SpaceX to Feed Its Engineering Data Into Grok's Next Model

HAWTHORNE, Calif. — Elon Musk is handing his AI model something no competitor can buy: decades of real-world rocket engineering. Musk said this week that SpaceX will pour its internal engineering data into the next major training run for Grok, the model now housed inside SpaceX following February's merger with xAI.

In a post on X, Musk wrote that SpaceX's "massive corpus of world-class engineering data" — excluding anything restricted under U.S. arms-export law — will be added during supplemental training of what he called the "2T run," a roughly two-trillion-parameter model that would nearly double the scale behind the current Grok 4.5. It is the clearest example yet of the strategy underpinning SpaceX's transformation into an AI powerhouse.

A Data Moat No One Else Has

The excluded material falls under the International Traffic in Arms Regulations, which restrict export of technical data tied to defense and space hardware. That likely walls off propulsion specifics for the Merlin and Raptor engines and guidance-and-control details for SpaceX's launch vehicles.

But it leaves an enormous amount on the table: manufacturing know-how, materials science, and Starlink hardware design accumulated since 2002. No other AI lab has access to a proprietary trove of aerospace engineering at that depth — the kind of hard, physical-world data that is far scarcer than the text scraped from the open internet.

The Same Playbook Across Musk's Companies

Feeding SpaceX data into Grok extends a pattern Musk has run across his empire. Tesla supplies real-world driving data and manufacturing expertise, X supplies conversational data, and now SpaceX contributes aerospace engineering. Each company becomes a unique data source that sharpens the shared model — the same logic that has SpaceX aggressively expanding its compute business, from Colossus in Memphis to its server orders that lifted suppliers like Super Micro.

The buildout has accelerated since SpaceX's Nasdaq debut in June, when Grok and the Colossus supercomputer were folded into the investor pitch, followed by a $60 billion all-stock acquisition of coding startup Cursor that added a stream of real-world developer data.

Why It Matters

Musk said the two-trillion-parameter run, referred to elsewhere as Grok 4.6, is in its final training phase and expected to wrap this week. If SpaceX's engineering corpus lands as planned, it should sharply improve Grok's ability to reason about physical design, materials, and manufacturing — domains where general-purpose models still struggle.

As Teslarati reported, the move underscores why owning the companies that generate the data may prove as valuable as owning the compute that trains on it. For SpaceX and Grok, the combination is becoming a moat that rivals will find very hard to cross.