Broadcom Just Quietly Became OpenAI’s Preferred Choice for AI Inference. Here’s What That Means for AVGO Stock.

In the second half of last year, OpenAI and Broadcom (AVGO) announced a deal for 10 gigawatts worth of compute capacity. Just nine months later, the chipmaker has unveiled a purpose-built chip that will run inference workloads for large language models (LLM). The chip, named Jalapeno, was designed by OpenAI and implemented on silicon by Broadcom, along with the networking. Celestica (CLS) handled the board, rack, and system integration. It was this collaboration that helped achieve targets in such a short span of time.

As AI inference takes center stage, Broadcom’s early entry into inference-specific chips is a positive development for shareholders. The chip is optimized around memory movement, kernels, and serving patterns that are unique to frontier LLM inference rather than general AI workloads. The reason this matters is that inference-specific chips can be fine-tuned to work efficiently with AI models, getting as close to theoretical limits as possible. The resulting higher throughput per chip lowers the costs in the long run, which eventually helps companies like OpenAI deliver their services at a lower cost than the competition.