Huawei Launches Agentic AI Infra Stack, Rivals Nvidia
Summary
Huawei Cloud has launched an "Agentic Infra" stack. This is a full set of compute, storage, and networking products designed for large-scale AI agents. What's interesting is this looks like Huawei's direct bid to compete with Nvidia-anchored AI infrastructure. At its Inspire event, Huawei unveiled AICS, an AI Cluster Service. It can support clusters with 100,000 cards. This platform offers five million tokens per second throughput across 1,000 cards, with total computing power up to 200 exascale floating-point operations per second. The stack also includes AMS, a storage solution for memory expansion and reduced inference costs. CCE VolcanoNext is a scheduler that claims over 30% better resource utilization. Plus, there's AgentSphere, a secure sandbox for spinning up hundreds of thousands of agent instances per minute. This move comes as China aims to build domestic alternatives following U.S. import bans. Huawei is using its own scaling principle, Tau, for semiconductor design. This allows for improving chip designs without shrinking transistors. The company has already used this concept for 381 chips. The bottom line is this new stack allows enterprises to build and run AI agents on homegrown chips, potentially rivaling American competitors.
This is an AI-generated audio summary. Always check the original source for complete reporting.