The potential speed improvement vs JAX for large training runs is over an order of magnitude.
💡 Inside Track & Deep Insight
Elon Musk's tweet reveals SpaceX is developing a proprietary AI training stack written in C, optimized for a massive 220,000 GB300 GPUs connected via 800G NICs. This is a direct challenge to established frameworks like JAX, which are built on Python and often introduce overhead. By using C and pipeline parallelism at near-bare-metal level, SpaceX aims to achieve over 10x speedup for large training runs. This is not just about AI—it's about vertical integration in hardware-software stacks. If successful, it could disrupt the AI infrastructure market, potentially driving demand for custom silicon and networking solutions from companies like NVIDIA (GB300) and InfiniBand/Ethernet providers. For stocks, this could boost NVIDIA's data center narrative but pressure software AI companies that rely on JAX/TensorFlow. Crypto-wise, no direct impact, but broader AI compute efficiency could influence mining or DePIN projects. The move also signals SpaceX's seriousness about in-house AI for autonomous systems, from Starlink to Mars missions.
👇 Original Post on X
SpaceX has almost finished writing V1.0 of an in-house AI training stack in C that exact-maps to 220k GB300s with 800G NICs, making heavy use of pipeline parallelism and getting as close to bare metal as possible.
The potential speed improvement vs JAX for large training runs is…
— Elon Musk (@elonmusk) May 28, 2026