Broadcom is reportedly collaborating with OpenAI to develop a custom artificial intelligence server chip, codenamed “Jalapeno,” as the chipmaker seeks to diversify its revenue beyond its standard networking hardware. The partnership aims to optimize infrastructure for large-scale AI models, potentially reducing OpenAI’s reliance on Nvidia’s dominant graphics processing units.
Why OpenAI is moving toward custom silicon
OpenAI is pursuing custom chip development to address the massive computational costs associated with training and running large language models. According to reports from Reuters, the organization has evaluated various paths to secure its supply chain, including potential acquisitions and the design of internal hardware. By partnering with Broadcom, OpenAI gains access to specialized expertise in Application-Specific Integrated Circuits (ASICs). These custom chips are designed for specific tasks, offering higher efficiency and lower power consumption compared to general-purpose GPUs.
How the “Jalapeno” chip impacts Broadcom’s market position
Broadcom’s pivot toward custom AI silicon represents a strategic shift to compete in the high-growth data center market. While Broadcom is traditionally known for its networking switches and custom chips for companies like Google and Meta, the OpenAI collaboration signals a deeper entry into the AI infrastructure space.
Industry analysts at CNBC note that successful deployment of the Jalapeno chip could provide a significant revenue tailwind for Broadcom, as the company looks to offset slower growth in its legacy software and semiconductor segments. The move mirrors a broader industry trend where hyperscalers and AI-first companies move away from off-the-shelf hardware to gain better control over performance margins.
Comparison of AI hardware strategies
The shift toward custom silicon places OpenAI in a unique position relative to other tech giants.
| Company | Strategy | Primary Focus |
| :— | :— | :— |
| Nvidia | Off-the-shelf GPUs | High-performance, general AI training |
| Google | Custom TPU (Tensor Processing Unit) | Proprietary AI workload optimization |
| OpenAI | Custom ASIC (Jalapeno) | Model-specific efficiency and cost reduction |
While Nvidia remains the industry standard for AI training, the move by OpenAI to co-design hardware suggests a long-term goal of optimizing its specific model architectures, such as GPT-4 or its successors, to run more efficiently on internal silicon.
What happens next for the partnership?
Development of the Jalapeno chip is currently in the engineering phase. Broadcom’s role involves managing the complex supply chain and manufacturing process, which typically includes partnering with foundries like TSMC. Success for this project depends on whether the resulting hardware can achieve a superior price-to-performance ratio compared to the latest Blackwell-series chips from Nvidia. If the project meets internal benchmarks, it could signal a shift in how AI-native companies prioritize hardware procurement, moving toward a model where the software design and the physical transistor architecture are developed in tandem.