AMD Teams Up With Cerebras to Challenge Nvidia’s Grip on AI Inference

(WS News) – AMD has struck a partnership with chip startup Cerebras to build a new AI inference system called Helios, a move aimed squarely at challenging Nvidia’s dominance in the data center accelerator market.

The announcement came from AMD chief executive Lisa Su at the company’s Advancing AI 2026 event, where AMD unveiled a broad refresh of its compute lineup, including sixth-generation EPYC CPUs, Instinct MI400 series GPUs, and the new Helios rackscale platform, which is already in production. According to AMD, Helios delivers up to 30 percent more inference tokens per dollar than competing systems, a pitch designed to appeal to companies racing to control the soaring cost of running large-scale AI models.

The Cerebras partnership pairs Helios with Cerebras’ wafer-scale engine chips to build what the companies describe as a workload-optimized approach to ultra-low-latency AI inference, using different hardware for different stages of AI reasoning. The joint system is expected to be installed in Cerebras data centers later this year. AMD said its expanding silicon push, spanning data centers, PCs, edge devices, and embedded processors, is targeting a total addressable market of roughly $2 trillion by 2030.

Beyond the data center, AMD also introduced new hardware for physical AI and robotics, including the Ryzen AI Embedded X100 processors and the Kria AI robotics developer platform, alongside an open robotics partner network that includes companies such as Analog Devices, SICK, and Voyant Photonics. AMD executives framed the open approach as a deliberate contrast to closed ecosystems, saying no single company should control the future of physical AI. Anthropic, OpenAI, Meta, Cerebras, AT&T, and Cisco were among the companies AMD said are already collaborating on its AI infrastructure push.