OpenAI's first in-house custom chip will be released in 2025, and the secret weapon to break away from NVIDIA dependency finally appears



This article, originally posted in Japanese on 15:00 Feb 12, 2025, may contains some machine-translated parts.
If you would like to suggest a corrected translation, please click here.

Reuters reported that OpenAI is moving forward with plans to develop its own AI semiconductors and outsource their manufacturing to TSMC as part of a plan to reduce its reliance on NVIDIA. OpenAI is expected to '

tape out ' its custom chip designs and send them to the factory within the next few months.

Exclusive: OpenAI set to finalize first custom chip design this year | Reuters
https://www.reuters.com/technology/openai-set-finalize-first-custom-chip-design-this-year-2025-02-10/

OpenAI's secret weapon against Nvidia dependence takes shape - Ars Technica
https://arstechnica.com/ai/2025/02/openais-secret-weapon-against-nvidia-dependence-takes-shape/

According to sources who spoke to Reuters, the new chips, which will be developed with an emphasis on AI training, are positioned within OpenAI as a 'strategic tool to strengthen its negotiating power with other chip suppliers.'

DeepSeek's emergence has raised questions about whether AI can be built without large amounts of powerful chips, but generative AI model makers such as OpenAI, Google and Meta have demonstrated that the more chips connected to data centers, the smarter their models become, keeping chip demand high.



If OpenAI can successfully complete its first tape-out, the company could mass-produce its first home-grown AI chips and test its own AI accelerators to replace NVIDIA chips in the second half of 2025.

Reuters said, 'Typically, it takes years from when a chip is developed to when it can be manufactured, but OpenAI's plan to send its designs to TSMC within the year shows how quickly the company is moving forward with its first designs.'

Because training and operating AI requires huge computing resources, technology companies such as Microsoft , Amazon , Google , and Meta have developed their own custom AI chips in various forms.

Meta announces the second generation of its self-developed AI chip 'MTIA', achieving three times the performance of the previous generation and already introduced in data centers - GIGAZINE



OpenAI has also been rumored to be developing its own chips for some time, and it has been reported that it is raising huge amounts of money to build an AI chip factory in 2024 and has entered into a collaborative arrangement with Broadcom and TSMC.

OpenAI plans to develop and manufacture its first chips in collaboration with Broadcom and TSMC, scaling back ambitions for its own manufacturing - GIGAZINE



OpenAI's home-grown AI chips are reportedly capable of both training and running AI models, but they will reportedly be deployed on a limited scale initially and will be used primarily to run AI models, meaning they will likely play a limited role in the company's infrastructure for the time being.

Reuters said, 'OpenAI would need to hire hundreds of engineers to build an effort as comprehensive as Google and Amazon's AI chip programs.'

OpenAI and TSMC declined to comment.

in Hardware, Posted by log1l_ks