OpenAI and Broadcom Unveil Jalapeño Custom AI Chip for LLM Inference
window 24h
IN SHORT
OpenAI and Broadcom have partnered to develop Jalapeño, a custom AI chip for large language model inference, aiming to reduce reliance on Nvidia and improve efficiency, with deployment expected in late 2026. Meanwhile, Qualcomm is reportedly in talks with ByteDance for custom chip designs, potentially utilizing technology from its AlphaWave Semi acquisition. Qualcomm has also launched new data center chips to challenge Nvidia in the AI processor market and plans a China-specific chip to navigate export controls.
✉Newsletter
PiQ Daily
Pick your topics. Get only what matters, on your cadence.
Who's Involved
OpenAI
AI research and deployment company developing custom AI chips
Broadcom
Technology company collaborating on custom AI chips for OpenAI
Nvidia
Major provider of AI chips and competitor in the AI processor market
Qualcomm
Technology company developing data center chips and custom chip designs
ByteDance
Parent company of TikTok in talks for custom chip design services
AlphaWave Semi
Company whose technology may be used in Qualcomm's custom chips
1 / 2
Key facts
OpenAI and Broadcom are collaborating on a custom AI chip called Jalapeño.
Jalapeño is designed for large language model inference.
The chip aims to enhance efficiency and reduce OpenAI's dependence on external providers like Nvidia.
Deployment of Jalapeño chips is planned from late 2026.
Qualcomm is in discussions to design custom chips for ByteDance.
These chips may use technology from AlphaWave Semi, acquired by Qualcomm.
Qualcomm has launched new data center chips to compete with Nvidia.
Qualcomm plans to design a China-specific chip.
The China-specific chip is intended to comply with US export controls.
OpenAI and Broadcom have announced a collaboration to create Jalapeño, a custom AI inference chip specifically designed for large language models (LLMs). This partnership aims to boost efficiency and decrease OpenAI's reliance on third-party chip manufacturers, notably Nvidia. The deployment of these custom chips is anticipated to commence in late 2026.
In parallel, Qualcomm is reportedly engaged in discussions with ByteDance, the parent company of TikTok, to provide custom chip design services. This potential agreement could involve chips incorporating technology from AlphaWave Semi, a company Qualcomm acquired in the previous year. The focus of these chips might be on video processing units.
Furthermore, Qualcomm has introduced a new series of data center chips, marking its entry into the AI processor market and positioning itself as a competitor to Nvidia. The company also intends to develop a chip tailored for the Chinese market, ensuring compliance with existing US export controls. This move signifies a broader strategy by Qualcomm to expand its footprint in the AI and data center sectors.
↳ Why This Matters
OpenAI and Broadcom have announced a collaboration to create Jalapeño, a custom AI inference chip specifically designed for large language models (LLMs). This partnership aims to boost efficiency and decrease OpenAI's reliance on third-party chip manufacturers, notably Nvidia. The deployment of these custom chips is anticipated to commence in late 2026.
Frequently asked questions
The partnership aims to develop and deploy custom AI chips for OpenAI, reducing its reliance on Nvidia and securing computing power for future AI models and products.
Broadcom will begin developing and deploying the chips starting in the second half of 2026, with full deployment expected by the end of 2029.
The agreement involves 10 gigawatts worth of custom chips.
Yes, the partnership is seen as a move to challenge Nvidia's dominance in the AI chip market, although analysts suggest custom chips may not pose a short-term threat.
What Happens Next
01Broadcom will begin developing and deploying custom AI chips in the second half of 2026.
02Future generations of the chip will support gigawatt-scale AI infrastructure starting in 2026.
03Full deployment of the custom chip systems is expected by the end of 2029.
Get the newsletter.
Pick the topics you actually care about. We'll email when there's news worth your time, on the cadence you choose. Cancel any time from your account.
