A conceptual diagram illustrating the architecture of OpenAI’s proprietary AI chip, codenamed ‘Jalapeño’, designed in collaboration with Broadcom. (Illustrative AI-generated image).
OpenAI Unveils Its First Custom AI Chip, Jalapeño, Built with Broadcom
OpenAI, the company behind ChatGPT, has unveiled its very own custom computer chip named Jalapeño. This chip was built in collaboration with chip giant Broadcom. This development marks a significant move away from OpenAI’s previous reliance on Nvidia for its hardware needs, representing a strategic pivot toward vertical integration in the rapidly evolving artificial intelligence industry.
The announcement was made on June 24, 2026. TechCrunch was the first to report the news, with other major outlets like CNBC, Bloomberg, and Quartz quickly following with their analyses. OpenAI also released a statement describing the chip as an LLM-optimized inference processor, providing official technical framing that underscores its specialized design.
The Jalapeño chip is designed for a specific, crucial task: running AI models after they have been trained. This process is known as inference. It’s what happens when you ask ChatGPT a question and it generates a response. The custom chip aims to make this inference step faster and more cost-effective, addressing one of the most significant operational challenges facing large-scale AI deployment today.
While the name Jalapeño might sound informal, the technology it represents is serious. This is OpenAI’s first major foray into designing its own silicon. For years, the company heavily relied on Nvidia’s graphics processing units (GPUs) for both training and running its AI models. Now, it has a custom alternative in development, signaling a new chapter in its hardware strategy.
The chip’s unveiling comes at a time when the AI industry is grappling with soaring computational costs and supply chain bottlenecks. Nvidia’s high-end chips, such as the H100 and the newer B200, have been in extremely high demand, leading to long lead times and escalating prices. By developing its own silicon, OpenAI aims to gain more control over its infrastructure and reduce its vulnerability to external market forces.
Broadcom, a leading semiconductor company known for its expertise in custom chip design and networking solutions, was a natural partner for this project. The collaboration leverages Broadcom’s extensive experience in creating application-specific integrated circuits (ASICs) for major technology firms. This partnership allows OpenAI to tap into Broadcom’s manufacturing and design capabilities without having to build those capabilities from scratch.
What Is the OpenAI Custom Chip Jalapeño?
Jalapeño is a specialized computer chip, not a general-purpose processor like those found in standard computers. Its primary function is to run large language models (LLMs) quickly and efficiently. It’s optimized for inference tasks, which are the most common operations in production AI systems serving millions of users daily.
Think of it like a highly tuned engine built specifically for a race car. While a standard car engine can handle various tasks, a race car engine is designed for maximum speed and performance in its specific domain. Jalapeño follows this principle, focusing solely on inference optimization rather than general computation or training. This specialization allows for significant gains in efficiency and throughput.
Inference is the stage where a trained AI model applies its learned knowledge to generate outputs. For instance, when you use ChatGPT to draft an email, the model is performing inference, creating new text based on the patterns it learned during its training phase. This process demands substantial computing power, and Jalapeño is engineered to handle this load more effectively than general-purpose chips, reducing latency and energy consumption.
OpenAI’s statement highlighted that Jalapeño is an LLM-optimized inference chip. This means it’s specifically tuned for the unique mathematical operations and memory access patterns characteristic of large language models. It’s capable of handling tasks such as text generation, image recognition, and other model-based outputs. The chip’s architecture likely includes specialized matrix multiplication units and high-bandwidth memory interfaces to accelerate these workloads.
The Jalapeño chip is currently in production. OpenAI intends to deploy it within its own data centers, signaling a shift towards using proprietary hardware to serve its popular products like ChatGPT to users. This deployment will likely begin with a limited rollout for testing and optimization before scaling to handle a larger share of inference traffic.
Industry analysts have noted that custom inference chips can offer substantial cost savings compared to using general-purpose GPUs. For a company like OpenAI, which operates at massive scale, even modest per-query efficiency improvements can translate into millions of dollars in annual savings. Bloomberg’s reporting emphasized these cost and speed advantages, highlighting the chip’s potential to improve OpenAI’s bottom line while also enhancing user experience through faster response times.
Why OpenAI Developed Its Own AI Chip
OpenAI’s operations have historically depended on Nvidia’s GPUs, which are the industry standard for AI development due to their power and flexibility. However, these chips are also expensive and have faced significant supply constraints, creating a bottleneck for companies that need to scale their AI services rapidly.
The recent surge in AI development led to unprecedented demand for Nvidia’s high-end chips, such as the H100 and the newer B200. This resulted in lengthy waiting times for deliveries and escalating prices, making it difficult for companies, including OpenAI, to acquire sufficient hardware. Some reports indicated that Nvidia’s GPUs were selling for tens of thousands of dollars each on the secondary market, far above their list prices.
As ChatGPT gained immense popularity, OpenAI faced a growing need for more computing power. The cost of renting or purchasing Nvidia chips ran into billions of dollars, impacting profitability and potentially slowing down development cycles. By designing its own chip, OpenAI can potentially reduce these costs significantly over the long term, especially as its user base continues to expand.
Developing a custom chip like Jalapeño provides OpenAI with greater control over its hardware infrastructure. The company can design the chip to meet its specific requirements precisely, avoiding the compromises often necessary for general-purpose hardware. This tailored approach can lead to improved performance and reduced operational costs, as the chip can be optimized for the exact workloads OpenAI runs most frequently.
According to CNBC, this initiative is part of OpenAI’s broader strategy to ‘build the full stack.’ This means controlling every layer of its technology, from the underlying hardware to the software and user interface. Owning custom silicon is a critical component of this vertical integration, allowing OpenAI to optimize the entire system for its specific needs rather than relying on off-the-shelf components designed for a broad market.
Furthermore, developing its own chip reduces dependence on a single supplier like Nvidia. Relying solely on one vendor introduces risks related to pricing, supply chain disruptions, and strategic alignment. Having its own chip offers OpenAI more flexibility and options, even if it continues to purchase Nvidia chips for specific needs like model training. This diversification is a prudent risk management strategy in an industry where hardware availability can make or break a company’s growth plans.
Bloomberg has pointed out the potential cost and speed advantages. Running AI models on custom-designed chips can be more efficient because the hardware is tailored to the specific mathematical operations that LLMs require. This can lead to faster inference times and lower energy consumption per query, which are critical factors for a company serving billions of requests each month.
The partnership with Broadcom also provides access to advanced manufacturing processes and design expertise. Broadcom has a long history of creating custom chips for major technology companies, including networking and storage solutions. This experience likely accelerated the development timeline for Jalapeño, allowing OpenAI to bring its first custom chip to market more quickly than if it had attempted the project alone.
Looking ahead, the success of Jalapeño could pave the way for future custom silicon projects at OpenAI. The company may eventually develop chips for training as well, further reducing its reliance on external suppliers. However, for now, the focus remains on inference, where the immediate cost and performance benefits are most pronounced.
The broader implications of this move extend beyond OpenAI. Other AI labs and large technology companies are likely watching closely. If Jalapeño proves successful, it could accelerate the trend toward custom silicon in the AI industry, potentially reshaping the competitive landscape and reducing the dominance of general-purpose GPU vendors like Nvidia.
In summary, the unveiling of the Jalapeño chip represents a pivotal moment for OpenAI. It marks the company’s transition from a pure software and services provider to a vertically integrated technology firm with its own hardware capabilities. This strategic shift could give OpenAI a significant competitive advantage in terms of cost, performance, and control over its infrastructure, positioning it for continued growth in the years ahead.
