A new wave of artificial intelligence startups is turning to Amazon's custom-designed chips to power their work, according to a report published by About Amazon, the company's own news site.
The chip at the center of the story is Trainium, Amazon Web Services' purpose-built processor for training AI models. According to About Amazon, a growing group of startups is choosing Trainium specifically to train models that simulate the physical world — a category of AI that reaches beyond the text-based chatbots that have dominated headlines.
The report notes that this interest extends past Amazon's best-known AI partners. Anthropic and OpenAI are named as established players, but About Amazon frames the trend around a separate, emerging cohort of companies opting for Trainium as they build these world-simulating systems.
It is worth being clear about the source: this account comes from Amazon's corporate newsroom rather than an independent outlet, so it reflects how Amazon describes the adoption of its own hardware. The sources provided do not include specific figures, named startups beyond Anthropic and OpenAI, or direct comparisons of performance or price.
Why it matters: the chips used to train AI are among the most contested resources in the technology industry, and any sign that startups are picking a custom alternative rather than relying solely on industry-standard processors signals how competition over AI infrastructure — and who controls it — is taking shape.