OpenAI has introduced its first custom inference chip, named Jalapeño, which reportedly outperforms Nvidia's latest Blackwell and Rubin chips in both throughput and energy efficiency. This development was highlighted during the Hot Chips conference, marking a significant milestone in AI hardware innovation.
According to The Decoder, tests conducted by SemiAnalysis show that Jalapeño leads Nvidia's offerings in key performance metrics. SemiAnalysis CEO Dylan Patel noted that first-generation chips typically struggle to compete, yet OpenAI's Jalapeño has surpassed Nvidia’s Blackwell and even Rubin models.
This advancement is particularly relevant for Japanese markets, where efficient AI processing hardware can influence the competitiveness of technology firms and impact equities in the semiconductor and AI sectors.