SpaceX's Potential IPO Is Not What You Think It Is
Musk math strikes again!
Last week, reports began circulating that SpaceX is preparing to go public late next year, with an IPO set to value the company between $1 trillion and $1.5 trillion and raise well over $25 billion in the process. Why does SpaceX need this massive injection of cash? Well, according to Reuters, SpaceX plans to use the funds to buy AI chips and then build and launch space-based orbital data centres. All of this sounds typical and run-of-the-mill in our current techbro hype economy. But, as always with Musk, if you dig a little deeper and start asking questions, the entire narrative falls apart. You see, this IPO is almost certainly not about space-based AI infrastructure but about something far more sinister.
Before we carry on, none of this is financial advice, obviously. But hopefully, you guys have your heads screwed on, and none of you take financial advice from randos online anyway.
Is SpaceX Worth $1.5 Trillion?
Let’s start with the hard question: is SpaceX worth $1.5 trillion?
It might seem so at face value. Over 66% of satellites in orbit today are SpaceX’s Starlink satellites. Over 90% of US space launches are carried out by SpaceX. The company is expected to haul in $15 billion in revenue this year. Surely such a dominant player is highly valuable?
Well, no. For one, the global space launch market is valued at less than $10 billion a year, and it has been stagnant for decades (excluding Starlink). There is also little sign that Starship will change that, as the craft itself is still light-years away from its promised targets. Not to mention there is little demand for it anyway, outside of a handful of NASA missions and SpaceX’s own needs. Moreover, over 80% of SpaceX’s launches are for Starlink, and Starlink’s profitability is, at best, highly questionable. It gained a positive EBITDA (Earnings Before Interest, Taxes, Depreciation & Amortisation) in 2024, but considering SpaceX has likely taken on a lot of debt to build out Starlink, and Starlink satellites only last a few years, that does not mean they are operationally profitable at all! Indeed, some accountants believe that Starlink is still far, far from profitability.
SpaceX, with its huge revenue and gargantuan market share, might seem like a highly valuable company. But in reality, it is a giant hollow tree in a very small forest. So no, I do not think it is worth anywhere near $1.5 trillion or even $1 trillion.
But does its actual value matter anymore? SpaceX is AI now! Surely that will make it worth more than a developed country in our twisted modern world?
Orbital Data Centres
Indeed, the techbro justification of orbital data centres sounds good. Up there, there is constant, free, clean solar energy. What’s more, an orbital data centre won’t take up land, impact local communities’ energy prices, or affect the local water supply, as terrestrial ones do. In orbit, the social, environmental, and operational cost issues of datacentres can, in theory, be reduced to zero. That is pretty damn appealing, given that these factors make current AI data centres wildly unprofitable, and this unprofitability is threatening to pop the AI bubble. This concept is also somewhat proven, as a startup has already trained a basic AI on a tiny prototype orbital data centre (read more here).
So, could SpaceX resolve the AI bubble by building space-based datacentres? That would surely make it worth $1.5 trillion.
Well, for many, many reasons, no. You see, orbital data centres make no sense at all (as I have covered before).
Because AI chips perform significantly better when wired together, these orbital data centres would need to be part of a single, giant satellite, not a constellation of thousands of interconnected small satellites, each with its own GPU rack. Indeed, one proposal for a relatively small 4 GW orbital data centre satellite is a massive four km by four km, with the majority of that size coming from the solar array and radiators. If it had ever been launched, it would be by far the largest thing ever put into orbit by a colossal margin!
This size alone causes immense problems, not just in getting such a gargantuan object into orbit, but also because it makes no orbit suitable.
Low Earth Orbit (LEO) makes sense, as launch costs per kg are the lowest. But that orbit regularly shifts into the shadow of the Earth, cutting off power. So it would have to stay in a special polar-orientated LEO to remain in the sunlight 24/7. Unfortunately, both typical LEO and polar orientations of LEO are full of other satellites and tonnes of space junk flying around at Mach 22. Adding a giant four km by four km satellite to that mix is a recipe for it getting torn to pieces. Furthermore, in LEO, satellites are at risk of geomagnetic storms, which can not only damage the chips through radiation but also deorbit the satellites through induced atmospheric drag. Such a giant satellite will have far more drag and therefore far more risk of deorbiting and burning up in the atmosphere than typical LEO satellites. So, LEO is a challenging location, to say the least.
But while higher orbits, like GTO, get loads of sunlight and have far fewer other satellites or space junk to contend with, the cost per kg to these orbits is substantially more, and out there, they are exposed to far, far more radiation, which can cause critical damage to these chips. More likely, this radiation will corrupt files, which is why many satellites in these orbits run parallel computing and data systems to help mitigate this issue. Needless to say, file corruption in a data centre is unacceptable, and doubling the number of excruciatingly expensive chips in an orbital data centre compared to its earthly counterparts to protect against this problem isn’t viable at all.
Simply put, there is no orbit where this kind of data centre would be safe, even if you spent the extra cash to put it in a higher orbit.
Then there is the cooling issue. AI data centres run their chips at the rugged edge and push them to breaking point. This generates a lot of heat, so to keep the chips efficient and avoid critical damage, they need a lot of cooling. In fact, up to 40% of a data centre’s energy consumption can come from cooling alone.
You might think this isn’t a problem in orbit, as space is very cold. But, while space is ‘cold’, there is not enough atmosphere for convective cooling (which terrestrial data centres use to cool themselves), so only the far less efficient radiative, or black body, cooling can take place. This means objects in space can rapidly accumulate heat if not adequately cooled.
Furthermore, while space is cold, the unfiltered Sun is not! That is why surface temperatures on the Moon can reach 127 degrees Celsius. Because orbital data centres will use solar power, they will be exposed to this crippling heat 24/7.
This means an orbital data centre will have far greater cooling requirements than a terrestrial one. Therefore, it will need larger radiators and greater energy demand, which will increase the satellite’s mass and, in turn, increase launch costs. Even after all that, this additional heat stress, combined with the increased exposure to orbital radiation, could significantly shorten the operational lifespan of these chips.
Then there is the fact that even simple maintenance, such as replacing a damaged chip, is cost-prohibitive on an orbital data centre. Overall, the data centre’s efficiency could rapidly plummet as these issues mount.
And what about space junk? Sure, in LEO, a single fragment could destroy the entire satellite. But an AI data centre only has a three-year lifespan. So, putting a meaningful number of these giant data centres in orbit would either create serious issues of orbital overcrowding and space infrastructure falling to Earth or take us significantly closer to Kessler syndrome if they are left to become space junk. In terms of e-waste problems, orbital data centres are significantly worse than their terrestrial counterparts.
Then there is the risk of losing the data centre to a launch explosion. Historically, this was a very low risk for the space industry. But it isn’t for Starship, as a sizeable number of launches have ended in a fireball.
Purely from a practical point of view, orbital data centres make no sense. But somehow, it gets so much worse.
