The "space AI data center concept," which has been making headlines with companies like SpaceX, is generating intense debate – the reality is not so simple.

Is the "Space AI Data Center Concept" a promising future, or is it simply unrealistic?

This time, I'd like to discuss the recently popular idea of "building an AI data center in outer space," which has been generating a lot of buzz lately, and the resulting debate.

If you're unfamiliar with the "Space AI Data Center Concept," please read on, as we will explain that as well.

This news article was the inspiration for writing this piece.
⇒ Masayoshi Son questions Elon Musk's space AI data center concept: "What's the point?" | Business Insider Japan

As you all know, Masayoshi Son is the type of person who, based on his public image, would say, "Let's keep challenging ourselves with new things." In fact, one might even get the impression that he would wholeheartedly support it and invest trillions of yen. However, "that Masayoshi Son" has raised objections to the plan and has stated that he will continue to focus on developing data centers on Earth, not in space.
I decided to write this because I felt that this issue was starting to generate serious debate and mixed opinions.

What is the "Space AI Data Center Concept"?

First, let me briefly explain what the "Space AI Data Center Concept" is.

This concept has become widely known because it is supported by Elon Musk, who is well-known for his work with SpaceX and Tesla. Even SpaceX, which made headlines with its unprecedentedly large-scale IPO, has the space AI data center concept as a major project it plans to undertake in the future.

The boom in generative AI is driving a construction boom in data centers.

As of this writing, there is a massive global boom in generative AI. Construction of data centers for generative AI is progressing at an alarming pace, and the rate of construction is so high that even with NVIDIA producing high-end GPUs at full capacity, some costing over 5 million yen each, they are selling so fast that it's not enough. The power supply necessary for operating these data centers is also becoming strained, leading to a construction boom of power plants and transmission lines. Everything related to generative AI data centers is becoming critically strained.

While there are certainly strong opinions that this situation is a "bubble," if you ask me what the current situation is, I would say that "it's a situation where it feels like no matter how many AI data centers we build, it won't be enough."

"If there aren't enough data centers, let's build one in outer space."

In that situation, the idea emerged: "If there aren't enough data centers, let's build them in space." Proponents cited the following advantages:

• Solar power generation can provide a free and inexhaustible supply of energy.
  • Some argue that deploying solar panels in outer space could allow for constant power generation using energy from the sun, thus solving the power problems plaguing AI data centers.
• It can be cooled in the cold space.
  • Some argue that the extreme cold of outer space, unlike anything on Earth, could solve the cooling problems (caused by massive power consumption) that plague AI data centers.

Furthermore, it was Elon Musk who started pushing this concept forward. The combination of the high profile of SpaceX, which launches a large number of space rockets, and the hugely popular phenomenon of generative AI resulted in an explosion of public attention.

Doubts about the "concept of creating an AI data center in outer space" (Is it technically impossible?)

On the other hand, from the outset, there have been criticisms and doubts that this concept is "technically unrealistic." I also think there are quite a few difficulties involved.

Cooling is actually difficult in outer space.

Outer space is a cold environment, but cooling a data center there would likely be difficult because there is no air.

When cooling GPUs in ground-based data centers, the GPU (semiconductor) generates heat, that heat is transferred to a heatsink, the heat is then transferred from the heated heatsink to cool air, and the hot air is then drawn out and exhausted by a fan.

What you should pay attention to is that heat is transported by materials "through adjacent materials." The "GPU semiconductor," "heat sink," and "air" are in contact with each other, and heat is transported outward through adjacent materials (through "convection" and "conduction"), thus achieving cooling.

However, in outer space, which is almost a vacuum, there is no matter around the data center (it's more like a thermos flask insulated by the vacuum). The same methods used for data centers on Earth cannot be used to release heat to the outside.

In space, cooling can only occur through "radiative cooling".

If you want to cool something in the near-vacuum of space, you would use "radiative cooling" (thermal radiation). This is a method of releasing heat energy into space by emitting infrared radiation (electromagnetic waves) from the surface of a hot object. However, this method requires attaching a huge heat sink, and it is considered difficult to release heat to the outside as efficiently as "convection" or "conduction".

If adequate cooling is not possible, even if the system is operated as a data center and attempts to perform a large amount of computation, the heat generated by the computation will not be able to be dissipated sufficiently, preventing it from operating properly (or it will overheat and break down).

Launching rockets into space is expensive.

Launching something into space by rocket is inherently expensive. I recall it being said that launching just one kilogram of material used to cost around 1 million yen.

Currently, significant reductions in launch costs are being achieved through measures such as rocket reuse. However, even with these dramatic cost reductions, this is still clearly a disadvantage compared to the fact that ground-based data centers can be constructed with virtually no concern about weight.

Measures against strong cosmic radiation are necessary.

In outer space, intense radiation is present. Not only is it harmful to the human body (due to exposure), but it can also cause malfunctions and deterioration in electronic devices, and can even corrupt data in memory (for example, a bit that should be 0 might become 1).

It may be necessary to take measures such as executing processing while taking into account the likelihood of processing errors (which reduces computational efficiency), installing mechanisms to block radiation (which increases weight), or using specially designed semiconductors (which would require the development of new GPUs, etc.). These are all factors that are generally not necessary to consider in ground-based data centers, making space a disadvantage.

Electricity costs account for only 7% of the operating costs of an AI data center (Masayoshi Son).

The concerns raised so far are commonly cited as problems with the concept, but Masayoshi Son says, "There are also doubts about the cost aspect from the very beginning."

Even if electricity were free in space thanks to solar power, the cost of electricity only accounts for 7% of the operating costs of an AI data center (according to Masayoshi Son), so a 7% cost reduction wouldn't be a game-changer. On the other hand, cost increases due to the concerns mentioned above are certainly to be expected.

Decreased efficiency due to increased communication latency (Masayoshi Son)

Furthermore, Masayoshi Son has pointed out that communication delays resulting from communication between space and Earth (or between space data centers) are expected to reduce the efficiency of data centers.

Given these circumstances, they cannot currently confirm whether there is any advantage over ground-based data centers, and even if they conclude that space is advantageous, it is thought that it will take many years. Therefore, they cannot explain to their shareholders why they are investing in space.

• merit
  • Solar power could potentially make electricity costs free.
  • Outer space is cold.
• Concerns
  • Cooling itself is likely to become difficult.
  • Cost of launching into space
  • Countermeasures against cosmic radiation
  • The theory that electricity costs only account for 7% of expenses.
  • Increased communication delays are unavoidable.

However, I get the feeling that Elon Musk is doing it intentionally.

Now, regarding the doubts I've raised so far, I don't think Elon Musk is completely unaware of them; I believe he is certainly aware that these are legitimate criticisms.

Furthermore, if clear solutions to these concerns already exist, then the space AI data center concept would have already been explained, and I believe Masayoshi Son would be reasonably satisfied with it.

It appears that SpaceX has already considered the technical challenges involved in resolving issues such as cooling and launch costs. They seem to be exploring how to proceed with these efforts and believe there is room for improvement, but I suspect they still don't think they have a sufficient chance of success (hence, no clear solutions have been presented).

The space rocket business depends on launch demand.

It might seem irrational to spend a huge amount of money to promote a space AI data center concept when there isn't a sufficient chance of success, but SpaceX might be saying that it doesn't matter even if Masayoshi Son's concerns prove true.

SpaceX is a private company that develops and launches space rockets, but in expanding its business, it is making something "amazing" a reality beyond just technology.
When private companies entered the space rocket business, there were fundamental limitations to begin with. The reality is that the maximum size of the business is capped by the global demand for satellite launches.

For example, in 2010, the total number of artificial satellites launched worldwide was approximately 100. Even if we assume that each satellite sells for 1 billion yen, monopolizing all satellite launches worldwide would only amount to 100 billion yen. With sales of that level, it seems difficult to continue developing expensive rockets aimed at reaching Mars, doesn't it?

So, I think that's why there were many rocket ventures at one point that aimed to realize space travel, hoping to make it happen through private sector efforts. However, even getting trained astronauts into space is extremely difficult, so realizing space travel for civilians was an impossibly difficult goal. Moreover, with the enormous costs involved, there would only be demand from a few wealthy people who are indulging in such whims.

If there's no demand, make it yourself: Starlink

However, SpaceX came up with an unexpected solution: "If there are no customers, we'll become customers ourselves."

SpaceX has announced that it will launch a large number of small communications satellites into low Earth orbit (LEO) and use them to launch a space internet service. This is the well-known "Starlink" project. With its new venture, the company has created unprecedented demand for satellite launches, putting it in a position to launch rockets at a rapid pace.

Initially, there were questions about how many people would actually want to use a service like Starlink. However, the subsequent Russian invasion of Ukraine led to the recognition that services like Starlink are technologies that can determine the fate of a nation. When SpaceX went public, it was revealed that annual revenues related to Starlink already amounted to $11.3 billion (1.8 trillion yen).

Regardless of the success or failure of the space AI data center concept, "the rocket will be launched."

Starlink has ultimately become a huge success and a major source of revenue for SpaceX, but beforehand, there was uncertainty about whether it would be a successful endeavor. If it weren't for the situation in Ukraine, it might have been seen as an interesting new service, but since most people weren't significantly inconvenienced by conventional internet connectivity services, it might still be a loss-making venture despite generating buzz.

Similarly, it remains unclear whether the space AI data center concept will be a successful endeavor, and it seems that Masayoshi Son doesn't see it as a promising idea at the moment. I also share the same view as I've already written; I think there are many difficult factors involved, so like Masayoshi Son, I wonder if it wouldn't be better to locate the AI data center on Earth?

Even if, as feared, "many satellites carrying GPUs are launched into space, but the service remains unprofitable as predicted," rocket launches will continue at a rapid pace. And this situation itself is not bad for SpaceX, which aims to reach Mars. Even if the new business remains unprofitable, they can still dominate the rocket launch market.

What might be the "real advantages" of a space AI data center

In my personal opinion, the explanations for the advantages—that there are no electricity costs and that it's cold in space—are questionable, and it seems difficult to create a cost advantage over ground-based data centers. However, on the other hand, I also think that aspects that are not being actively mentioned now may be marketed in the future.

First, since outer space "doesn't belong to any country," space AI data centers differ from ground-based data centers in that they are "independent of any government." As of this writing, "Claude Mythos" has become a hot topic as a generative AI service with a high ability to discover vulnerabilities, and then shortly after, the provision of Mythos was "suddenly stopped at the behest of the US government." For those who are concerned about situations like this happening at the behest of a government, the fact that it is "in outer space, which is not under the jurisdiction of any government" is a significant advantage.

Furthermore, in typical use, data is sent from Earth to space for processing, and the processing results are returned from space, which would cause communication delays. However, if we "consider Starlink usage as a premise," it becomes more rational for not only communication but also processing to be completed in space. In that case, we may see opportunities to market it to Starlink users, or to use it as a platform for providing cloud services to Starlink users.

If these speculations are correct, it seems there will be more demand for data centers that meet general usage needs than for AI data centers. Therefore, instead of a space AI data center, a space cloud service with more generally usable functions, like AWS, or a service for securely storing data in space, may be established in satellite orbit, and this may become a new business for SpaceX in the future.

An example where it's unclear what will happen next.

In other words, whether the space AI data center concept is likely to succeed or not depends on things that might seem "a bit surprising" to those who don't understand the situation, such as how cooling will work, weight and launch costs, and whether communication delays will be a problem.

And I think that in the world, there are quite a few cases where something "a little unexpected" is the crucial factor that separates success from failure.

The fact that something is "difficult to understand" is a "crucial" factor that separates success from failure.

For example, our file transfer middleware, HULFT, as you know, continues to sell "even now" as the overwhelming de facto standard product in Japan. However, it's difficult for people to understand why a transfer platform that can perfectly realize safe, secure, and reliable file transfer has been so necessary for the world. The reaction tends to be something like, "file transfer?"

A similar situation exists with the space AI data center concept. You might think, "Is it about cooling?", but rather, the trends of various companies in cooling technology will likely be the key point when considering the future of this field. Or perhaps something "even more unexpected" than what we've discussed here will be the deciding factor in success or failure.

The ability to experiment and figure out what to "connect" with will likely determine the future of our society.

Assuming Elon Musk shares the same understanding as Masayoshi Son (it's unlikely he "doesn't understand anything"), if so, why do they arrive at different conclusions despite having the same understanding of the current situation? Perhaps the difference lies in whether or not Elon Musk alone possesses the means to "connect" the two initiatives—the "space rocket business + space AI data center plan."

I also considered that we should pay attention to initiatives in space other than AI data centers, and my opinion was that a winning strategy could be created by "connecting" these with other things, such as "space AI data center plan + existing Starlink" or "space AI data center plan + sovereign cloud needs."

Our company provides products such as "DataSpider" and "HULFT Square," which allow users to freely data integration various types of data, software, and cloud services using only a GUI. These products have been supported for many years because "connecting and combining" data—such as "it would be so convenient if I could use that data from Salesforce in kintone"—is actually a common and successful way to meet real-world IT needs.

These "connectivity" needs aren't always identified through prior analysis; they can also emerge as "great insights" from the field after IT system implementation. Alternatively, "good usage methods" can be discovered through proactive trial and error in the field. Our product allows data integration to be built "solely through a GUI" so that business users can utilize it themselves and successfully extract results from such possibilities.

Like Masayoshi Son, I also find the current plan to "build an AI data center in outer space" somewhat unconvincing. However, as I've already written, I believe there are motivations for the plan to proceed due to the rocket development side's circumstances, so for now, I think that "a situation where a large amount of computing resources are deployed in space" will "become a reality for the first time in human history."

The AI data center project is (as expected) somewhat underwhelming, but we are now in an unprecedented situation where "computational resources already exist in space." Is there a good use for them? Perhaps someone other than SpaceX will discover an amazing use for it, leading to a major breakthrough (it was the Ukrainian military, not SpaceX, that discovered the numerous practical uses of Starlink).

Whether things will proceed as SpaceX has currently explained, or whether other uses will be discovered later, remains to be seen. However, if the era of space data centers arrives, we hope you will consider using "DataSpider" and "HULFT Square." At that time, our products will likely also have "connection adapters (connection connectors) for space data centers," allowing business users to connect and experiment with different uses using only a GUI, and to utilize "space IT" in their own businesses.

The person who wrote the article

Affiliation: Marketing Department, Digital Marketing Division

Ryo Watanabe

・2017: Transferred from Appresso Co., Ltd.
After majoring in information engineering (artificial intelligence lab) at university, I struggled in the development department of a startup.
・Small and medium-sized enterprise management consultant (as of 2024)
・Image: I took over the "Fukusuke" name that was previously used by our company.
(Affiliations are as of the time of publication)