OpenAI has introduced its first custom AI chip, Jalapeño, developed with Broadcom as part of a broader effort to build its own AI infrastructure and reduce reliance on external hardware providers.
Quick Summary – TLDR:
- OpenAI and Broadcom unveiled Jalapeño, OpenAI’s first custom AI inference chip.
- The chip is designed specifically for large language model inference, the process behind generating ChatGPT responses.
- Early testing suggests better performance per watt than current leading AI accelerators.
- OpenAI says the chip was designed and sent for manufacturing in just nine months, with AI helping accelerate parts of the development process.
What Happened?
OpenAI and Broadcom have officially revealed Jalapeño, a custom AI processor built specifically for inference workloads. The announcement marks OpenAI’s first major step into designing its own AI chips and expands the company’s ambitions beyond models and products into the hardware powering them.
The companies say Jalapeño is the first accelerator in a multi generation compute platform that will support OpenAI’s future AI services, including ChatGPT, Codex, API offerings, and upcoming agent based products.
We’ve designed and built our first AI chip: Jalapeño.
— OpenAI (@OpenAI) June 24, 2026
Designed from the ground up by OpenAI and brought to production with @Broadcom, Jalapeño is purpose-built for the LLM workloads powering ChatGPT, Codex, the API, and future agentic products.
Chips are foundational to the AI… pic.twitter.com/mHU7DaMMTi
OpenAI Moves Deeper Into AI Infrastructure
For years, OpenAI has relied heavily on Nvidia graphics processors to train and serve its AI models. However, surging demand for advanced AI services has pushed major AI companies to seek alternative hardware solutions that can offer greater efficiency and lower operating costs.
Jalapeño represents OpenAI’s effort to gain more control over the technology stack supporting its products. The company says it designed the processor from the ground up around the specific requirements of large language models rather than adapting an existing accelerator architecture.
According to OpenAI President Greg Brockman, the company sees hardware development as a key part of making AI faster, more reliable, and more affordable for users.
Built Specifically for AI Inference
Unlike general purpose AI accelerators, Jalapeño was designed specifically for inference, the process of generating responses to user prompts after a model has been trained.
OpenAI hardware chief Richard Ho said the processor was optimized around the memory movement, networking requirements, serving systems, and computational patterns used by modern large language models.
The company says engineering samples are already running workloads in its laboratories at target power and performance levels, including GPT 5.3 Codex Spark.
Broadcom CEO Hock Tan also claimed the processor can compete with leading AI hardware currently available in the market, including Nvidia’s Blackwell chips and Google’s Tensor Processing Units.
Developed in Just Nine Months
One of the most notable aspects of the project is its development speed.
OpenAI says Jalapeño moved from initial design to manufacturing tape out in approximately nine months, a timeline the company believes is among the fastest ever achieved for a high performance advanced semiconductor project.
The chip was manufactured by Taiwan Semiconductor Manufacturing Company, better known as TSMC. OpenAI also revealed that its own AI models helped accelerate parts of the design and optimization process.
The company views this as an early example of AI assisting engineers in creating better hardware that can ultimately power future AI systems.
Broadcom and Celestica Play Key Roles
While OpenAI led the chip architecture design, Broadcom provided silicon implementation expertise, networking technology, and production support. The company is also supplying technologies such as its Tomahawk networking platform to help scale deployments.
Meanwhile, Canadian electronics manufacturer Celestica will build the server systems that house the chips.
The infrastructure is expected to be deployed across large scale AI data centers, with OpenAI and Broadcom targeting expansion over multiple generations.
A Long Term Bet on AI Scale
OpenAI says Jalapeño is only the first step in a broader hardware roadmap. The company previously disclosed plans with Broadcom to deploy racks of OpenAI designed chips at massive scale, eventually supporting infrastructure measured in gigawatts of power consumption.
OpenAI has also recently expanded partnerships with Amazon Web Services, AMD, and Cerebras as it seeks additional compute capacity beyond Nvidia’s ecosystem.
As competition among AI companies intensifies, owning more of the underlying hardware stack could become a major strategic advantage.
SQ Magazine Takeaway
I think this is one of OpenAI’s most important announcements of the year because it signals a shift from being only an AI model company to becoming a full infrastructure company. Nvidia still dominates AI hardware, but OpenAI clearly wants greater control over costs, performance, and scalability. If Jalapeño delivers on its efficiency claims, it could help OpenAI serve more users while reducing dependence on third party chip suppliers. This feels less like a single product launch and more like the beginning of OpenAI’s long term hardware strategy.
