Amazon intends to integrate chips from the startup Cerebras Systems Inc. alongside its proprietary Trainium processors, a move the company claims will establish optimal ground for deploying large language models. Amazon Web Services (AWS), a leading provider of cloud computing capacity, is scheduled to introduce this novel service, stemming from its arrangements with Cerebras, in the latter half of 2026. Details regarding the financial aspects of this agreement have not been publicized.
This collaboration between Amazon and Cerebras represents another effort to address the massive demand for artificial intelligence (AI) computing infrastructure. Nafea Bshara, a Vice President at AWS, indicated that both organizations have been preparing for this partnership over several years. Furthermore, he asserted that the AWS platform is prepared to incorporate as many of these chips as necessary to scale up its computational resources based on client needs.
For Cerebras, which is preparing for an Initial Public Offering (IPO), securing Amazon as a customer is set to boost the company’s profile within a potentially vast market. AWS stands out as the first major data center operator to commit to employing Cerebras chips within its infrastructure stack.
Current information suggests that the processors from Amazon and Cerebras will function synergistically, particularly devoted to inference computation—that is, running large language models and formulating responses to incoming queries. Specifically, Amazon’s Trainium 3 chips will be utilized for interpreting user requests, followed by the Cerebras Wafer Scale Engine chips taking over to generate the actual answers. While this dual-component strategy often suffers from speed degradation due to inter-component communication, the companies aim to leverage the specialized nature of these chips to accelerate inference tasks. Performance gains are anticipated to be most apparent in user-facing applications, such as iterative software code generation.
Bshara noted, “While the service based purely on our Trainium chips might be more cost-effective, this new combined offering will be compelling where speed translates directly into value.” Amazon maintains its status as a major Nvidia client while simultaneously furthering the development of its custom AI silicon. These combined initiatives are intended to enhance the economic efficiencies of Amazon’s data centers and provide the capability to deliver distinctive service experiences to its clientele.
About the Author
