Large investment
The path to creating a custom AI chip requires considerable resources. Industry experts told Reuters that designing a single version of such a processor could cost $500 million, with the additional cost of developing support software and hardware double that amount. He said there is a possibility that it would be.
According to Reuters, the current OpenAI chip project, led by former Google Chip designer Richard Hoe, involves a team of 40 engineers working with Broadcom. Also, TSMC, a Taiwanese company that produces Nvidia chips, will use 3 nanometer process technology to manufacture Openai chips. The chip reportedly incorporates high-bandwidth memory and networking capabilities similar to those found on Nvidia’s processors.
Initially, Openai’s first chip focuses primarily on running AI models (often called “inference”) rather than training, with limited deployment. The timeline suggests that mass production could begin at TSMC in 2026, but initial tape-outs and production runs may require additional modifications and delay projects for months They face a number of technical risks.
Openai’s transition to AI hardware is due to large tech companies spending record amounts on AI infrastructure. Microsoft is planning to invest $80 billion in 2025, and Meta has secured $600 billion next year, Reuters said. Last month, Openai (with SoftBank, Oracle and MGX) announced its $500 billion “Stargate” infrastructure project aimed at building a new US AI data center.
