Key Highlights
- OpenAI and Broadcom have introduced Jalapeño, a custom-designed AI chip optimized for inference tasks.
- Preliminary benchmarks indicate approximately 50% cost reductions versus conventional AI GPUs, according to Broadcom CEO Hock Tan.
- Development required nine months; TSMC handles production, with first deployments scheduled for late 2026.
- Broadcom shares jumped approximately 2% post-announcement; Nvidia experienced a minor decline.
- A multi-generation development plan exists, with the successor chip slated for 2028.
On Wednesday, OpenAI and Broadcom revealed Jalapeño, a collaboratively engineered custom AI chip. Broadcom (AVGO) stock gained roughly 2% following the disclosure, whereas Nvidia (NVDA) declined 0.26%.
The processor targets AI inference operations specifically—the computational process behind generating outputs in applications such as ChatGPT. Classified as an ASIC (application-specific integrated circuit), it sacrifices versatility to deliver reduced costs and customization for particular computing tasks.
Broadcom CEO Hock Tan revealed that preliminary samples demonstrate cost reductions approaching 50% relative to standard AI GPUs. This figure carries significant weight in a sector where computational expenses represent a persistent challenge.
According to OpenAI, prototype chips are currently operational in its facilities, achieving intended power consumption and performance benchmarks when evaluated against the GPT-5.3-Codex-Spark model.
OpenAI’s engineering team finalized the chip architecture in approximately nine months, leveraging AI-assisted design tools. Manufacturing responsibilities went to TSMC.
Richard Ho, OpenAI’s hardware chief, characterized Jalapeño as “a very general purpose device” optimized for large language models yet engineered to “address future LLM innovations.”
Tan positioned Jalapeño as competitive with Nvidia’s Blackwell series and Google’s tensor processing units—currently among the most advanced AI accelerators available.
Design Philosophy
OpenAI President Greg Brockman positioned the chip within a comprehensive infrastructure strategy. “By designing more of the stack ourselves, we can serve more intelligence with greater efficiency,” he explained.
OpenAI emphasized that Jalapeño represents the inaugural entry in a multi-generation computing platform. The subsequent iteration is scheduled for 2028, with yearly releases planned thereafter.
Finished processors will be deployed across data centers operated by Microsoft and additional partners. Celestica, a Canadian electronics manufacturer, will construct the server infrastructure.
Industry-Wide Movement
As one of Nvidia’s largest chip purchasers, OpenAI faces fierce competition across the AI sector for silicon supply. Creating proprietary processors provides the organization with an alternative compute resource pathway.
Amazon, Google, and Microsoft are all developing or already utilizing custom AI processors. Meta similarly designs and implements its own silicon. Certain companies, including Amazon and Google, have begun offering these chips to external clients.
OpenAI and Broadcom initially disclosed their chip collaboration in October. Original projections anticipated sufficient silicon capacity to demand 10 gigawatts of electricity.
During Wednesday’s announcement, Tan suggested that demand has expanded sufficiently that his previous estimate of 1.3 gigawatts for next year’s chip deployments “may prove conservative.” He told Bloomberg: “We like to think we can do better because there is a lot of demand.”
Nvidia competitor AMD is likewise pursuing AI data center market share. Firms such as Qualcomm and Cerebras are also entering the arena.
First-phase deployment of Jalapeño chips is scheduled before 2026 concludes.
Get 3 Free Stock Ebooks
Discover top-performing stocks in AI, Crypto, and Technology with expert analysis.
- Top 10 AI Stocks - Leading AI companies
- Top 10 Crypto Stocks - Blockchain leaders
- Top 10 Tech Stocks - Tech giants
