Broadcom and OpenAI Unveil Jalapeño AI Chip with 50% Cost Savings, 10GW Deployment
AVGO•OpenAI and Broadcom completed the fastest high-performance semiconductor development cycle ever, delivering a custom AI inference chip called Jalapeño in nine months that promises roughly 50% lower costs than standard AI GPUs. They plan to deploy 10 gigawatts of these accelerators with Microsoft by 2029.
1. Chip Launch and Partnership
OpenAI and Broadcom announced the Jalapeño chip, marking their first joint custom inference accelerator built for AI workloads. The design-to-blueprint cycle completed in nine months reflects the fastest high-performance semiconductor development timeline to date.
2. Performance and Cost Efficiency
Early tests indicate Jalapeño delivers approximately 50% cost savings compared to standard AI GPUs when running inference workloads. The chip's architecture is optimized specifically for large language model inference rather than training.
3. Deployment Scale and Timeline
Broadcom and OpenAI plan to deploy 10 gigawatts of Jalapeño accelerators through Microsoft's data centers by 2029, representing one of the largest custom chip rollouts in data-center history. Commercial deployment is slated to begin by year-end.
4. Implications and Risks
Broadcom’s custom AI chip business, which has driven share gains since 2022, stands to benefit significantly from adding OpenAI as a client. Risks include scaling from lab prototypes to large-scale racks, though OpenAI has backup supply agreements with AMD, Cerebras and Amazon.
