AMD and Cerebras Partner to Challenge Nvidia in AI Inference
AMD and Cerebras Systems have formed a strategic partnership to compete with Nvidia by offering a disaggregated AI inference solution for high-speed processing.
While Nvidia continues to dominate the AI model training market with its graphics processing units, Advanced Micro Devices (AMD) and Cerebras Systems are targeting the growing AI inference sector to capture market share. The two semiconductor firms have established a partnership to offer a disaggregated solution designed to lower costs and increase efficiency. In this arrangement, Cerebras manages the memory-intensive decode phase of inference using its wafer-sized chips, while AMD handles the prefill phase.
Cerebras claims its massive chip architecture provides inference speeds 30 times faster than traditional GPUs. To further its expansion into AI inference and robotics, AMD has acquired MEXT, Taalas, and World Labs. These strategic moves have already resulted in significant commercial traction, including $100 billion deals with Meta Platforms and OpenAI, as well as agreements with Anthropic and Microsoft.
OpenAI has also diversified its hardware dependencies by signing a separate $10 billion deal with Cerebras Systems. By focusing on niches such as physical AI, robotics, and ultrafast response times, AMD and Cerebras aim to erode Nvidia's infrastructure dominance in the mainstream AI market.
