CloudWalk Bets on Inference Infrastructure
智东西 · wechat · 2026-07-19
This article introduces the AI inference infrastructure roadmap released by CloudWalk at WAIC 2026. The focus isn't on a single chip, but rather restructuring an entire system around inference workloads:
- Chip Layer: Introduced three high-compute cloud inference chips—DeepVerse100P, 100D, and 100L—optimized for different stages like long-context Prefill, Decode, and FFN.
- System Layer: Built a disaggregated inference infrastructure for 10,000-card heterogeneous clusters based on these chips, utilizing high-speed interconnects, resource pool partitioning, and scheduling to reduce tail latency and boost utilization.
- Software Layer: The IFWA software stack covers model adaptation, operator optimization, compilation deployment, and software-hardware co-design.
The article emphasizes that the focus of competition in the inference era has shifted from peak compute power to token generation efficiency and cost per token. CloudWalk also proposed a long-term goal of "one cent for 10 billion tokens" and launched the "1001 Initiative," aiming to foster full-link collaboration across chips, EDA, packaging, servers, models, applications, and research institutions.
More from Infra
- AI Infra Meme: While Countries Need TW-Scale Power for Datacenters, People Argue About Lightbulbs — mgostIH · 2026-10-27
- Legora runs 2,000 parallel agent evals on Modal Sandboxes for legal AI — blelbach · 2026-10-27
- Cloud release velocity scores: AWS 95, Azure 90, Google Cloud 82 — DavidLinthicum · 2026-10-27
- Selling a 5090 to scale local inference: one business weighs 5 hardware paths — Anarchaotic · 2026-10-27
- AWS tutorial: fleet-wide edge CV model lifecycle with SageMaker SDK v3, ~$0.07 per run — AWS ML Blog · 2026-10-27
- Qwen3.8-27B hits 40 tok/s on Jetson Thor with INT4 + MTP, accuracy beats FP16 — No-Dragonfly6246 · 2026-10-27