SpaceX has designed a way to equip its AI1 satellite with an interchangeable computing payload that allows the installation of hardware from any competent chipmaker. Unveiled in a video published on X, AI1 is the first ever orbital and satellite data center, designed to run computing operations from space. It is claimed to have a maximum power of 150 kilowatts, producing 120 kilowatts on average, with an output of 70 kilowatts per tonne. Power will be provided via a 150-kilowatt solar panel built into the craft, which is expected to provide 250 watts per square meter of power – all using technology made by SpaceX, minus the powerful payload.
The interchangeable chipset is a big deal here as this is how the company aims to address market challenges that could actually hold back the project. SpaceX previously reported, as shared by Tom’s Hardware, that it could not supply enough chips for this same orbital project, which is part of a joint venture with Tesla and its TeraFab solution. By allowing the platform to remain open to all competent suppliers, SpaceX can install what is most competitive at any time, whether it is readily available, cheaper, or more powerful.
In the video, Musk estimates that AI1’s computing payload is the equivalent of a single Nvidia GB300 rack, which requires about 140 kilowatts of power on Earth. For cooling (which is more difficult in the vacuum of space, since traditional methods are not viable because waste heat has no material to dissipate into), AI1 will also be equipped with 110 square meters of deployable liquid radiators. Additionally, pump loops are installed for redundancy purposes, as well as micrometer shielding, which should help repel debris.
What’s so powerful about the payload?
The modularity or open design of AI computer chips is what makes the AI1 so powerful to begin with. Simply put, an AI data center is the facility or location that hosts the IT infrastructure for “computing” operations. These operations consist of training, deploying and delivering AI applications and services. AI workloads have incredibly high demands on data storage (hard drives), high-speed memory (RAM), and processing (data center GPUs and CPUs). When you look at what differentiates Nvidia’s AI accelerators from traditional GPUs, like the H100, it’s all about computational tasks, processing, and math management, versus visual or hardware logic for graphics. Simply put, they need a lot of power. Lots of computing power, to be exact. Being able to trade for what’s available, what’s most powerful, or what’s most efficient will absolutely make AI1 satellites both extremely useful and scalable.
For now, the constraints are due to hardware shortages and market complications, but eventually it may be more about the most powerful hardware to include in satellites to maximize computing power. As Musk says: “It seems like a reasonable starting point is 150 kilowatts peak, 120 kilowatts sustained. » The initial models will be a “preliminary version of the first version of the SpaceX AI satellite.” It’s not difficult to imagine future versions or generations of the AI1, with more powerful hardware, thanks to its open design.
Why is AI1 so important in the first place?
AI1 will be SpaceX’s first-generation data center operating in low Earth orbit. With data centers consuming monumental resources on Earth – one tech company even wants to install mini data centers in your home – space data centers have been proposed as a viable alternative. But the idea has its detractors. Nvidia CEO Jensen Huang discussed several reasons why we don’t put AI data centers in space, namely cooling challenges and excessive costs of launching the hardware into orbit. Some critics, like the CEO of Amazon Web Services, say the technology required to achieve this is “simply not economical.”
AI1 is SpaceX’s answer to the challenge and will specifically run AI computing workloads from orbit, powered entirely by solar energy. If it works, more of that computing power could be allocated to orbital hardware rather than using valuable resources here on Earth, especially land and water. Companies are getting more creative with where data centers are located, such as the new underwater and wind-powered data center in China. However, we still have operations on Earth underway, such as Kevin O’Leary’s rural center in Utah.
