news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

In order to reduce the dependence on purchased AI chips, Microsoft is rumored to have invested in a generative AI application companyOpenAIOpenAI has already started to design and produce related chips on its own, and has contacted several chip design service giants including Broadcom. However, OpenAI may not build its own fab to produce chips, and TSMC will have the opportunity to provide OpenAI with chip manufacturing capacity.

Previously, the Wall Street Journal and other foreign media reported that OpenAI CEO Sam Altman is planning to raise $7 trillion to build a wafer factory to produce its own artificial intelligence (AI) chips. He has already held talks with many heavyweight investors such as Abu Dhabi's G42 and Japan's SoftBank Group, and has received support from companies such as Microsoft.

Although Sam Altman later denied the "7 trillion US dollars" statement and said "I think everyone underestimates the demand for AI computing", Sam Altman is not sure how many wafer fabs are needed to meet these demands. He believes that "AI computing is a different type of commodity, which is more like energy. There is a certain amount of demand at a certain price, but less demand at a higher price."

The latest report said that Sam Altman, who originally planned to raise $7 trillion to build a wafer factory to produce self-developed AI chips, changed his plan after negotiating with TSMC. Therefore, OpenAI plans to use the raised funds to set up a joint venture chip design company to implement the development and design of AI chips, and then hand over the design of its AI chips to TSMC for production.

At the same time, it is rumored that OpenAI has also recruited Richard Ho, the former head of Google TPU, and the former head of Lightmatter chip engineering to form an AI chip design department. In addition, OpenAI has also negotiated with Broadcom and has discussed related matters, possibly discussing cooperation on chip design services and related semiconductor IP.

In fact, like GPU giant Nvidia, Broadcom has become another hot AI chip company because of its network and ASIC business. Because the products developed by Broadcom can realize artificial intelligence server connection and allow developers to develop customized chips based on customer needs. Therefore, similar to Nvidia, Broadcom's stock price has also risen by 78% in the past 12 months.

Market participants expect that the performance of OpenAI's self-developed AI chip will be similar to that of Nvidia's products. However, because chip design requires years of experience, such products are not expected to be available until at least 2026. Although the process has been simplified due to the rise of foundry manufacturers such as TSMC, related AI design companies still have to invest a lot of time and money to complete these tasks.

The report emphasized that the discussion between OpenAI and TSMC played a role in changing Sam Altman's idea of ​​building his own fab to make chips, so the next focus will be on the production capacity allocation of OpenAI products and Nvidia's products. If OpenAI commits to placing an order, TSMC is also willing to provide production capacity.

The Financial Times also reported on the discussion between Sam Altman and Broadcom, and pointed out that OpenAI will not stand idly by while other competitors actively build AI computing infrastructure. Especially when they start to compete directly face to face. The report also pointed out that OpenAI is also involved in the development of a wider AI industry with Broadcom, and these plans also require a lot of capital investment. However, OpenAI did not comment on this statement.

Editor: Xinzhixun-Linzi