Broadcom (AVGO) on Solana

OpenAI and Broadcom have jointly unveiled "Jalapeño," a custom AI accelerator developed over nine months and purpose-built for large language model inference. The chip is architected to optimize memory movement, networking, and compute resources, with early testing indicating meaningfully better performance-per-watt than current alternatives. OpenAI applied its knowledge of LLM kernels and future model requirements to define the chip's architecture, and used AI models to accelerate the design process itself — marking a shift toward tighter integration between product development and custom silicon.

Broadcom is positioned as the manufacturing and semiconductor partner in the arrangement. Jalapeño is described as the first entry in a multi-generation compute roadmap, with gigawatt-scale deployment targeted for later in 2026. The partnership reinforces Broadcom's standing as a key infrastructure supplier in the custom AI silicon market, competing alongside other ASIC vendors for hyperscaler and AI-lab workloads.