OpenAI has dropped news about a collaborative design of AI chips with Broadcom, boasting performance that rivals Nvidia's Blackwell. OpenAI plans to roll out these chips by the end of the year. Celestica CLS will be on the hook for manufacturing server systems exclusively for OpenAI's internal use. OpenAI reports that the chip engineering samples have been running with the GPT-5.3-Codex-Spark model, hitting the expected power consumption and performance targets. The engineers spent around nine months perfecting the design before handing it off to TSMC 2330 for fabrication.