June 25th.OpenAI Yesterday Broadcom Joint launch of the first one in historychip「Galapeño(Mexican chili), mission design for large-language models。
OpenAI claims that Galapeño is the first generation of AI accelerators of the multigenerational platform. The chip structure is designed by OpenAI on the basis of model road maps, reasoning, kennel, service systems and product requirements, Broadcom for silicon realization, network and connectivity technology, Celestica participation card, cabinet and system integration。
It is worth noting that the Galapeño development cycle took only nine months and that, in addition to benefiting from the collaborative development of OpenAI and Chase at the soft and hardware level, OpenAI has accelerated some of the chip design and optimization processes using its own model。
IN OTHER WORDS, THE AI MODEL FOR SERVICE USERS IS ALSO INVOLVED IN THE DEVELOPMENT OF THE NEXT GENERATION OF AI HARDWARE. KEY OFFICIAL DISCLOSURES INCLUDE:
Engineering samples have been commissioned in laboratories to operate machines with target frequencies and power consumption, including GPT-5.3-Codex-Spark。
Early tests have shown that the Galapeño per-watt performance is above the current state-of-the-art level and that the completeness technical report will be released in the coming months。
The platform is planned to begin deployment at the end of 2026 and expand to multi-generation products in subsequent years; Broadcom CEO Hock Tan indicated that it would support the construction of a Giwa level data centre with data centre partners such as Microsoft。