🧠 Food for thought

Implications, context, and why it matters.

Musk’s ‘at scale’ ambitions involve building a million-GPU supercomputer

• Elon Musk described a large hardware spend, since his AI company xAI already runs the Colossus supercomputer with over 100,000 Nvidia H100 GPUs 1.
• xAI plans to grow that cluster past one million GPUs, which would put it among the world’s most powerful supercomputers, though Musk gave no timeline 1.
• Musk said training Grok 3 and later models will need 100,000 Nvidia H100 GPUs 2.
• xAI has raised $11 billion this year and added another $5 billion recently, as it races rivals such as OpenAI, which works with Microsoft for computing capacity 1.

Tesla’s chip push could make it a future Nvidia competitor

• Tesla and xAI buy large amounts of Nvidia hardware, while Tesla also builds its own custom AI processors 3.
• Musk set a 9-month design cycle for these chips, aiming to outpace the yearly rhythm he attributes to Nvidia and AMD 4.
• The approach uses Nvidia GPUs for today’s AI training, while Tesla develops its own silicon roadmap 2.
• Musk said Tesla’s in-house chips could become “the highest volume AI chips in the world,” with volume driven mainly by use in vehicles and possibly robots rather than data-center accelerators 4.