OpenAI Considers Founding a Chip Manufacturing Enterprise Amid AI Chip Shortage
According to a report published on the 21st by “The Mint,” which aggregated stories from various international media, OpenAI—an American artificial intelligence research company—is looking for solutions to the looming artificial intelligence (AI) chip shortage. OpenAI’s proposed remedy is to establish a semiconductor manufacturing company, and the organization’s CEO, Sam Altman, is actively engaging potential investors to support the initiative. The increasing interest in AI applications from both businesses and consumers has led to a surge in demand for AI chips. Currently, this demand is largely concentrated on a couple of industry-leading companies, creating a significant supply shortfall. This has prompted AI research and development firms, including OpenAI, to consider producing their own AI chips.
Ambitious Plans for a Global Semiconductor Factory Network
Bloomberg reported on the 20th that Sam Altman of OpenAI is in discussions with global investors to raise billions of dollars with the aim of establishing a network of semiconductor factories.
These talks are still in the early stages, and a comprehensive list of partners and investors has yet to be finalized. However, insiders reveal that the project involves collaborations with several top-tier chip manufacturers to construct a global manufacturing network. “The Mint” has reported that OpenAI seeks to decrease its reliance on Nvidia, a major U.S. chip provider, in favor of a more diverse supply chain.
Insiders have indicated that Altman has entered discussions with G42, an artificial intelligence company based in Abu Dhabi, and Japan’s SoftBank Group. He has also conversed with various investment firms in the Middle East. On the 21st, South Korea’s “Chosun Ilbo” highlighted that whether Samsung Electronics, a major Korean semiconductor company, will join this semiconductor network remains a subject of industry focus. It has been disclosed by sources that Intel in the United States, TSMC in Taiwan, China, and Samsung Electronics in South Korea are all potential partners for OpenAI.
Before being briefly relieved of his role as the CEO of OpenAI in November of the previous year, Altman was a driving force behind this project. Upon his return, he quickly revived the initiative. Two informed sources reported that Altman has explored Microsoft’s stance on this plan, to which Microsoft has expressed support.
Constructing a state-of-the-art semiconductor manufacturing plant alone could cost tens of billions of dollars, and developing a network of such magnitude would require even more time and funding. Sources told Bloomberg that OpenAI aims to attract an investment of $8 to $10 billion from G42, though the status of these negotiations remains unclear.
AI Craze Spurs Demand for High-End Chips
Since OpenAI released its general-purpose AI model ChatGPT, interest in AI research and applications has skyrocketed among businesses and consumers alike, fueling the demand for AI chips.
Insiders privy to Altman’s plans shared with Bloomberg that he believes the AI industry needs to act immediately to ensure an adequate supply of cutting-edge chips by the end of the century. He has repeatedly stated that current chips are insufficient to meet OpenAI’s AI research and development needs.
CNBC reported on the 19th that Meta, an American tech company, is spending billions of dollars to acquire Nvidia’s high-end H100 chips, which are central to AI development. Meta’s CEO Mark Zuckerberg has shared that the company’s AI roadmap involves building a “large-scale computing infrastructure” which, by the end of 2024, will encompass 350,000 Nvidia H100 chips. Industry insiders estimate that the H100 chip is priced at about $25,000 to $30,000, with secondary market prices reaching over $40,000. If Meta acquires the chips at a median price, the expense could approach $9 billion. Moreover, the H100 chip supply is limited.
U.S. stocks rose across the board on the 19th, with the S&P 500 reaching a historic high not seen in over two years. Reuters attributes this rally to the surge in semiconductor manufacturers and other heavyweight tech stocks, reflecting market optimism for AI’s prospects. The day before, TSMC forecasted strong demand for high-end AI chips, contributing to the rebound in chip stocks.
Data from the Semiconductor Industry Association indicates that global chip sales experienced growth for the first time in 15 months, showing a demand rebound. In November last year, global semiconductor revenue hit $48 billion, a month-over-month increase of 2.9% and a year-over-year rise of 5.3%. Deloitte, a prestigious accounting firm, predicts that by 2024, total AI chip sales will account for 11% of the global $576 billion chip market.
“Creating a Cutting-Edge Semiconductor Manufacturing Plant is Highly Challenging”
Among the top ten highest-valued companies, many, including Nvidia, Google, Apple, Meta (formerly Facebook), Amazon, Microsoft, and Tesla, are heavily involved in chip design. However, due to cost considerations, companies such as Amazon, Google, and Microsoft typically focus on their custom-designed silicon and outsource the manufacturing process.
The Center for Strategic and International Studies, an American think tank, posted on the 19th that chip design and R&D are becoming increasingly expensive—the costs of electronic design automation tool usage, intellectual property fees, and labor are all rising as semiconductor technology advances. For instance, crafting a 7nm chip requires approximately $223 million, while the next-generation 3nm chip demands about $650 million—nearly three times the cost of developing a 7nm chip. Additionally, designing an advanced central processing unit (CPU) typically takes several years by a skilled design company, and integrating the CPU onto a chip can take several additional years, escalating both time and financial costs.
Analysts argue that for OpenAI at present, rapidly establishing a cutting-edge semiconductor manufacturing facility presents a considerable challenge. Current reports do not clarify whether Altman’s strategy involves simply purchasing chips from mature manufacturing services or collaborating with manufacturers, but either approach would demand substantial capital and time.
Deloitte warns of several aspects to watch for the future of the semiconductor industry. Firstly, the current generative AI chip market for 2023 is characterized by having virtually a single designer who, in turn, relies on a limited-capacity manufacturer. As buyers acquire as many chips as possible and new suppliers enter, possibly reducing prices as capacity grows, this could affect revenues post-2025. Secondly, when customers are in a backorder status, they often place orders exceeding their actual needs. Once the supply and demand of AI chips balance out, buyers might end up with far more chips than needed and might decrease orders as new capacities come online. This forms part of the “bullwhip effect,” a phenomenon of amplified demand variability in the supply chain, and is one reason for the extreme cyclicity historically observed in the chip industry. Thirdly, practically all AI training and computation currently employ the same type of generative AI chip. However, as time progresses, more advanced GPUs (Graphic Processing Units), CPUs, or other new processors could emerge, leading to an oversupply of AI chips currently in use. Finally, some believe the robust demand for AI chips in 2023 and 2024 illustrates a bubble that may burst by 2025. Although this view is not mainstream, it nonetheless warrants vigilance.