Elon Musk: SpaceX engineering data to train Grok's 2-trillion-parameter model

Elon Musk said on X on July 21 that SpaceX's world-class engineering data — excluding anything restricted under ITAR export-control rules — will be folded into supplemental training for Grok's next large language model. Musk said the data would be added during training of a 2-trillion-parameter run, calling it a move that will dramatically improve Grok's engineering capabilities.

The 2-trillion-parameter model would be roughly a third larger than the 1.5-trillion-parameter V9 architecture currently powering the public Grok 4.5. Feeding in SpaceX's hands-on rocket and satellite engineering data is expected to sharpen Grok's performance on engineering-heavy tasks like code and design problem-solving.

The announcement matters to Tesla shareholders because Grok is already built into Tesla vehicles as the "Hey Grok" voice assistant — just hours earlier, Tesla's 2026 Summer Update expanded Grok's in-car duties to phone calls, music and climate control. A stronger base model doesn't automatically translate into a better in-car assistant, but it's another data point on how Musk is pooling data and talent across SpaceX, Tesla and xAI to push his AI ecosystem forward.

That said, this is primarily a SpaceX/xAI story with no direct financial impact on Tesla's results. Coming one day before Tesla's July 22 Q2 earnings call, it's useful background on the broader AI ecosystem Musk is building around Tesla's Robotaxi and Optimus "physical AI" narrative, rather than a market-moving Tesla headline itself.