Hot Take: Elon Might Win the AGI Race and Nobody Is Paying Attention

I'm going to make a prediction that most people in the AI space would laugh at right now. I think Elon Musk has a legitimate shot at winning the race to AGI. And the reason has nothing to do with Grok being the best model.

Let's get the obvious part out of the way. Grok is not the consensus best AI model. The current AI race is dominated by Anthropic, OpenAI, and Google, with xAI still trying to close the gap. On coding benchmarks, Claude leads the official SWE-bench leaderboard, and Grok trails the leaders. And xAI has had a rough start to 2026 — all 11 of its non-Musk co-founders have now departed the company.

So why do I think Elon could still win?

TeraFab.

On March 21st, Musk unveiled TeraFab — a $20-25 billion chip fabrication plant in Austin, Texas, jointly developed by Tesla, SpaceX, and xAI. Intel joined the project on April 7th. The goal is to produce more than one terawatt of AI compute capacity per year. Bernstein analysts have estimated that scaling to the full one-terawatt target could require around $5 trillion across hundreds of fabs — a number that gives you a sense of just how ambitious the vision is.

But here's the part that makes my brain melt. Musk said 80% of TeraFab's compute output would go to space-based orbital AI satellites. His argument is that solar energy collection in orbit is significantly more efficient than on Earth's surface — roughly 3x when you factor in 24/7 availability versus day/night cycles on the ground. The engineering challenges are real though. Cooling in a vacuum is actually harder, not easier — there's no air for convection, so you have to rely entirely on thermal radiation, which requires massive radiator panels. That's a problem SpaceX will need to solve. And Musk has floated longer-term ideas about lunar manufacturing, including the possibility of using Optimus robots, though that's very much a far-future vision at this point.

This is where the rest of the AI industry should be nervous.

Right now, the AGI race looks like a competition over who can build the best model. Who has the smartest researchers. Who can squeeze the most capability out of the current hardware. And on that front, Anthropic, Google, and OpenAI are all doing incredible work.

But what if the real bottleneck isn't talent or algorithms? What if it's compute?

Musk is the only person in the race who is vertically integrating the entire stack — from chip fabrication to energy infrastructure to the physical deployment of compute in space. Everyone else is buying GPUs from Nvidia and renting data center space. Musk is building the factory that makes the chips and the rockets that could eventually put them in orbit.

If TeraFab delivers even a fraction of what's been announced, xAI will have access to more raw compute than any other AI lab on Earth. And at some point, raw compute wins. It's the same principle behind every arms race in history: whoever has the most resources eventually overwhelms whoever has the best tactics.

I know this is a hot take. I know people will point to the fact that Grok isn't the best model today and say compute alone doesn't get you to AGI. And they're right — today. But we're talking about a 3-5 year horizon. Models improve. Algorithms get better. And when they do, the lab with effectively unlimited compute is the one that can iterate fastest.

Everybody's watching the model benchmarks. I'm watching the factory.

Sources: TeraFab Wikipedia, Tom's Hardware, Bloomberg, FinTech Weekly, Bernstein Research