"Science and Technology Innovation Board Daily" reported on January 22 (Editor Zheng Yuanfang) The ai industry has been struggling with chip supply for a long time. This time even openai "can't sit still." Sam Altman, CEO of
openai, has repeatedly complained publicly about the shortage of NVIDIA GPUs and that the company's GPU supply is "severely constrained." As the "computing power overlord", NVIDIA has almost monopolized the AI chip market. With the shortage of its GPU, the cost of superimposed AI computing power has soared. Even as powerful as OpenAI, it is constantly seeking new ways to avoid being "stuck" by the chip for a long time.
According to media reports, Altman is discussing cooperation with several large investors, and hopes to raise billions of dollars to build chip manufacturing plants/wafer fabs .
Companies participating in cooperation discussions include Abu Dhabi-based AI company g42 and Japan's SoftBank Group . As discussions are still at an early stage, the full list of partners and investors has yet to be finalized. It is worth mentioning that openai and g42 have previously held negotiations. Openai plans to raise 8 billion to 10 billion US dollars from the latter, but the progress of the discussions is currently unknown.
Another person familiar with the matter revealed that Altman also consulted Microsoft on the plan, and the latter was very interested in it.
Altman's plan, also includes cooperation with well-known chip manufacturers in the industry, so that the production network covers the global . The report pointed out that Intel, TSMC, and Samsung may become potential partners of openai.
According to South Korea's Dong-A Daily News today, Altman will also visit Seoul, South Korea this week, during which he is expected to meet with SK Group Chairman Cui Taiyuan. The two parties may discuss AI chip cooperation - before SK Hynix, a subsidiary of SK Group It has always been Nvidia's sole supplier of hbm. Although Nvidia has begun to expand hbm suppliers to the three major storage leaders, SK Hynix is still one of the main suppliers.
Building an advanced semiconductor manufacturing plant can cost tens of billions of dollars, and building the associated production network can take years. In this case, most other AI companies have chosen another convenient path: Amazon, Google, and Microsoft all design their own customized chips and then outsource the manufacturing business to chip production companies.
However, hardware has always been one of the main constraints on the development of AI, especially chips. Altman has said many times before that chip supply is not enough to meet OpenAI's demand.
’s plan to raise funds to produce chips this time also further reflects the concerns of the leading AI CEO - as the AI wave gradually evolves, there will be a lack of sufficient chips for deployment. Judging from the current supply and demand expectations for some AI-related chips, in the future Related chips are likely to be in short supply.
▌ US chip stocks rose together
Openai and other customers here are worried about insufficient chip supply, while US semiconductor stocks are booming over there.
On the latest U.S. stock trading day (January 19, local time), amd rose by more than 7%, and Nvidia rose by 4.17%. The stock prices of both companies once again set a new record high ; Qualcomm, GlobalFoundries, Micron, Intel, TSMC, etc. followed suit. rise.
Taking amd, which has the largest increase, as an example, many institutions have recently raised their target prices and are optimistic about the prospects of product demand. Beth Kindig, a technology stock analyst at
, predicts that shipments of mi300 chips will reach 300,000-400,000 units this year, and Nvidia h100 shipments will reach 1.5 million-2 million units.
keybanc raised the target price of AMD from US$170 to US$195. Major companies have strong demand for AI servers and are confident of future potential. Barclays Bank raised its target price for AMD's share price from US$120 to US$200, predicting that its AI chip sales will reach US$4 billion this year. U.S. investment banks Cowen and Susquehanna Financial raised AMD's target prices to US$185 and US$170 respectively. In the view of analysts, under the customer's demand and expectations for the second supplier besides Nvidia, mi300 is becoming a model for generating artificial intelligence. The market for “increasingly powerful” alternatives.
TSMC also said last week that the demand for AI and HPC will bring a huge boost to its performance. The company expects revenue growth in U.S. dollars in 2024 to fall in the 20%-25% range; inventory will return to a stable level in 2024 , will be a steady growth year; in the next few years, TSMC's sales are expected to have a compound annual growth rate of 15% to 20%.
(Science and Technology Innovation Board Daily Zheng Yuanfang)
