"Concerned about the shortage of artificial intelligence chips, (U.S. artificial intelligence research company) OpenAI is discussing solutions." On the 21st, India's "Mint" compiled reports from multiple foreign media that OpenAI's solution is to establish a chip manufacturing company, and OpenAI CEO Sam Altman is persuading potential investors to join this plan. As businesses and consumers become increasingly interested in artificial intelligence (AI) applications, demand for AI chips is also rising strongly.
However, at present, these demands are almost all concentrated on the top one or two companies in the industry, and the supply of related chips exceeds the demand. This has prompted AI R&D companies represented by OpenAI to begin to consider embarking on the path of producing their own AI chips.
Ambitious plan for network of semiconductor manufacturing plants
Bloomberg reported on the 20th that OpenAI CEO Sam Altman is negotiating with global investors, hoping to raise billions of dollars to create a network of factories for manufacturing semiconductors.
Negotiations are still at an early stage and the full list of partners and investors involved in the project has not yet been determined, the report said. However, some people familiar with the matter revealed that the project involves cooperation with a number of top chip manufacturers, and the completed network of manufacturing factories will cover the world. The Mint reported that OpenAI hopes to get rid of its dependence on US chip manufacturer Nvidia and diversify its supply.
Altman has started discussions with Abu Dhabi-based artificial intelligence company G42 and Japan's SoftBank Group, and has also held talks with some investment firms in the Middle East, people familiar with the matter said. South Korea's Chosun Ilbo reported on the 21st that whether South Korean semiconductor company Samsung Electronics will participate in the establishment of this semiconductor manufacturing factory network is the focus of the industry. Sources revealed that U.S. chip company Intel, Taiwanese chip foundry TSMC and South Korean chip manufacturer Samsung Electronics are all potential partners of OpenAI.
Altman had been working on the project before he was briefly dismissed as CEO of OpenAI in November. After his return, he quickly restarted the project. Two people familiar with the matter said Altman had tested Microsoft's attitude towards the plan, and Microsoft expressed support for it.
Building a state-of-the-art semiconductor manufacturing plant alone could cost tens of billions of dollars, and creating a network of manufacturing plants of that scale would take longer and require more money. People familiar with the matter told Bloomberg that OpenAI hopes to raise $8 billion to $10 billion in investment from G42, but it is unclear how the negotiations are progressing.
Artificial intelligence craze drives demand for high-end chips
Since OpenAI released the general AI large model ChatGPT, the interest of enterprises and consumers in artificial intelligence research and applications has skyrocketed, which in turn has stimulated the demand for artificial intelligence chips.
People familiar with Altman's thinking told Bloomberg that Altman believes that the artificial intelligence industry needs to take immediate action to ensure that there is an adequate supply of cutting-edge chips by the end of the century. He has publicly stated many times that the current chips are not enough to meet OpenAI’s AI research and development needs.
The US CNBC website stated on the 19th that US technology company Meta is spending billions of dollars to purchase Nvidia’s high-end chip H100, which is the core of AI research and development. Meta CEO Mark Zuckerberg said the company’s AI future plans include building a “large-scale computing infrastructure” that will include 350,000 Nvidia H100 chips by the end of 2024. According to industry analysts, Nvidia's H100 chip sells for about US$25,000 to US$30,000, and can be sold for more than US$40,000 on second-hand platforms. Even if Meta purchases it at an intermediate price, the expenditure will be close to US$9 billion. Moreover, the supply of H100 chips is limited.
U.S. stocks closed higher across the board on the 19th, with the S&P 500 index reaching a two-year historical high. According to Reuters analysis, this is because the market is optimistic about the development of AI, driven by the rise of chip manufacturers and other heavyweight technology stocks. The day before, TSMC predicted strong demand for high-end AI chips, which also contributed to the recovery of chip stocks.
The latest data from the U.S. Semiconductor Industry Association shows that global chip sales increased for the first time in 15 months and demand is rebounding. Global semiconductor revenue in November last year was US$48 billion, a month-on-month increase of 2.9% and a year-on-year increase of 5.3%. Deloitte, an internationally renowned accounting firm, predicts that by 2024, total AI chip sales will account for 11% of the global US$576 billion chip market.
"Building a cutting-edge semiconductor manufacturing facility is extremely challenging."
Among the top ten companies by market value, many companies including Nvidia, Google, Apple, Facebook parent company Meta, Amazon, Microsoft and Tesla are heavily involved in chip design. However, due to cost considerations, Amazon, Google and Microsoft have typically focused on customizing their own silicon and then outsourcing manufacturing.
The website of the Center for Strategic and International Studies, an American think tank, published an article on the 19th stating that the current focus of chip design and R&D is that designing semiconductor circuits requires a large amount of R&D expenses - the use of electronic design automation tools, the use of intellectual property rights and labor costs, etc., all of which will continue to increase with the advancement of semiconductor technology. For example, developing a 7-nanometer chip requires approximately US$223 million, while developing the next-generation 3-nanometer chip requires US$650 million, which is approximately three times the cost of developing a 7-nanometer chip. In addition, making an advanced central processing unit (CPU) usually requires a skilled design company to spend several years, and it also takes several years to integrate the CPU onto the chip, which drives up the cost of time and money.
Some analysts say that it is extremely challenging for OpenAI to quickly establish a cutting-edge semiconductor manufacturing factory. According to the current news, it is not known whether Ultraman’s plan is to directly purchase mature foundries to manufacture chips, or to cooperate with chip manufacturers to produce chips, but either way will require a lot of money and time.
Regarding the future development of the semiconductor industry, Deloitte warned that there are several points to note. First, the market for generative AI chips in 2023 is characterized by basically having only one designer, and the designer relies on a manufacturer with limited production capacity. As buyers acquire as many chips as possible and new suppliers enter the market and increase production capacity, prices may fall, impacting revenue in 2025 and beyond. Second, when customers are waiting to receive their goods, they often overorder. Once supply and demand for AI chips become balanced, buyers may be able to acquire far more chips than they need and then order less as new capacity comes online. This is part of the "bullwhip effect" (a phenomenon of amplifying demand variation in the supply chain), which is also a reason for the extreme cyclicality in the history of the chip industry. Third, almost all AI training and calculations are currently performed using the same generative AI chip, but as time goes by, more advanced GPUs (graphics processing units), CPUs, or other new processors may appear, resulting in an oversupply of currently used AI chips. Finally, some people believe that the strong demand for AI chips in 2023 and 2024 is a bubble, and there is the possibility of a depression by 2025. Although this voice is not a mainstream view, it is worthy of vigilance.