Anthropic models reportedly optimize kernels for NVIDIA, TPU, Trainium — but not Apple
GabGarrett · x · 2026-11-16
A claim circulating that Anthropic's models are internally used to optimize kernels on NVIDIA GPUs, Google TPUs, AWS Trainium, and AMD GPUs — but not Apple hardware. The stated reason: optimizing Apple silicon only benefits local and open-source inference, which offers Anthropic no strategic advantage. Unverified, but points to labs using their own models to auto-optimize low-level kernels.
More from Infra
- Who absorbs failed AI datacenter deals? Microsoft took Abilene's 900MW in three weeks — tengyanAI · 2026-11-16
- Microsoft open-sources TokenEconomizer, cutting coding agent costs 18–31% — aigclink · 2026-11-16
- EE asks Claude to price 1,000 human brains in silicon: 39 sq-mile data center, 27x global nuclear output — Recent-Day3062 · 2026-11-16
- Intel 170hx cards unlocked to 80GB VRAM, price jumps from hundreds to $4k — GabGarrett · 2026-11-16
- Track cost per finished task, not token price: the agent metric that matters — arpit_bhayani · 2026-11-16
- Nebius and CoreWeave already used customer prepayments to secure GPUs — context WSJ missed — firstadopter · 2026-11-16