Chad Smith
OpenAI's first custom AI chip, called Jalapeño, was developed in partnership with Broadcom and is now in early testing, where the companies say it outperforms current state of the art processors. The chip targets inferencing workloads and puts OpenAI in direct competition with Nvidia for the first time on silicon.
At a Glance
- OpenAI and Broadcom co-developed Jalapeño, a custom chip built for running AI models
- Early testing results reportedly beat current leading chips, a pointed challenge to Nvidia
- Development took nine months from start to finish
- The chip is the first in a planned multi-generation platform, with rollout starting later this year
- Broadcom shares rose more than 1% on the news
What Jalapeño Is and Why It Matters
OpenAI unveiled Jalapeño on Wednesday, describing it as an inferencing chip designed both for its own AI models and for broader industry use. Inferencing is simply the process of running a trained AI model, and it is where most of the real world compute cost lives. Building a chip optimized for that task, rather than buying off the shelf from Nvidia, gives OpenAI far more control over cost and availability.
OpenAI president Greg Brockman framed the chip as part of a longer term infrastructure ambition. "By designing more of the stack ourselves, we can serve more intelligence with greater efficiency and keep pushing advanced AI toward broader access," he said in a statement. The company says the goal is to make compute more abundant, which in turn makes AI faster, more reliable, and cheaper for both consumers and businesses.
Nine Months From Concept to Silicon
The speed of development stands out. OpenAI says Jalapeño went from concept to working chip in just nine months, a timeline that would be considered aggressive even for established semiconductor companies. The chip is positioned as the first in a multi-generation platform that will begin reaching customers later this year and continue expanding in the years ahead.
For Broadcom, the partnership extends its already substantial role in custom AI silicon. The company has worked with Google on its Tensor Processing Units for years and has become the go-to manufacturing and design partner for hyperscalers that want custom chips without building an in-house semiconductor team from scratch.
A Crowded Push Away From Nvidia
OpenAI is one of Nvidia's largest customers, but that relationship comes with a real constraint: every major AI company is chasing the same GPUs, and supply does not always meet demand. Developing its own chip gives OpenAI a parallel path to computing power that does not depend on Nvidia's production schedule or allocation decisions.
The broader trend is hard to miss. Amazon, Google, and Microsoft all use or are actively developing proprietary AI processors. Meta builds and deploys its own chips for AI and other workloads, and has floated the idea of entering the cloud computing market, which would put it in direct competition with Nvidia's chip rental business. AMD is pressing its case in the AI data center space, while companies like Qualcomm and Cerebras are working to carve out positions of their own.
What This Means for the Chip Market
Nvidia still dominates AI compute by a wide margin, but the competitive picture is changing quickly. When OpenAI, arguably the most prominent AI company in the world, announces a chip that it says beats current state of the art processors, that carries real signal value regardless of whether Jalapeño eventually displaces Nvidia at scale. The more immediate effect is to reduce OpenAI's dependence on a single supplier and give it leverage in pricing conversations.
Amazon and Google have already started renting their custom chips to outside customers, turning what began as an internal cost-saving measure into a new revenue line. OpenAI has not announced similar plans, but the multi-generation roadmap suggests the company is thinking well beyond a single product.
Frequently Asked Questions
What does the Jalapeño chip actually do?
Jalapeño is built for inferencing, which means running AI models rather than training them. OpenAI says it is designed for its own models as well as others across the industry, though commercial availability details have not been announced.
Why is OpenAI building its own chips if it already buys from Nvidia?
Nvidia's chips are in high demand across the entire AI industry, making supply tight. A proprietary chip gives OpenAI a secondary source of compute that it can optimize specifically for its workloads and access without competing against its own rivals for allocation.
Is Jalapeño available to other companies?
OpenAI has described the chip as built for its models and those across the industry, but it has not announced a plan to rent or sell access to outside customers. The rollout is expected to begin later in 2025.
How does Broadcom fit into this?
Broadcom is the manufacturing and co-design partner on Jalapeño. The company has a long track record building custom AI chips for major tech companies, including Google, making it a natural collaborator for OpenAI's first silicon effort.
OpenAI's Infrastructure Ambitions Take Shape
Jalapeño is a concrete step in what OpenAI describes as a full-stack infrastructure strategy. The company is not just competing on models anymore. Owning more of the hardware layer gives it control over cost, speed, and reliability in ways that pure software cannot. Whether the chip lives up to its early testing results at production scale is the question that will matter most in the months ahead.
