OpenAI and Broadcom (NASDAQ: AVGO) today unveiled Jalapeño, OpenAI’s first "Intelligence Processor": a chip specifically designed for large language model (LLM) inference that reflects OpenAI’s vision for future inference workloads. Jalapeño is the first component of a multi‑generation compute platform the two companies are building to make advanced AI faster, more reliable, and more widely accessible.
The chip was handed to OpenAI CEO Sam Altman and President Greg Brockman by Broadcom President and CEO Hock Tan and President Charlie Kawwas, underscoring a milestone in OpenAI’s plan to build the full stack behind its models and products.
Design and collaboration
OpenAI designed the chip from the ground up based on its deep knowledge of LLM fundamentals, informed by its roadmap of models, kernels, serving systems, and product requirements. Broadcom and Celestica supported industrializing the platform: Broadcom contributes silicon implementation and networking technologies, including Tomahawk networking silicon, while Celestica provides board, rack, and system integration expertise and scalable production.
Jalapeño is intended to be flexible enough to work with all LLMs, guided by OpenAI’s insights into current and future inference needs across the industry. Engineering samples of the chip are running machine learning workloads in the lab at production target frequency and power, including GPT‑5.3‑Codex‑Spark.
Performance and efficiency
OpenAI is still finalizing performance measurements, but early tests indicate that Jalapeño will deliver substantially better performance per watt than current state‑of‑the‑art solutions. A detailed technical report on performance is expected in the coming months. The architecture reduces data movement and balances compute, memory, and networking resources to achieve realized utilization closer to theoretical peak performance.
Broadcom’s silicon implementation and networking technologies help prepare the platform for large‑scale production deployment.
Role in the full stack and strategic impact
OpenAI emphasizes that Jalapeño is a blank‑slate design for modern LLM inference rather than a general‑purpose accelerator repurposed from earlier AI workloads. The objective is to combine the power and throughput of today’s leading AI accelerators with latency closer to the fastest specialized inference systems, making Jalapeño suitable for interactive LLM products at scale.
OpenAI argues that better infrastructure increases compute efficiency, which in turn enables better training and serving, powering more capable models and better products. Improved products drive usage, customers, and revenue, enabling reinvestment into the next generation of infrastructure—creating a reinforcing cycle toward more capable, reliable, and affordable intelligence.
Development speed and next steps
The Jalapeño program progressed from initial design to manufacturing tape‑out in nine months; OpenAI describes this as among the fastest ASIC development cycles achieved in high‑performance advanced semiconductors. That speed reflects close software‑hardware co‑development with OpenAI engineering teams, Broadcom’s silicon expertise, and use of OpenAI models to accelerate portions of the design and optimization process.
Jalapeño is the first step in a multi‑generation compute platform planned for initial deployment by the end of 2026 and expansion in subsequent years. The platform will combine OpenAI‑designed accelerators with Broadcom’s silicon and networking technologies and Celestica’s board, rack, and system competencies.
Why it matters
OpenAI states that inference is where AI reaches people: improvements in cost, speed, and reliability translate into faster ChatGPT responses, Codex workflows that can take more steps with less waiting, cheaper API products, and more dependable access under high demand. Making advanced models more available and affordable is central to democratizing AI, and Jalapeño is presented as a tool to help OpenAI move more of its infrastructure into production use for students, developers, small businesses, researchers, enterprises, and others seeking to learn, create, or solve difficult problems.
