Skip to main content

AWS and Cerebras Partner to Revolutionize AI Inference Speed on Amazon Bedrock

23 MARCH 2026·9 MIN READ·14 SOURCES

Amazon Web Services has teamed up with Cerebras Systems to deploy a groundbreaking AI inference architecture that promises up to tenfold speed improvements in generating AI model outputs, transforming real-time applications across industries.

AWS and Cerebras Partner to Revolutionize AI Inference Speed on Amazon Bedrock

Key takeaways · 5

  • 01

    AWS and Cerebras introduce disaggregated inference architecture to accelerate AI responses.

  • 02

    Cerebras’ wafer-scale WSE-3 chip offers 900,000 cores and 27 PB/s on-chip memory bandwidth, dwarfing traditional GPUs.

  • 03

    Prefill stage runs on AWS Trainium processors; decode stage runs on Cerebras WSE-3 chip, optimizing for distinct workload needs.

  • 04

    Integration into Amazon Bedrock enables developers to access faster inference with no code changes.

  • 05

    Expected launch in H2 2026 with capabilities benefiting enterprises deploying AI agents, real-time voice AI, and agentic workflow automation.

AWS has partnered with AI chipmaker Cerebras to deliver an unprecedented inference speed boost by integrating Cerebras’ wafer-scale WSE-3 chips within Amazon Bedrock’s managed infrastructure. This collaboration represents the first hyperscaler deployment of wafer-scale AI chips in the cloud.

Disaggregated inference splits the AI model inference process into 'prefill' and 'decode' stages, assigning each to a processor optimized for its workload. AWS's Trainium chips manage the parallel, compute-intensive prefill, while Cerebras’s WSE-3 chip handles the sequential, memory-intensive decode phase, significantly reducing inference latency.

Unlike traditional GPUs composed of multiple small chips, Cerebras designs a single monolithic wafer-scale chip integrating 900,000 AI cores and 44 GB of on-chip SRAM. With 27 petabytes per second of memory bandwidth, WSE-3 outperforms GPUs by a large margin, enabling rapid token generation that exceeds 3,000 tokens per second.

The enhanced inference speed directly benefits workloads such as real-time AI coding assistants, conversational agents, and voice AI, where low latency is critical. Faster inference translates into better user experiences, increased productivity, and new possibilities for agentic automation.

The AWS-Cerebras partnership signals a significant advance in AI inference infrastructure, addressing critical latency bottlenecks in deploying large language models at scale. By leveraging specialized hardware for distinct inference phases, it enables enterprises to run AI applications faster and more efficiently, enhancing user experience and opening new frontiers in real-time AI. This development could reset industry performance standards and influence future cloud AI hardware strategies.

Why it matters
Daily session

Put this to work — one session a day, built for your industry.

Create a free account for a daily session — eight questions and one real-work challenge, on the news that affects your role.

Start free

Sources

AI fluency, one session a day, built for your work.

Get the appStart free