OpenAI Reveals 'Jalapeño' Custom Inference Chip Built with Broadcom, Joining the Anti-Nvidia Wave

OpenAI has unveiled plans for a custom AI inference chip codenamed "Jalapeño," developed in partnership with semiconductor giant Broadcom, as reported by TechCrunch. The move places OpenAI alongside Google, Apple, and SpaceX in a growing list of tech giants building their own silicon to reduce reliance on Nvidia.

Nvidia has dominated the AI chip market for years, with its GPUs serving as the de facto standard for both training and inference workloads. However, the soaring demand for AI compute has driven major customers to seek alternatives, motivated by supply-chain risk, pricing pressure, and the need for specialized architectures.

Jalapeño is positioned as an inference-specific chip — designed to run already-trained models efficiently rather than train new ones. This strategy mirrors Google's Tensor Processing Units (TPUs) and Apple's Neural Engine, both of which target specific workload profiles to improve power efficiency and cost per query.

The trend toward in-house chip development poses a long-term threat to Nvidia's market dominance. While Nvidia's CUDA ecosystem and raw performance remain formidable, an increasing number of hyperscale customers are opting for custom silicon, potentially compressing Nvidia's addressable market over time. The era of single-supplier dependence in AI hardware may be drawing to a close.