OpenAI’s new Jalapeño chip aims to optimize artificial intelligence operations and cut expenses
OpenAI, developer of the renowned ChatGPT, announced the creation of its first custom artificial intelligence chip, called Jalapeño. This semiconductor, the result of a strategic collaboration with Broadcom, represents a crucial step for the company, which seeks greater autonomy and efficiency in the hardware responsible for executing its complex AI models. The initiative projects significant operational cost savings, estimated at up to 50%, by adapting the computing infrastructure to the specific requirements of artificial intelligence platforms.
Strategic partnership with Broadcom boosts initial chip testing
Presented jointly by OpenAI and Broadcom, the Jalapeño chip is already in an advanced stage of testing. The first samples of the accelerator are under rigorous evaluation, and preliminary data reveals a considerable reduction in costs, around 50%, when compared to graphics processing units (GPUs) traditionally used in artificial intelligence tasks. Hock Tan, CEO of Broadcom, expressed optimism, suggesting that the deployment potential could exceed the initial forecast of 1.3 gigawatts in capacity next year.
Cost reduction and performance optimization in data centers
The integration of the definitive chips is expected to take place at the end of the year, focusing on the large data centers of Microsoft, one of OpenAI’s main investors, and other strategic partners. The overarching goal is to ensure more efficient and cost-effective execution of vast artificial intelligence workloads. Optimizing hardware for the inference phase — the time when AI models actually perform their tasks after training — is critical to the scalability and economic sustainability of services like ChatGPT at scale. Continued reliance on general-purpose GPUs, while powerful, can incur prohibitive costs for the routine operations of large language models (LLMs).
OpenAI seeks to diversify semiconductor suppliers
Although OpenAI still maintains a heavy dependence on chips from Nvidia, the market leader in AI processors, the company has been actively investing in expanding its network of hardware suppliers. This diversification is vital to meet the growing demand for its artificial intelligence services and models. The company has already established significant financial agreements with other chipmakers, including Advanced Micro Devices (AMD) and Cerebras Systems, signaling a conscious effort to mitigate risks and ensure a steady supply of specialized components.
Leia Também
Messi reaches 19 goals in FIFA World Cups with great free kick for Argentina in clash with Jordan
Cristiano Ronaldo shines beyond the pitch at the 2026 World Cup with business empire and family luxury
At 41, Cristiano Ronaldo reaffirms his historic legacy and prepares to shine in the 2026 World Cup
Billion-dollar investment and financial support for new infrastructure
OpenAI has plans to invest tens of billions of dollars in acquiring Broadcom chips, highlighting the magnitude of its contributions to physical infrastructure to maintain its AI operations, even in a scenario of no profit. To facilitate this strategic movement, Broadcom created a financing vehicle in collaboration with important asset managers in the market:
- Apollo Global Management Inc.
- Blackstone Inc.
Earlier this year, OpenAI had already secured additional funding of US$122 billion, strengthening its ability to expand into chips, data centers and attract specialized talent.
Jalapeño chip design breakthrough for large language models
The development of the new Jalapeño chips was completed in record time, from scratch, highlighting the agility and intense collaboration between the two companies. Although still in the evaluation phase, preliminary data released by OpenAI indicates that Jalapeño delivers performance per watt of energy “substantially superior to the current state of the art”. The chip is designed specifically for large language models (LLMs) and optimized for the crucial inference phase. Richard Ho, director of hardware at OpenAI, emphasized that despite its focus on LLMs, the device has a very general purpose, capable of supporting future innovations in this area.
Development roadmap and the future of hardware autonomy
OpenAI and Broadcom have already outlined a clear development roadmap for the next generations of the Jalapeño chip. A new version is expected to be released in 2028, with subsequent annual updates expected. While the current Jalapeño focuses primarily on inference, future editions could be designed to cover other artificial intelligence workloads. For OpenAI, this initiative marks the beginning of a quest for greater control over every level of the technological infrastructure it depends on, a move that Broadcom’s Hock Tan believes will be replicated by other major developers of AI models, especially outside of China, looking for accelerators and custom networking solutions. This advancement represents a continuous search for lasting competitive advantage and greater technological independence.
