Chamath's warning goes viral: 'zero data retention' is no guarantee — run open models yourself
demian_ai · x · 2026-09-13
Trigger
On the All-In podcast, Chamath advised that sensitive workloads belong with a vendor you trust, naming Nebius as an example.
The argument
- Companies paste crown-jewel IP (protein designs, deal strategies) into shared chat APIs, but "zero data retention" usually means best efforts, not a hard guarantee.
- The real leak isn't names — it's solution patterns: a thumbs-up here, terms allowing training on "insights," or the next model version just "knowing" the shape of your private answer (the Navier-Stokes case).
The recommendation
Don't avoid AI; stop putting core work on shared frontier APIs. Use open-weight models on hardware you control, provisioned by vendors like Nebius, Fireworks, or AWS-class providers.
Boardroom angle
"We signed ZDR" looks thin next to "do we actually control where this runs" — and the CIO who took the easy API deal becomes the problem.
More from Infra
- Patched vLLM+FlashInfer Pushes Gemma 4 31B to 150 tok/s on a Single B300, Beating SGLang — abhijithneil · 2026-09-13
- MLX vs GGUF on Mac: which local model format and engine wins? — Ok_Warning2146 · 2026-09-13
- The Hugging Bay Brings Torrent Downloads to Open LLM Weights — csuwildcat · 2026-09-13
- Anthropic has lined up compute deals worth up to $517 billion, far above its $180 billion plan — citrini · 2026-09-13
- Google DeepMind team publishes deep dive on scaling Transformer inference — zainhas · 2026-09-13
- Open weights need orders-of-magnitude cheaper training or it's a LARP, researcher argues — jd_pressman · 2026-09-13