OpenAI’s 700W Jalapeño ASIC outpaces 1,400W Nvidia flagship GPU

OpenAI unveiled its first in-house chip, Jalapeño, co-developed with Broadcom, at Hot Chips, demonstrating superior performance over NVIDIA's GB300 GPU. The 700W chip achieved 1.5 to 1.9 times higher ... Read original →

OpenAI’s unveiling of the Jalapeño chip marks a pivotal shift in AI hardware dynamics. Built on TSMC’s 3nm process, it outperforms NVIDIA’s GB300 GPU in throughput per watt by up to 1.9x and latency by up to 3.6x, demonstrating strong inference capabilities. This breakthrough challenges NVIDIA’s dominance in data center AI chips and accelerates demand for HBM4 memory, benefiting suppliers like SK hynix and Micron. Despite a $105B data center deal with NVIDIA, OpenAI’s move signals a strategic pivot toward in-house chip development, potentially inspiring other AI players to follow suit. Over the next 12–24 months, the adoption of ASICs in inference workloads will accelerate, forcing NVIDIA to expedite Blackwell advancements. Meanwhile, geopolitical tensions around semiconductor supply chains intensify, especially amid U.S.-China tech decoupling, where performance-cost trade-offs will define competitive advantage.