Elon Musk says SpaceX is about to give Grok a very large dose of engineering knowledge.

In a post on X, Musk said SpaceX’s “massive corpus of world-class engineering data” will be added during the supplemental training of Grok’s upcoming 2 trillion parameter model. He said the data will “dramatically improve Grok’s engineering capabilities.”

There is one important limitation. Musk said the training data will not include material blocked by ITAR, the US rules that restrict the sharing of certain military and aerospace technology. So this will not mean handing over every SpaceX document or design to an AI model.

Even with that restriction, the amount of useful information could be significant. SpaceX works across rocket design, spacecraft, communications, manufacturing, flight operations and satellite systems. That kind of data could help Grok deal with technical questions in a way that general internet text cannot.

Of course, Musk’s post is still a claim about what the training will achieve. He has not shared details about how much SpaceX data will be used, what form it will take or how the company will measure the improvement. We also do not know whether the data will make Grok better at solving real engineering problems or simply better at explaining engineering concepts.

That said, the model in question is already shaping up to be one of the biggest upgrades. As we reported earlier, Musk said the 2 trillion parameter model, expected to be called Grok 4.6, was due to finish its initial training the following week. This latest update suggests the work will continue after that first stage.

It also fits with xAI’s wider push to make Grok useful beyond ordinary chat. The company has been moving Grok Build toward a more complete app-building service, with possible voice input and direct deployment features. We covered those details here.

Whether Grok becomes a genuinely useful engineering assistant will depend on more than the size of its training set. It will need to give accurate answers, understand constraints and avoid confidently suggesting designs that do not work. SpaceX’s data may help, but the real test will come when people start using the model for serious technical work.