SpaceX Engineering Data to Train Next-Gen 2-Trillion-Parameter Grok LLM
Elon Musk announced that SpaceX will contribute its non-ITAR engineering data to train the next-generation Grok LLM, which features approximately 2 trillion parameters. This model, informally referred to as Grok 4.6, is in its final training stages and is expected to complete training soon. This move highlights Musk's data synergy strategy across his companies, combining Tesla's real-world driving data, X's conversational data, and SpaceX's aerospace engineering data. It provides xAI with a unique source of physical-world engineering data, potentially giving Grok a competitive edge in scientific and industrial applications. To comply with U.S. ITAR regulations, sensitive technical data regarding rocket propulsion systems, guidance, and control will be excluded, while manufacturing, materials science, and Starlink hardware design data will be included. Additionally, SpaceX recently acquired AI coding startup Cursor for $60 billion to further bolster its AI training capabilities.
## BACKGROUND
ITAR (International Traffic in Arms Regulations) is a set of U.S. government regulations that control the export of defense and space-related technologies to safeguard national security. xAI, founded by Elon Musk, develops the Grok series of AI models and utilizes the Colossus supercomputer, one of the world's largest AI training clusters, to train its models.