OpenAI, Broadcom (AVGO) Unveil “Jalapeño” AI Accelerator for Enhanced LLM Inference

OpenAI and Broadcom jointly announced 'Jalapeño,' a custom AI inference accelerator designed specifically to handle the compute demands of large language model inference at scale. The chip represents OpenAI's first disclosed custom silicon effort built with an outside foundry partner, positioning Broadcom's custom ASIC unit (formerly CA Technologies ASIC group) as the design and manufacturing orchestration engine behind it. No silicon volumes or commercial timeline have been publicly disclosed yet.

For Broadcom, the announcement matters because custom AI ASICs are the fastest-growing segment of its already strong AI revenue mix. AVGO posted $63.9B in revenue for FY2025 with 23.9% YoY growth and a 67.8% gross margin — metrics that reflect the high-value nature of custom silicon contracts versus commodity networking gear. OpenAI's scale of inference demand is enormous, which would translate to a material recurring revenue stream if volumes ramp.

The bull case rests on Broadcom becoming the default custom silicon partner for the largest AI labs — Google, Meta, and now OpenAI — creating a near-oligopoly position in a market that is structurally shifting away from merchant GPUs toward purpose-built inference chips. Every incremental hyperscaler win compounds the revenue visibility.

The bear case is that no volume or revenue guidance has been attached to this announcement, making it a headline event rather than a confirmed earnings catalyst. NVIDIA's inference roadmap (Blackwell, Rubin) remains highly competitive, and Broadcom's valuation already prices in significant AI upside after a strong run. The setup is real but the timeline to revenue contribution is unclear.

The key watchpoints are: any formal volume or revenue disclosure from either company, AVGO's next earnings call commentary on AI ASIC pipeline, and whether NVIDIA responds with inference-specific pricing or product positioning.