DeepSeek bottleneck is compute, not demand: 1GW cluster could train a far stronger model, says leak account
teortaxesTex · x · 2026-10-21
AI leak account teortaxesTex argues the claim that DeepSeek lacks demand is wrong — demand is effectively infinite and the bottleneck is compute.
Key claims (unverified):
- With a 1GW cluster, DeepSeek could train a vastly stronger model than V4.1F-1017
- Even at the same footprint, it could charge far more than the current $0.003/$0.15/$0.6 pricing
- More compute would enable long-range autonomy
- $200B/GW/year in revenue is plausible
Image context: the screenshot shows Teortaxes describing DeepSeek's DSec sandbox platform as a "Software Machine" — 380K concurrent agents coordinating in shared VMs (already done in DSH) could tackle one big problem like "AUDIT ALL LINUX". The figure shows DSec's architecture: libdsec Python SDK, cluster services (IAM, API server, placement engine, watcher) and Container/MicroVM/FullVM sandbox backends on 3FS storage with AppArmor+eBPF isolation.
More from Infra
- How a 30B model fits on a 32GB GPU: quantization plus speculative decoding — TimDarcet · 2026-10-21
- PyTorch details KTIR and Dataflow Scheduler behind IBM's Spyre accelerator backend — PyTorch · 2026-10-21
- Indie dev replaces $1k OpenAI bill with a YAML pipeline on spot GPUs and vLLM — cmm324 · 2026-10-21
- DeepSeek open-sources DeepInfer, the vLLM-based engine behind V4.1-Flash — The Decoder · 2026-10-21
- Robotics startup launches Deployments to serve fine-tuned HF checkpoints to robots in real time — arnie_hacker · 2026-10-21
- ByteDance, Alibaba, Tencent order 3M+ Nvidia RTX PRO 5500 cards worth $35B, Reuters reports — SumitGup · 2026-10-21