AI chip competition intensifies as OpenAI and Samsung challenge Nvidia

The semiconductor landscape is undergoing significant shifts as major players develop custom AI hardware to challenge Nvidia’s dominance. OpenAI has unveiled ‘Jalapeño’, a custom AI inference chip developed with Broadcom and TSMC. Technical assessments indicate the chip offers approximately 50% lower cost per token compared to Nvidia’s GPUs and demonstrates superior performance in terms of processing per watt and end-to-end latency.

In a separate strategic move, Samsung Electronics and Arm have partnered to develop specialized system-on-chips (SoCs) for on-device AI using Samsung’s 2nm manufacturing process. This collaboration aims to reduce cloud dependency and latency for terminal devices like smartphones. Industry speculation suggests OpenAI could be a potential end customer for these advanced chips.

Nvidia continues to see massive market valuation, with analysts predicting its market cap could reach $10 trillion due to sustained AI demand. However, the company faces growing competition. While Nvidia acts as an ‘AI bank’ by providing financial support to help customers purchase GPUs, critics raise concerns about potential bubbles. Meanwhile, Nvidia is expanding its influence in the ASIC market through its NVLink Fusion ecosystem, partnering with companies like MediaTek to integrate its interconnect technology into custom AI chips, directly challenging competitors like Broadcom and Marvell.