Nvidia Corporation and Cerebras Systems Inc. Forge AI Inference Partnerships

Major AI hardware providers are forming strategic alliances to resolve infrastructure bottlenecks related to memory bandwidth and inference latency. Nvidia Corporation signed the largest memory deal in history with SK Hynix Inc. to co-develop future generations of AI memory and secure high bandwidth memory (HBM) supplies. As part of this agreement, SK Telecom will build a 2-gigawatt AI cloud data center in South Korea using the Nvidia Vera Rubin Platform. This follows Nvidia Corporation's acquisition of Groq and its language processing units to further optimize AI inference phases.

Simultaneously, Cerebras Systems Inc. partnered with Advanced Micro Devices, Inc. to create a disaggregated AI inference solution. The architecture combines the Helios rack-scale architecture from Advanced Micro Devices, Inc. for compute-intensive pre-fill with the Cerebras Wafer-Scale Engine for low-latency decode. Cerebras Systems Inc. Chief Marketing Officer Julie Choi stated this combination delivers five times higher tokens per second per watt than existing solutions. Cerebras Systems Inc. plans to deploy Helios systems within its own data centers by the end of 2026 to support high-demand workloads such as real-time voice generation and agentic coding.