Musk has been talking about SpaceX’s mission in his own words, and the filing that started this whole run-up came with a number that made even jaded market watchers blink: a path to a $28.5 trillion revenue opportunity. That is the kind of figure you put in a deck when you want people to stop thinking in quarters and start thinking in decades.
The market’s reaction, though, was less poetic. Nvidia shares have spiked nearly 10% in the last month. SpaceX stock has fallen more than 13% as the excitement around its heavily anticipated debut cooled off. Analysts covering both names lean toward Nvidia having more upside from here.
As someone who spends most days wiring up bots, agents, and inference pipelines, that split makes intuitive sense to me. Not because I have a strong opinion on rocket economics, but because of where the actual bottleneck sits in every AI project I have ever shipped.
The bottleneck is never the location of the server
Orbital compute is a genuinely interesting idea. Put data centers where cooling is free and solar is constant, and you sidestep some ugly terrestrial constraints. I get the appeal. But when I think about what actually slows down the bots I build, the physical address of the rack has never once been on the list.
What is on the list, consistently:
- Available GPU capacity when I need to fine-tune something on a deadline
- Whether my framework of choice has solid kernel support for the hardware I am renting
- Latency between my agent’s decision loop and the model serving it
- Cost per token at the volume where a side project turns into a product
Every one of those is a compute-layer problem. Moving the compute layer to orbit does not solve any of them and adds a brutal new one: round-trip time and bandwidth to a moving target hundreds of kilometers up. For batch training jobs, maybe you can eat that. For an agent that needs to make a tool call and respond before a user gets bored, you cannot.
Why the picks-and-shovels position keeps winning
Barron’s framed Nvidia as the big winner from SpaceX earnings, and the logic there is not complicated. If someone announces an ambitious plan to put AI infrastructure in space, the very first thing that plan requires is a large order of accelerators. The moonshot generates demand for the shovel regardless of whether the moonshot lands.
That asymmetry is the whole story. SpaceX has to actually execute an unprecedented engineering feat to justify its valuation. Nvidia gets paid during the attempt. Commentary out of CNBC pointed to multiple large long-term growth paths for Nvidia: sovereign AI, enterprise AI, AI-native startups. Those are three different customer bases with three different budget cycles, and none of them depend on a single technical bet paying off.
I see a smaller version of this in my own work. When I build a bot, the interesting part feels like the agent architecture, the prompt orchestration, the tool routing. The part that determines whether it works is the model underneath and the silicon underneath that. The flashy layer sits on top of the layer that everyone actually pays for.
What builders should take from a stock story
I am not giving investment advice, and you should not take portfolio construction tips from someone whose main skill is debugging retry logic. But there is a lesson in how this played out that applies directly to technical decisions.
Announcements move sentiment. Capacity moves projects. The SpaceX filing generated enormous attention and a June launch date and a trillion-dollar addressable market claim. What it did not generate, as far as anything verified goes, is compute I can rent this afternoon for a fine-tuning run.
When you are choosing what to build on, the boring question beats the exciting one every time. Can I get the hardware. Does the tooling work. What does it cost at scale. Will it still exist in eighteen months. Ask those four and you tend to end up choosing the same infrastructure that the market has already decided has more upside, for roughly the same reasons.
Keeping an eye on the actual signal
The thing I will watch is not the share prices. It is whether any of this orbital work produces developer-accessible endpoints. The moment there is an API, a pricing page, and a latency figure I can measure, this stops being a market narrative and becomes a real option in my stack.
Until that day, my bots run on the same accelerators everyone else’s do, and the space story is a fascinating thing to read about between builds. The rally in one stock and the slide in the other is the market saying something builders already know: the layer that everything else depends on is the layer worth owning.
🕒 Published: