Author: Li Dan
Source: Hard AI
Sometimes, the more idealistic it is, the more realistic it appears. In February this year, the news of OpenAI CEO Sam Altman raising $70 billion to build a chip empire shocked the world. Recent reports indicate that OpenAI has chosen a more practical path, starting by bringing in partners to develop its own chip instead of establishing a chip factory network.
On Tuesday, October 29th, Eastern Time, Reuters reported, citing sources familiar with the matter, that due to the cost and time required to establish a chip factory network, OpenAI has abandoned its ambitious chip factory plans and is now focusing on internal chip design work.
The report mentioned that OpenAI has assembled a chip team of about 20 people, led by top engineers who previously worked on building Tensor Processing Units (TPU) for Google. They are collaborating with Broadcom and TSMC to create the first self-designed chip aimed at supporting their artificial intelligence (AI) systems, with a focus on inference.
It was reported that OpenAI has been working with Broadcom for several months and has reserved TSMC's production capacity through Broadcom, planning for TSMC to manufacture OpenAI's first custom-designed chip in 2026. OpenAI has yet to determine whether to develop or acquire other elements of its chip design, as well as potentially collaborating with more partners to jointly develop chips.
At the same time, the report mentioned that in addition to currently using NVIDIA chips, OpenAI will also adopt AMD chips to meet the increasing infrastructure demands.
This is not the first time OpenAI has been reported to be collaborating with Broadcom to develop AI chips. In July this year, Wall Street CN mentioned that media reports indicated OpenAI has been recruiting former Google employees who were involved in producing Google's AI chip TPU, seeking to develop an AI server chip, and has been in talks with chip design companies including Broadcom to develop new AI chips.
At that time, it was reported that Altman discussed with TSMC executives whether TSMC could increase production capacity to produce more NVIDIA chips or the new AI chips proposed by Altman. TSMC executives told Altman that if he or OpenAI could commit to a large number of new chip orders, TSMC would be willing to expand chip production.
Developing new chips is just part of Altman's plan. Altman reportedly also intends to establish one or more companies with external investors to cover the costs of real estate, electricity, data centers, and the deployment of dedicated AI chip servers.
The media believes that the possibility of OpenAI developing servers chips comparable to NVIDIA's is slim, and it will take years to see results, which may pose a risk of upsetting NVIDIA, OpenAI's most important chip supplier, but could also provide leverage for future pricing negotiations with NVIDIA
Regardless, cooperating with OpenAI is a positive development for Broadcom. After news of the collaboration in July, a report by Citigroup analysts stated that following Google, Meta, and ByteDance, OpenAI will become Broadcom's fourth largest ASIC (Application-Specific Integrated Circuit) customer, with Broadcom expected to deliver to OpenAI after the second half of next year.
After news of the collaboration between OpenAI and Broadcom surfaced again this Tuesday, Broadcom's stock price saw a rapid increase during Tuesday's midday trading. In less than 20 minutes, the intraday gain expanded from less than 1.7% to nearly 3.7%, reaching a new daily high by the closing bell with an intraday gain of almost 4.8%, ultimately closing up 4.2%.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/acdcb969-7bdf-4bc6-802c-05ed0930c614.jpeg?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/acdcb969-7bdf-4bc6-802c-05ed0930c614.jpeg"/>
It wouldn't be surprising if OpenAI indeed gives up on its dream of a chip empire, as some previous reports have deemed this dream as somewhat crazy.
The $7 trillion, equivalent to 10% of the global GDP, is nearly 14 times the total revenue of the entire semiconductor market last year. Based on the market value calculation at the stock price in February, this amount of money could essentially acquire the entire semiconductor ecosystem, including NVIDIA, TSMC, and Broadcom, with the remaining funds enough to acquire Meta and still bring back home an additional $300 billion.
Media reports have pointed out that many executives in the chip industry believe that Altman's factory expansion ambitions are unlikely to materialize, as it would require a significant amount of capital and specialized labor. For example, when asked about the prospects of a new factory in June, TSMC's CEO C.C. Wei expressed disbelief in Altman's plans, stating, &#34;Altman is too aggressive, I can't believe it.&#34;

OpenAI

AVGO

Media reports that OpenAI has been collaborating with Broadcom for several months to develop AI inference chips, and has reserved TSMC's production capacity. The plan is for TSMC to manufacture OpenAI's first custom chip in 2026. In addition to NVIDIA chips, OpenAI will also use AMD chips to meet its needs. Broadcom's stock price saw a rapid increase during trading hours, rising nearly 5% at one point

- OpenAI has shifted its strategy from building a chip factory network to focusing on internal chip design, collaborating with partners like Broadcom and TSMC.  
- A team of about 20 engineers, previously involved in Google's TPU development, is leading the effort to create a custom chip for AI systems, with production expected by 2026.  
- Analysts suggest that while OpenAI's ambitions may be overly optimistic, the partnership with Broadcom could significantly benefit both companies, as evidenced by a rise in Broadcom's stock price following the news.

Abandon the dream of a 7 trillion chip empire? OpenAI reportedly plans to "shrink" and collaborate with Broadcom to develop the first AI chip