On July 21, 2024, OpenAI made a significant adjustment to its self-research and production plans for AI chips. In an effort to reduce its reliance on externally sourced AI chips, OpenAI plans to design and produce its own chips. Although it had initially engaged with multiple chip giants such as Broadcom, it ultimately decided to outsource production to TSMC.

Originally, OpenAI CEO Sam Altman intended to raise $7 trillion to build a fabrication plant for the research, development, and production of its own AI chips. However, after negotiations with TSMC, Altman revised this plan, deciding to use the funds raised to establish a joint venture chip design company focused on AI chip development, with TSMC handling production.

According to Wccftech, this decision was primarily due to TSMC's ability to provide sufficient capacity to enable OpenAI's production plans. Additionally, OpenAI's experience working with Broadcom and NVIDIA led it to realize that outsourcing production to a specialized chip foundry could effectively reduce costs and risks.

OpenAI's in-house AI chips are expected to perform similarly to NVIDIA's products, but due to the years of experience required in chip design, the earliest market availability is projected for 2026. Market observers predict that despite streamlining the process, AI design companies will still need to invest significant time and costs.

The partnership with TSMC has altered Altman's thinking. TSMC's willingness to provide capacity for OpenAI not only addresses the issue of the high costs associated with building a fabrication plant, but also allows OpenAI to focus on chip design. The Financial Times in the UK also reported on Altman's discussions with Broadcom, noting that OpenAI will not stand idly by as competitors actively build AI computing infrastructures.

In addition to engaging with chip design firms like Broadcom, OpenAI has recruited former members of Google's TPU division to form a powerful chip team led by Richard Ho, former Senior Director of Engineering at Google TPU. This team may opt to collaborate with an American company like Broadcom to develop new AI chips. Altman has also discussed chip plans with memory chip manufacturers Samsung and SK Hynix, considering various chip packaging and memory components to maximize chip performance.

Developing new chips is just one part of the plan. Altman aims to build new data centers to house these chips and establish joint ventures with external investors to cover the associated costs. OpenAI has committed to leasing servers from these data centers. It is understood that Altman is also working with the US Department of Commerce and foreign governments such as those in the Middle East to seek approval for such projects.

While OpenAI's partnership with TSMC provides a viable path for its in-house chip plans, challenges remain. Developing a new server chip comparable to NVIDIA's could take several years and may impact OpenAI's relationship with NVIDIA.

NVIDIA's dominant position and high profit margins in the AI GPU market make it a key supplier to customers like OpenAI. If OpenAI or the new joint venture continues to design new AI chips, it could potentially give OpenAI leverage in future pricing negotiations.

OpenAI's shift from building its own fabrication plant to partnering with TSMC for chip production demonstrates its strategic flexibility and adaptability. By collaborating with industry leaders like TSMC and Broadcom, OpenAI can not only reduce costs and risks but also focus on chip design and innovation, laying the foundation for future AI technology developments.

*Disclaimer: The above content is compiled from the internet and is for communication and learning purposes only. If there are any content or copyright issues, please contact us for removal.