Amazon Web Services announced Friday a new partnership with artificial intelligence chip manufacturer Cerebras Systems to enhance AI-powered applications including chatbots and programming assistance tools.
The collaboration brings together the $23.1 billion chip company with Amazon’s cloud computing division. Cerebras has positioned itself as a competitor to Nvidia by developing AI processors that operate without the costly high-bandwidth memory required by Nvidia’s leading chips. The startup recently secured a massive $10 billion contract to provide processors to OpenAI, the company behind ChatGPT.
The new arrangement will place Cerebras processors within Amazon Web Services data centers, where they’ll work alongside Amazon’s proprietary Trainium3 AI chips through specialized networking infrastructure.
“Every customer large or small is on AWS, from individual developers to the largest banks in the world,” Cerebras CEO Andrew Feldman explained to Reuters, adding that the partnership will “make it easy as a click to get on Cerebras.”
Neither company revealed the financial value of their agreement.
The partnership focuses on improving “inference” operations, where trained AI systems process user requests and generate responses. Amazon and Cerebras plan to divide this process into two distinct phases: “prefill,” which converts human language into computer tokens, and “decode,” where the AI generates the final answer.
Amazon’s Trainium3 processors will manage the prefill phase, while Cerebras chips will handle decoding operations in what Feldman described as a “divide and conquer strategy.”
This approach mirrors expectations for Nvidia’s upcoming announcement next week, where the company is anticipated to reveal how it will integrate its graphics processing units with chips from Groq, a startup Nvidia acquired for $17 billion in December.
Amazon expressed confidence that its service will provide superior value compared to Nvidia’s forthcoming offering, though detailed comparisons aren’t yet possible.
“The timeline for that (Nvidia-Groq) pairing remains unclear while our Trainium3 program is just months away from running production workloads,” Amazon stated. “What we can say is that we believe (Trainium3)—and future (Trainium4)—will continue to lead in price-performance versus merchant GPUs.”
The Amazon-Cerebras service is expected to launch during the second half of this year.
