Broadcom (AVGO) and OpenAI New Custom AI Chip “Jalapeño” Takes Aim at Nvidia

TLDR

• OpenAI and Broadcom have unveiled a custom AI chip called Jalapeño, designed for AI inference workloads.
• Early testing shows roughly 50% cost savings compared to typical AI GPUs, per Broadcom CEO Hock Tan.
• The chip took nine months to develop and is manufactured by TSMC; initial deployment is targeted for end of 2026.
• Broadcom stock climbed around 2% following the announcement; Nvidia dipped slightly.
• A roadmap for future chip generations is already in place, with the next version planned for 2028.

OpenAI and Broadcom announced a jointly developed custom AI chip on Wednesday called Jalapeño. Broadcom (AVGO) stock rose around 2% on the news, while Nvidia (NVDA) slipped 0.26%.

Broadcom Inc., AVGO

The chip is built specifically for AI inference — the process of generating responses in products like ChatGPT. It belongs to a category called ASICs, which trade flexibility for lower cost and the ability to be tuned to specific workloads.

Broadcom CEO Hock Tan said early samples are showing cost savings of roughly 50% compared to typical AI GPUs. That’s a pointed number in an industry where compute costs are a constant pressure point.

OpenAI said samples of the chip are already running in its labs at target power and performance levels, tested against its GPT-5.3-Codex-Spark model.

OpenAI’s engineers completed the chip design in about nine months, aided partly by AI tools. It was then sent to TSMC for manufacturing.

OpenAI hardware chief Richard Ho described Jalapeño as “a very general purpose device” designed with large language models in mind, but built to “address future LLM innovations.”

Tан placed Jalapeño on par with Nvidia’s Blackwell line and Google’s tensor processing units — two of the most powerful AI accelerators currently in use.

What It’s Built For

OpenAI President Greg Brockman framed the chip as part of a broader infrastructure push. “By designing more of the stack ourselves, we can serve more intelligence with greater efficiency,” he said.

OpenAI noted it’s the first chip in a multi-generation computing platform. The next version is planned for 2028, with annual releases following after that.

The finalized chips will be integrated into data centers run by Microsoft and other partners. Canadian electronics manufacturer Celestica will build the server systems.

A Growing Trend

OpenAI is one of the largest buyers of Nvidia chips, but it has to compete with the entire AI industry to secure supply. Developing its own silicon gives the company a second lane for compute access.

Amazon, Google, and Microsoft are all working on or already using custom AI processors. Meta develops and deploys its own chips as well. Some, like Amazon and Google, have started renting those chips to outside customers.

OpenAI and Broadcom originally announced their chip partnership in October. At the time, plans called for enough silicon to require 10 gigawatts of power.

On Wednesday, Tan said demand has grown enough that his earlier projection of 1.3 gigawatts of chip deployments next year “may prove conservative.” He told Bloomberg: “We like to think we can do better because there is a lot of demand.”

Nvidia rival AMD is also working to take market share in AI data centers. Companies like Qualcomm and Cerebras are also pushing into the space.

Initial deployment of Jalapeño chips is targeted for before the end of 2026.

The post Broadcom (AVGO) and OpenAI New Custom AI Chip “Jalapeño” Takes Aim at Nvidia appeared first on CoinCentral.