuously bombard electronic devices, causing data flips or damage.

To solve this problem, one either has to dress the servers in heavy lead "bulletproof vests" (but this would greatly increase launch weight), or use expensive and lower-performance "radiation-resistant chips."

Moreover, low Earth orbit has become a "dumping ground." Space debris traveling at 15,000 miles per hour, even if only the size of a marble, would be catastrophic if it struck a massive heat dissipation panel.

Who Holds the Key? Geopolitical Concerns

Setting aside technology and costs, there is a more serious question: who will control this infrastructure?

If SpaceX's Starship program succeeds and monopolizes cheap space transportation capabilities; if American tech giants control infinite energy and computing power in orbit...

Jermaine Gutierrez, a researcher at the European Space Policy Institute (ESPI), warns: "If you let all space-based infrastructure be dominated by Americans, that in itself is a risk."

According to the U.S. Cloud Act, American companies may be forced to cut off services anywhere in the world. For Europe, this means that after falling behind Amazon AWS and Google in cloud infrastructure, it may again become a vassal in this round of the "space computing power" competition.

Himanshu Tyagi, co-founder of Sentient, bluntly stated that the real risk is not a super AI out of control as in science fiction, but oligopolistic monopoly. "When a small group of people controls multiple chokepoints such as launches, communications, AI, robotics, and consumer platforms, you will face an oligopoly that is difficult to regulate and compete against."

The Starting Gun of the Computing Power Race Has Been Fired

Musk's predictions have always been radical, as industry veteran Steve Collar said: "Elon is always years ahead of reality, but that doesn't mean he's wrong."

Are space data centers the next trillion-dollar blue ocean, or a mirage under the laws of physics? The current economic calculations are primarily based on the extreme assumption that "launch costs drop to $67 per kilogram." For investors, this is a field that requires extreme caution. After all, rather than building expensive radiators in space, investing in more efficient nuclear or geothermal data centers on Earth at this moment might be the "smarter" money.

But in any case, the migration of computing power to space has already been set in motion