Lenore unveils hybrid Token factory in Zhangjiakou: 15,000P cluster, 2 quadrillion tokens a year
量子位 · wechat · 2026-11-14
Lenovo launched its "Lenovo Wanquan Hybrid Token Factory" brand in Zhangjiakou, Hebei, signing with the local government for the first mother factory. Phase one: a 15,000P heterogeneous compute cluster and 100MW AI data center, targeting 2 quadrillion tokens/year (5.5 trillion/day) and a 200M tokens/sec peak, with scale-up 10x+ planned by 2030.
Context: China's daily token usage grew 1000x in two years, surpassing 500 trillion by June 2026, yet enterprises struggle with costs — one mid-size bank spent 9M yuan on tokens in H1 (8% of its annual IT budget), and McKinsey finds 20% of organizations cap AI use over cost.
Key ideas
- "Green power in, tokens out": hybrid supply mixes private/public tokens, edge-cloud collaboration, and open/closed models per task.
- TokenEngine handles dynamic scheduling and model routing; xCloud unifies private/public cloud management, TokenHub routes inference requests.
- Three security layers: model whitelist, TokenTrust transmission filtering, and privacy computing/TEE.
- Three-level cost optimization: 30%+ compute efficiency gains, all-liquid-cooling modular data centers (PUE<1.1), plus business-level token effectiveness; targets include 70% cost-per-million-token reduction and >99.9% "yield."
Deployments include an AI short-drama token factory with Hunan Broadcasting (targeting 50% cost cut, 1200+ hours of finished content per month) and an education project covering 600+ schools.
More from Infra
- Running DeepSeek 4.1 Flash on a 32GB Mac via SSD streaming with open-source Local Code — alexcovo_eth · 2026-11-14
- SGLang Summit talk details battle-tested NVFP4 4-bit RL across thousands of GPUs — ericzelikman · 2026-11-14
- Will local models keep up with frontier? Users debate a hardware-bound gap — AshAtlasDev · 2026-11-14
- Anthropic IPO roadshow slide leak sparks Blackwell training speculation — and a block — teortaxesTex · 2026-11-14
- New Yorker interview: AI is a brute-force problem that will 'cook the planet' — kevinnbass · 2026-11-14
- LexiSuite: a Rust-built inference engine beats llama.cpp 1.0.0 at 262k context — W61k3r · 2026-11-14