OpenAI and Broadcom Unveil First Proprietary AI Chip Jalapeño
OpenAI's move to custom silicon reduces its dependence on Nvidia GPUs and gives it direct control over cost, power, and supply for its AI services, marking a strategic shift toward vertical integration in AI infrastructure.
Reporting from 2 sources: ASCII.jp, GIGAZINE.
OpenAI announced on June 24 that it has jointly developed with Broadcom a custom AI chip named Jalapeño, its first proprietary processor. The chip is optimized for inference processing in large language models, powering services like ChatGPT and Codex. Engineering samples are already running machine learning workloads, including GPT-5.3-Codex-Spark, at expected performance levels. The chip progressed from design to production in nine months, with OpenAI's own models used to accelerate development. Jalapeño is designed to reduce data movement between logic and memory, a key bottleneck in inference. Broadcom CEO Hock Tan stated the chip is comparable to Nvidia's Blackwell and Google's TPU. Manufacturing is handled by TSMC, with system integration by Celestica. OpenAI plans initial deployment by the end of 2026, aiming for gigawatt-scale data center integration with partners like Microsoft. The company describes Jalapeño as the first in a multi-generational computing platform, signaling a long-term strategy to reduce reliance on external semiconductors like Nvidia's GPUs.
The chip's architecture was designed from scratch for LLM inference, not repurposed from general-purpose AI workloads, according to OpenAI. Richard Ho, OpenAI's hardware program lead, said the team optimized the architecture around the kernels, memory movement, and serving patterns most critical to state-of-the-art models. Early test results show Jalapeño performing near theoretical hardware limits on key workloads. Greg Brockman, OpenAI president, framed the chip as part of a long-term full-stack infrastructure strategy to make computing resources more abundant and AI more affordable. Broadcom's Hock Tan emphasized that the collaboration represents a fundamental effort to scale physical infrastructure for AI over the next decade, enabling gigawatt-scale data centers starting in 2026 with Microsoft and other partners. The partnership was first announced in October 2025, with a multi-billion yen strategic agreement. OpenAI has not disclosed detailed chip specifications, but noted the architecture reduces data movement, a common inference bottleneck.
Synthesized by Yomimono from the 2 cited sources below, including Japanese-language reporting where cited, then editorially reviewed before publishing.