News / Korea / yonhap_finance
OpenAI to Deploy Proprietary 'Jalapeno' AI Chip This Year, Likely Featuring Samsung HBM4
Electronic Technology · Semiconductors · yonhap_finance · 2026-08-26
OpenAI plans to launch its custom 'Jalapeno' AI inference chip this year, which reportedly outperforms Nvidia hardware and likely utilizes Samsung's HBM4 memory.
What Happened
Proprietary Chip Development: OpenAI has announced that its custom-developed AI inference chip, codenamed 'Jalapeno,' has outperformed Nvidia's GB300 in internal performance benchmarks. Richard Ho, OpenAI's head of hardware, confirmed that the chip excels in both power efficiency and response speed, with plans to integrate it into production workloads before the end of the year.
Technical Specifications: Developed in partnership with Broadcom, the Jalapeno chip is designed to deliver high performance at a low 700-watt power draw, significantly reducing data center operational costs. Industry reports from SemiAnalysis suggest that the chip utilizes Samsung Electronics' HBM4 memory, marking a significant potential win for the Korean semiconductor giant.
Strategic Focus: Unlike Nvidia's flagship hardware, the Jalapeno is specifically optimized for inference tasks rather than model training. OpenAI is already advancing its roadmap, with the second generation of the chip nearing the 'tape-out' phase and conceptual design for a third generation already underway to further lower infrastructure costs.
Market Implications: While OpenAI intends to maintain its strong partnership with Nvidia, the move toward internal hardware reflects a broader strategy to diversify supply chains and manage the massive computational demands of AI. The industry is closely watching how this shift will impact the competitive landscape for AI hardware and the future demand for high-bandwidth memory suppliers.