OpenAI and Broadcom reveal a chip for cheaper model inference

OpenAI and Broadcom introduced Jalapeño, OpenAI’s first accelerator designed to run large language models. This is not an ordinary chip announcement. OpenAI is showing that AI competition no longer happens only between models, but across full technical chains: application, model, compute scheduling, memory, network and data center. The chip is meant to be part of a multi-generation platform. OpenAI says first samples are running in the lab at target frequency and performance, processing workloads including GPT-5.3-Codex-Spark. The company has not released final performance figures, but says early tests show substantially better performance per watt than the current state of the art. The pace is notable. OpenAI says the chip moved from design to tape-out in nine months and that OpenAI models also accelerated development. If AI speeds up chip design, it creates a loop: better models help build better infrastructure, and that infrastructure then enables cheaper and more powerful models. The business impact lies mainly in costs. Running models is becoming one of the largest expenses for AI companies. If OpenAI reduces inference cost, it can serve ChatGPT, Codex, developer interfaces and future agentic products more cheaply. That improves margins, expands availability and reduces dependence on general-purpose accelerators. Infrastructure concentration also increases. OpenAI discusses deployment in gigawatt-scale data centers with Microsoft and other partners from 2026. That shifts the debate toward electricity, permitting, supply chains and semiconductor geopolitics. AI is becoming industrial capacity, not only a digital service. Terms: An accelerator is a specialized chip for fast AI computation. Inference is operational model use for an answer or agentic task. Performance per watt shows how much work a chip does per unit of energy.

OpenAI and Broadcom reveal a chip for cheaper model inference

The full article is available to users registered at Hard Skills