I think Musk’s remarks this time to the entire SpaceX workforce are worth rewatching, because he has already laid out very clearly his judgments about the next 5 to 10 years.

He said that SpaceX’s AI revenue will surpass other lines of business as early as September, and in the next five years, AI could even account for 99% of the company’s value. SpaceX currently has about 1.4 gigawatts of AI computing capacity, and its goal is to reach 10 gigawatts by the end of 2027. According to Musk’s own projections, this scale could correspond to annual revenue of $300 billion to $500 billion. Of course, these figures are more like management’s long-term targets right now—the real realization will depend on capital expenditures, construction speed, and computing capacity utilization.

But more important than the digital layer is his blueprint.

Previously, when we understood SpaceX, it was rockets + Starlink. Now what Musk wants to do is to string the entire system together: Starship will deliver infrastructure into space; Starlink will provide global communications; super data centers on the ground will handle AI training; in the future, space data centers will take on part of AI inference; Grok will handle the model; and finally, AI will be installed into cars and robots to enter the real world. So what he said is actually very important—“In the long run, the future fundamentally belongs to AI and robots.”

If you follow Musk’s line of thought, I think future wealth opportunities will gradually become clear.

Don’t just focus on which AI model is stronger. Instead, look for what AI is increasingly lacking as it develops.

The stronger AI gets, the more computing power it needs, so you need chips like Nvidia and AMD. The more computing power there is, the more HBM is required, so SK hynix and Micron benefit. To connect millions of GPUs, you need high-speed networks and optical communications like Arista and Lumentum. As chip demand keeps exploding, you can’t do without TSMC, Intel, and semiconductor equipment. And when data centers enter the gigawatt level, the bottleneck holding back AI may no longer be GPUs, but electricity—this is also why areas like GE Vernova, power grids, gas turbines, and nuclear power are becoming increasingly worth paying attention to. Musk has even started accelerating AI data center construction by building his own power.

After that, I believe the biggest space for imagination is in physical AI.

AI solves the “brain,” while robots solve the “body.” If AI ultimately moves from chatbots on screens to factories, logistics, cars, and even homes, then what humanoid robots replace won’t be just a software market, but a portion of the physical labor of billions of people worldwide. Tesla’s Optimus, at its core, is a bet on this future.

So in what Musk said this time, the “wealth code” I see is not a single stock, but an entire industrial chain:

AI → computing power → chips → storage → optical communication → data centers → electricity → space computing → robots.

Over the past twenty years, the biggest opportunity on the internet came from connecting people with information; over the next ten years, AI may directly reconstitute productivity.

Whoever controls the foundational infrastructure that’s indispensable in the AI era—and whoever can truly bring AI into the physical world—will be the next Nvidia.