Amazon Data Center Attacked: Can Moving to Space Avoid Missiles?

On April 3, multiple overseas media outlets reported that an Iranian military missile attack hit the data center in Bahrain operated by Amazon again. This is the second time this year.

No matter how sturdily the data center on the ground is built, it can’t stop missiles. That suddenly makes a topic that was once purely in the realm of science fiction feel real: if data centers are moved to space, could it reduce this kind of ground-based threat to some extent?

In fact, quite a few technology companies are already doing it.

Google CEO Sundar Pichai claims that within ten years, space data centers will become “the new normal.” The company is planning to launch two prototype satellites in 2027 to validate key technologies for in-orbit computing performance, high-speed communication between satellites, and the system’s long-term stability.

Bezos’s Blue Origin has already filed an application with regulators in March this year, planning to launch 50k solar-powered data-center satellites.

There’s also another U.S. space computing startup, Starcloud, which completed a key validation in December 2025: for the first time, it trained a small AI model on an orbital satellite, and for the first time demonstrated that AI training can be done in space.

A race centered on “orbital computing power” is heating up.

Supporters argue this is the only way out of AI’s power-hungry problem. Opponents say the math simply doesn’t add up. Others have even thought of a tougher issue: moving computing power into orbit means exposing the most critical digital infrastructure to predictable, targetable space coordinates. If a conflict escalates, will these high-value targets become targets for the other side?

A more critical question is: when your computing power floats in an orbit that the other side’s missiles can reach, who will foot the bill for its security?

01

Google’s “Project Loon Catchers” plan: small satellites, laser interconnects, TPU in the sky

Google’s “Project Loon Catchers” plan is the most detailed one among the currently disclosed information.

Google’s approach is a fleet of small satellites equipped with Google’s in-house TPU chips, interconnected via free-space optical links to form a constellation. The orbit is chosen as a dawn-dusk, sun-synchronous low Earth orbit, so the satellites can be in sunlight almost all the time.

The biggest technical challenge is communication between satellites. In ground data centers, data transfer between chips runs over fiber optics—high bandwidth and low latency. But in space, satellites have to rely on wireless connections. Google’s calculation is that to achieve data-center-level performance, inter-satellite links need to support transmission rates of dozens of terabits per second.

They have already achieved bidirectional 1.6 terabits-per-second transmission rates in the lab using a pair of transceivers. The key is that the satellites need to fly very close—within a few kilometers, even just a few hundred meters.

Google’s orbital dynamics model shows that at an altitude of 650 kilometers, a constellation of 81 satellites results in a distance between adjacent satellites of only about 100 to 200 meters. Fortunately, to maintain such a formation, in Google’s words, it only requires “moderate station-keeping maneuvers,” meaning you just occasionally adjust position.

Another major problem is radiation. High-energy particles in space bombard chips; lightly, they cause data errors (bit flips), and seriously, they damage the chips. Google bombarded their v6e Cloud TPU with a proton beam of 67 mega electron volts and found that these chips are more robust than expected. High-bandwidth memory is the most sensitive component, but anomalies start only after the accumulated dose reaches 2 krad.

Rad is the unit of absorbed radiation dose; 2 krad is roughly several thousand times the dose of a full-body CT scan. Google estimates that the radiation dose for its five-year mission is 750 rad, and its real endurance is about three times that. One single chip was tested up to 15 krad with no hard failures.

Google says the “Project Loon Catchers” plan can only succeed if the TPU can run for at least five years, which lines up exactly with a radiation exposure of 750 rad.

02

Not just Google—Musk and Bezos are all fighting for position

Caption: From Google to SpaceX, from startups to national space agencies, a new infrastructure race over “orbital computing power” is unfolding in space.

Google isn’t the only one.

In late January this year, SpaceX applied to the U.S. Federal Communications Commission to launch as many as 1 million satellites. According to the documents, this is part of a bigger goal: building a solar-satellite network to “meet the explosive growth in AI-driven data demand.”

In December 2025, Starcloud, backed by Y Combinator and Nvidia, launched its first AI-equipped satellite. CEO Philip Johnston predicts that even after counting emissions generated by launches, the carbon emissions produced by off-world data centers are 10 times lower than those of ground data centers.

In March this year, Blue Origin requested permission from the federal government to send a network composed of 50k solar-powered data-center satellites into orbit. In its application, it wrote that moving data centers to space would help ease the pressure that energy and “water-intensive computing” place on U.S. communities and natural resources.

Nvidia is also taking action. The company has already released hardware for space data centers. In February this year, CEO Jensen Huang said that the economics of space data centers “will improve over time.”

Investors are equally enthusiastic. After Starcloud completed a $170 million Series A round led by Benchmark and EQT Ventures, its valuation reached $1.1 billion. Baiju Bhatt, a co-founder of Robinhood, has been investing in space data centers since 2024; his space solar company Aetherflux is raising a new round at a valuation of $2 billion.

03

There are plenty of people opposed to space data centers, and their reasons are quite solid.

Matthew Buckley, associate professor of physics and astronomy at Rutgers University and a theoretical physicist, estimates that each space data center requires solar panels sized like 450 football fields to generate power. Just that portion of the cost is 10 billion dollars, and another 10 billion dollars to send everything up. This doesn’t include maintenance costs.

Buckley added, “You can put a data center in space. This is not physically impossible. I just don’t understand why you would do it.”

OpenAI CEO Sam Altman was even more direct, saying the idea is “too absurd.”

Kathleen Curlee, an analyst at the Center for Security and Emerging Technology at Georgetown University, said: “It’s an extremely crazy idea.” She pointed out that the design lifetime of space data centers is not long—at most only five years. “In the end, the cost required to send data centers into space is simply unreasonable. It’s a long-term goal that’s been packaged as something that can be achieved within a few years.”

Quentin A. Parker, director of the Space Research Laboratory at the University of Hong Kong, believes: “If you do a serious cost-benefit analysis, you’ll find it simply doesn’t hold up. The ground方案 is already there, and it’s very likely to be cheaper than sending anything to space. Sending data centers to space comes with all kinds of problems.”

Among them, launch cost is the biggest variable. Google’s analysis of historical and projected launch pricing data shows that by around 2035, space launch prices may fall to below $200 per kilogram. At that price, the operating cost of a space-based data center (calculated per kilowatt per year) would be roughly comparable to electricity costs for ground data centers.

But CNN reported that Lonestar Data Holdings signed a $120 million contract with Sidus, a U.S. company for the design, manufacturing, launch, and data services of commercial satellites, to build six data storage satellites—each carried on SpaceX’s Falcon rocket. The cost per launch is roughly $10 million, while the storage capacity of these satellites is only a fraction of a ground data center’s.

There’s an even bigger risk: an AI bubble. A report from McKinsey in April 2025 warned that overinvesting in data centers risks stranded assets, while underinvesting means falling behind. In 2025, Alphabet, Amazon, Oracle, Meta, and Microsoft issued $121 billion in new debt through bond offerings, while in 2020 that number was only $40 billion.

04

Easy to become a missile target?

He wrote that tracking and targeting satellites in low Earth orbit is very straightforward, and many countries have already demonstrated anti-satellite missile capabilities. Ground data centers enjoy defense protection, physical security, and an implicit deterrence that operates within national borders—but none of that exists in space. Orbits are fixed, and defense is nearly impossible.

“Those who champion space data centers would better come up with a security plan—and that plan can’t require taxpayers to spend trillions or tens of trillions of dollars to replicate the same level of protection in space for their commercial business.”

Others have pointed out additional problems. Space debris is a real and tangible threat— even fragments the size of coins can damage a satellite’s core components. Space weather such as solar flares could interrupt service. According to reports, some countries are developing “anti-space technologies,” such as jamming systems that can target satellites.

In a 2024 statement, Golestan Radwan, chief digital officer at the United Nations Environment Programme, said: “There are many unknowns about AI’s long-term impacts on the environment, but the data we already have is unsettling. He called for, before deploying AI technologies at scale, settling the score first: whether its positive effects on Earth actually outweigh its negative effects.”

Conclusion

Looking through these latest developments, you’ll find an interesting split in thinking.

On one side are technology companies. Google, Starcloud, Aetherflux—their logic is straightforward: the power bottleneck for AI is right there, and there’s unlimited solar energy in space. Why not use it? Musk wrote on the SpaceX website: “There’s a reason space is called ‘space’,” followed by a laughing-with-tears emoji.

On the other side are security experts. They don’t deny the appeal of solar energy in space, but they see a different picture: satellites in orbit fly at a steady speed along fixed paths, and the time and position can be predicted precisely, making them easy targets.

The two groups don’t think along the same dimension. Technology companies calculate electricity costs per kilowatt-hour; security experts calculate the probability that each satellite gets shot down.

One possible area of consensus lies in how sensitive the data is. If it’s just edge computing—processing tasks where interruptions don’t matter—the logic of a space-based solution may hold. But if you need to run critical missions, that problem can’t be sidestepped: who will protect those servers floating in orbit?

Worth noting is that two technical routes are moving forward in parallel:

One is sending computing power upward, and the other is preparing for possible conflicts.

In the next decade, we may see not only computing power going into the sky, but also the offense-and-defense battle of computing power in the sky.

This article is sourced from Tencent Technology.

Risk warning and disclaimer

        There are risks in the market; invest cautiously. This article does not constitute personal investment advice, and it does not take into account any special investment objectives, financial situations, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are consistent with their specific circumstances. Investing based on this is at your own risk and responsibility.