MediaTek may take Google's 1M training TPU business as Broadcom keeps inference
firstadopter · x · 2026-09-03
- Citing Hot Chips details: Broadcom said on its earnings call it handles TPU v8 inference, while MediaTek is doing the TPU training chip.
- The author speculates the "one million training TPUs" Google mentioned at Hot Chips may go to MediaTek rather than Broadcom — a significant blow to Broadcom.
- Also notes TPU SRAM details are worth attention from players like Cerebras.
More from Infra
- Dragonfly rethinks Redis with sharded multi-threading to scale across modern multi-core servers — techNmak · 2026-09-03
- Zeiss exec: China is about 15 years behind in cutting-edge chipmaking tools — broodsugar · 2026-09-03
- Engineer tried 70B prefill + 7B decode back in 2023: combining spec dec with disagg was useless — suchenzang · 2026-09-03
- Blogger argues Zeiss's mirror-coating know-how is only ~30 years deep, not magic — teortaxesTex · 2026-09-03
- Tenstorrent and AI & Inc launch JapanFold: free inference for open-source drug discovery models — DavidBennett__ · 2026-09-03
- Voice as default input: local STT with Mac Parakeet, Whisper vs Parakeet tradeoffs — LeatherRub7248 · 2026-09-03