Radar / AI infrastructure / Cerebras Disaggregated Inference
Cerebras Disaggregated Inference: GPUs prefill, wafer-scale decodes
Why it matters
Splits LLM inference across the best hardware for each phase instead of one chip doing both - a production deployment of prefill/decode disaggregation, with AWS, Fireworks AI, Gimlet Labs and General Compute as early-access partners.
What you could build with it
An inference provider could adopt the same prefill/decode split - cheap GPUs for prompt processing and a fast decoder for generation - to serve more tokens per megawatt from the same power budget, instead of scaling a single homogeneous fleet.
Does it hold up?
Promising but vendor-measured - the 5x throughput-per-megawatt figure comes from one shared GPT-OSS-120B endpoint against Cerebras's own earlier configuration, with no absolute numbers or independent measurement yet.
Built with Cerebras Disaggregated Inference
- Cerebras Puts Disaggregated Inference Into Production With GPUsunite-ai · Nov 17 report: production deployment, 5x throughput per megawatt, measured on a shared GPT-OSS-120B endpoint on Nov 14.
- Cerebras launches production disaggregated inference pairing GPUs with wafer-scaleletsdatascience · Nov 17 writeup noting the 5x claim is a ratio against Cerebras's own CS-3-only baseline, with no absolute throughput figures.
- Cerebras (CBRS) Q3 2026 Earnings Call Transcriptfool · Q3 earnings call (Nov 12) where Cerebras disclosed production-scale disaggregated inference and the AMD prefill / WSE decode split.
Learn more
First spotted on article: source.
More AI infrastructure
vLLM v0.34.0: extensible KV cache by default, Model Runner V2 only, and 10 new modelsThe open inference engine's latest release spans 1,061 commits from 412 contributors, adds 10 models including Mistral…infra · JEV 0.86llama.cpp v1.0.0: the local-AI engine's first stable releasellama.cpp and ggml hit v1.0.0, promising semantic versioning so breaking C API changes only arrive with major bumps.infra · JEV 0.84Graphify: turn any codebase into a knowledge graph your agent queriesA CLI plus agent skill that parses code with tree-sitter into a queryable knowledge graph of functions, classes…infra · JEV 0.77FreeLLMAPI: one free endpoint across 87 LLM providersA self-hosted router that puts 87 free LLM providers and 1,313 free model endpoints behind one OpenAI-compatible /v1…infra · JEV 0.73DeepSeek App API: the whole consumer app, now as developer APIsDeepSeek opened its app's capabilities through its API: web search, file parsing, URL reading, safety checks, voice…infra · JEV 0.67InferSSD: run 800B-class MoE models from an SSD on consumer hardwareWiCi's open-source inference engine streams MoE expert weights from NVMe so models larger than RAM or VRAM can run…infra · JEV 0.64
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.