SpaceX’s earnings call last week took a bigger turn as Elon Musk unveiled a major Nvidia tie-up.
Musk said SpaceX will deploy Nvidia’s Vera Rubin NVL723 AI systems both in space and on the ground. SpaceX will also work with Nvidia to develop the Starmind AI1 satellite computing payload. “Each of the Starmind satellites will include Nvidia Rubin GPUs and Vera CPUS for data centre-class space compute”
Noting that SpaceX will go exclusive with Nvidia, Musk noted that it would mark the next step in SpaceX’s plans to put AI-centric data centres in orbit.
“Now, that’s exactly the kind of signal Wall Street loves because it shows AI demand is expanding beyond the usual hyperscaler,” Sidharth Sogani is CEO of Blue Aster Capital (Bahrain) and CREBACO Global.
But, the investors need a reality check. “SpaceX isn’t really a ‘beyond Big Tech’ customer in the way people are framing it. They already operate Colossus data centres and have been renting out spare GPU capacity to companies such as Anthropic and Google. It’s more about the definition of a hyperscaler quietly expanding.”
Adding to this, Ankush Tiwari CEO and Founder of pi-labs, says, “I would read the word exclusive carefully. When a customer of this size commits entirely to one vendor, that is rarely a purely technical decision. There is almost always commercial architecture behind it, whether allocation priority, pricing, or co-design access, that is not visible in the announcement.”
Could non-traditional sectors such as aerospace, manufacturing, healthcare and energy become the next leg of AI spending?
“Yes, and I am really excited about that part,” says Sogani, adding, “Putting GPUs in orbit for real-time geospatial processing and flight analytics opens up an entirely new deployment environment.”
But, think of this market in layers, explains Tiwari.
The first layer is building capacity, which is the hyperscalers, and that phase is well underway. The second layer is building solutions, which is the model builders. But neither of those is where the volume ultimately sits.
Beyond capacity and models, the compute gets consumed by application builders. That is where the scale comes from. Aerospace, manufacturing, healthcare and energy will show up as the next leg of demand, but mostly not as direct GPU buyers. But they are unlikely to buy GPUs directly. Instead, they will use AI applications built for their needs.
This will also change the nature of AI spending. Application demand will have a much larger user base and grow over time. Most of that demand will come from inference—using AI models to generate answers and perform tasks—rather than training models, which has driven much of the spending so far.
“These industries are still tiny compared to the roughly $300 billion in capex the four largest hyperscalers spent in 2025,” concludes Sogani, adding, “So these sectors matter because they extend the duration of AI demand, not because they meaningfully change its size today.”