It reveals the boom that models are witnessing artificial intelligence A parallel challenge that is no less important than developing the models themselves, which is providing a computer architecture capable of operating them efficiently with the increasing number of users and the increasing demand for computing power. In this context, custom electronic chips have become one of the most prominent areas of competition between companies artificial intelligence.
In this context, OpenAI revealed, in an official announcement, its first dedicated electronic chip named Jalapeño, describing it as the first “Intelligence Processor” developed by the company. The chip was developed in cooperation with Broadcom, as part of a long-term strategy to build an integrated infrastructure aimed at providing greater computing capacity and improving the efficiency of operating artificial intelligence models on a large scale.
We’ve designed and built our first AI chip: Jalapeño.
Designed from the ground up by OpenAI and brought to production with @BroadcomJalapeño 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 (@OpenAI) June 24, 2026
According to a report published by the AI Business website, based on financial data and estimates, the cost of operating GBT Chat servers amounted to about $8.4 billion during the past year, and it is expected to rise to about $14 billion this year with the number of users reaching 900 million users weekly. The report also indicates that OpenAI intends to invest about $1.4 trillion in computing power over the next eight years, while its current annual revenues are estimated at about $25 billion.
The Jalapeño chip is specifically designed to run large language models in the inference phase, not for general purpose or model training. OpenAI designed the chip architecture based on its model development map and operating systems, while Broadcom managed the chip architecture and integration of high-performance networking technologies. The actual manufacturing is handled by TSMC in Taiwan, while Celestica assembles the boards and systems used within the data centers.
GPT chat logo (sites)
According to OpenAI, prototypes of the chip are already running advanced workloads, including an unannounced model named GPT-5.3-Codex-Spark, while achieving production target operating frequencies and power consumption. The head of the company’s hardware program, Richard Ho, also explained that the chip’s design focuses on reducing data movement within the system, which increases the efficiency of using computing power and reduces data transfer bottlenecks facing the operation of large interactive language models.
To achieve this, Broadcom’s Tomahawk networking technologies were integrated, allowing the chips to operate efficiently within huge data centers containing large numbers of processors.
The Jalapeño project is not limited to developing a new segment, but rather falls within the company’s broader strategy. According to what OpenAI announced, the chip represents part of a long-term strategy to build a full-stack infrastructure strategy, which depends on designing more components of the system internally, with the aim of improving the efficiency of operating artificial intelligence models, and making them faster, more reliable, and less expensive, allowing for expanding access to them.
