Broadcom and OpenAI heat up AI chip market with inference processor Jalapeño - MSN

Broadcom and OpenAI have unveiled Jalapeño, a new AI chip specifically designed for large language model (LLM) inference, intensifying competition in the AI chip market. Developed through a close coll... Read original →

Broadcom and OpenAI’s Jalapeño marks a strategic pivot from AI training to inference optimization. Technically, its 3nm EUV design with co-optimized memory and networking will pressure TSMC to accelerate CoWoS capacity and spur upgrades in data center optical interconnects and Ethernet standards. Geopolitically, tightening U.S.-EU export controls on advanced computing may compel Celestica to establish redundant assembly lines in Mexico or Vietnam, raising BOM costs by 10–15%. NVIDIA will likely counter with enhanced Blackwell Ultra software stacks, while AMD doubles down on MI400 integration with Microsoft Azure. Within 18 months, custom inference ASICs will become standard for hyperscalers, eroding GPU pricing power and shifting semiconductor value toward vertically integrated hardware-model co-design.