When investor Jim Chanos questioned Tesla Inc. (TSLA)'s AI spending strategy, CEO Elon Musk had a pretty straightforward answer: training wasn't the problem.
The discussion kicked off Sunday on X, where Atreides Management investor Gavin Baker jumped in to defend Tesla's approach. Baker's argument is actually quite clever when you think about it. Tesla customers are essentially paying for the inference compute hardware because the AI processing happens inside their cars, not on some massive server farm somewhere.
The Economics of Training AI in Cars
This setup gives Tesla a unique cost advantage. Real-world video data scales infinitely as more cars hit the road, which naturally lowers the capitalized training costs. You're not paying to build bigger and bigger data centers when your customers are driving around generating training data with hardware they already bought.
Musk agreed with Baker's reasoning and added some context. "The reason Tesla hasn't yet spent more on AI training is that it wasn't yet the Limiting Factor for FSD," the CEO explained. Translation: why throw money at something that isn't your bottleneck?
But here's where it gets interesting. Musk said training will become the limiting factor for Optimus, Tesla's humanoid robot project. "It will be that for Optimus, so we're building Cortex 2 for training," he noted.
Cortex 2 is Tesla's dedicated supercomputer being constructed as a companion site near the Gigafactory in Texas. Musk has positioned Optimus as a potential solution to labor shortages and even floated it as a path to universal basic income, though those are ambitious claims for a robot still in development.
Taking Shots at Waymo
Meanwhile, Musk couldn't resist taking a jab at Alphabet Inc. (GOOGL) (GOOG)'s autonomous vehicle service. When Waymo announced its fleet had grown to 2,500 robotaxis across multiple cities, Musk dismissed it with a curt "Rookie numbers."
The Tesla CEO has been talking up FSD recently, claiming it could spread faster than "any technology ever" and suggesting the company could enable self-driving for "millions of pre-existing cars" through a simple software update. That's a pretty bold claim, especially given the current regulatory environment.
The NHTSA Investigation Problem
Speaking of regulation, Tesla's FSD system is currently under investigation by NHTSA for multiple traffic violations and accidents involving vehicles operating on FSD (Supervised) or Autopilot. The probe affects over 2.88 million vehicles, which is a significant chunk of Tesla's total fleet.
That investigation adds some context to the competitive posturing. While Musk mocks Waymo's 2,500-vehicle fleet, those vehicles are operating commercially in geofenced areas with regulatory approval. Tesla's approach is different: deploy the technology broadly as a driver assistance system, collect massive amounts of real-world data, and iterate.
It's a fundamentally different strategy with different cost structures and different regulatory challenges. Whether it ultimately proves more successful than Waymo's approach remains to be seen, but Musk's explanation of why Tesla hasn't spent more on training infrastructure at least makes economic sense given their model.
Price Action: TSLA rose 0.59% to $404.35 at the end of regular trading on Friday and gained 0.27% to $405.45 in after-hours trading.