# AMD and Cerebras Pair Helios With the Wafer-Scale Engine for Ultra-Low-Latency AI Inference

**AMD and Cerebras say combining AMD Helios with the Cerebras Wafer-Scale Engine can deliver up to 5x more tokens per second per watt for AI inference.**

## What was announced

AMD and Cerebras Systems announced a technical partnership at AMD's Advancing AI 2026 event to build a disaggregated AI inference solution that combines AMD Helios rackscale systems with the Cerebras Wafer-Scale Engine. The two companies say the goal is to deliver the ultra-low latency that the most demanding AI applications need while sharply increasing throughput and efficiency. [AMD Newsroom](https://newsroom.amd.com/news/aai-2026-cerebras-inference/) described the pairing as a way to give each stage of an inference workload a chip built specifically for it.

## How the two systems split the work

Rather than running one chip for an entire AI request, the joint design separates inference into two stages and hands each to a different engine. AMD Helios processes prompts and large context windows, acting as a high-throughput engine. The Cerebras Wafer-Scale Engine takes over the memory-bandwidth-heavy work of generating tokens one at a time, aiming for the fastest possible response once generation begins. AMD and Cerebras say connecting the two engines in a single workflow lets them target ultra-low-latency use cases without giving up throughput or scale.

## The performance claim

AMD and Cerebras say the combined system is expected to deliver up to 5x higher tokens per second per watt than using either engine alone. Dr. Lisa Su, AMD's chair and CEO, said the partnership extends Helios's performance into the most latency-sensitive applications and creates a platform for real-time agentic AI. Andrew Feldman, Cerebras's CEO and co-founder, said pairing with AMD gives Cerebras a chance to bring its inference speed to more customers. These figures come from the companies themselves and have not been independently verified.

## Why this matters for inference workloads

AI inference needs differ by task: high-volume workloads want maximum token generation, while coding assistants, live agents and other agentic workflows need fast turnaround on each response. AMD and Cerebras frame their joint solution as infrastructure built for that split, with Helios as the foundation for high-throughput and balanced workloads and the Wafer-Scale Engine reserved for the most latency-sensitive segment.

## Key facts

- AMD and Cerebras announced the partnership at AMD's Advancing AI 2026 event on July 23, 2026.
- AMD Helios handles prompt processing and large context windows; the Cerebras Wafer-Scale Engine handles token generation.
- The companies say the combination can deliver up to 5x higher tokens per second per watt.
- Cerebras plans to deploy AMD Helios systems in its own data centers.
- The joint solution is expected to become available first through Cerebras Cloud in the second half of 2026.

## What happens next

Cerebras plans to deploy AMD Helios racks in its data centers, and the companies say the joint solution will become available first through Cerebras Cloud in the second half of 2026.

## Check your understanding

- Which of the two systems is designed to handle prompt processing and large context windows?
- Where do AMD and Cerebras say the joint solution will first become available?

## Sources

- [AMD Newsroom: AMD and Cerebras Announce Ultra-Low-Latency AI Inference Solution](https://newsroom.amd.com/news/aai-2026-cerebras-inference/)

---

- Category: know (announcement)
- Affects: —
- Published: 2026-07-26T07:31:45.633Z
- Original: https://newsroom.amd.com/news/aai-2026-cerebras-inference/
- Generated by tinker news