Nvidia has a moat but not control: Google's TPUs now hold a quarter of global AI compute
AryHHAry · x · 2026-10-06
A long-form analysis argues Nvidia's dominance in AI training (80-90% revenue share, but H100-equivalent share down from 90% in early 2024 to 70% by Sep 2025) is a moat, not control. Google owns a quarter of global AI compute via its own TPUs; custom ASICs surge in inference; Huawei Ascend claims domestic lead but stays under 4% of Nvidia's global output by 2026. The result: a broken supply symmetry, not a monopoly.
More from Infra
- Long-running benchmarks find Strata inference server failing full-build scenarios — julianharris · 2026-10-06
- AMD R9700 owner: ROCm on Windows cripples local i2v, Vulkan runs fine — vladomkd · 2026-10-06
- CutBCE: TPU Kernel Eliminates OOM in Large-Vocabulary Recommendation Training, 91.9% Faster — _reachsumit · 2026-10-06
- Domain specialists orchestrated by a general model: a local-LLM architecture pitch for 8-16GB GPUs — CyberExplore · 2026-10-06
- Run a 37GB Qwen MoE in a browser tab: LocalMind streams expert weights from disk, matches llama.cpp output — naklitechie · 2026-10-06
- Local LLM reality check: 4090 caps Qwen context at ~32k before VRAM runs out — BLUECOW009 · 2026-10-06