Cerebras has announced that it will collaborate with all companies in the AI hardware sector, with the exception of NVIDIA.

      The strategy is characterized by a singular exclusion. During the Bloomberg Tech conference on Wednesday, Cerebras CEO Andrew Feldman stated that the AI chip manufacturer is collaborating with all major hardware producers in the industry, except for one: NVIDIA. This statement was not a grievance but rather a sales pitch.

      Feldman argues that key buyers seek an alternative to a single dominant provider, and that Cerebras can serve as the connecting element for those assembling an AI stack from various components. Whether customers are utilizing AMD, cloud services, custom silicon, open models, or specialized accelerators, Cerebras aims to be part of the dialogue.

      The company is positioning itself as the choice for buyers who believe relying on a single vendor for the most costly segment of their budget is a risk. This framing is carefully crafted in what it omits. Feldman does not claim that NVIDIA is weak; NVIDIA continues to be the leading supplier of GPUs that train the majority of the industry’s models.

      The argument is that concentration is a liability, and that cloud providers, model laboratories, and enterprise buyers are looking for a reliable alternative source, rather than a total replacement—an apprehension heightened by NVIDIA’s $40 billion in AI equity investments tightening its hold on the market. Cerebras is marketing itself as a hedge, not a substitute.

      Cerebras is moving to collaborate with a diverse array of AI data-center component suppliers, which opens the door to more partnerships like the existing one with Amazon. This “everyone but NVIDIA” approach acts as a market strategy: design the playing field as NVIDIA versus not-NVIDIA, and position Cerebras as the organizing partner for the latter group.

      The arrangement with Amazon serves as the model Cerebras seeks to replicate. By engaging with a broad set of data-center component suppliers instead of limiting itself to one platform, the company aims to make itself an essential component in stacks created from various vendors, serving as the inference engine for those who choose not to standardize on NVIDIA.

      Each new supplier relationship expands the configurations that Cerebras can accommodate, providing the practical basis for their rhetoric.

      Cerebras is well-positioned to emphasize this point, having gone public last month in the largest US tech IPO in years, granting it both capital and the visibility that reassures enterprise buyers. A newly listed chipmaker with a clear narrative against NVIDIA presents a more credible counterbalance than a private one making the same assertion.

      The effectiveness of this strategy in generating sustainable revenue, however, remains uncertain. Positioning as an alternative to NVIDIA is a crowded endeavor, shared by AMD and numerous custom-silicon initiatives, and willingness to collaborate does not equate to securing the workloads.

      What Feldman conveyed in San Francisco is the narrative Cerebras intends to promote. It will be up to customers to determine if not-NVIDIA represents a genuine product or merely a strategic stance.