focus
Tencent Technology News on January 20, according to foreign media reports, many industry insiders familiar with the situation broke the news that Sam Altman, CEO of the artificial intelligence start-up openai, has recently been working non-stop around the world. Shuttle around the world and have close contact with various investors, aiming to raise a huge amount of funds to build a global network of artificial intelligence chip manufacturing factories.
These people familiar with the matter revealed that Altman has been in contact with a number of heavyweight potential investors, including the Abu Dhabi-based artificial intelligence and cloud computing giant G42 Group and Japan’s SoftBank Group.
In addition, this project also involves working with the world's top chip manufacturers to jointly build a seamless and efficient fab network.
Although there have been rumors in the market that Ultraman is planning to establish a chip company, the true scale and focus of this project (i.e., the chip manufacturing process) have not been revealed until today. However, negotiations are still at an early stage and the full list of specific partners and investors is still to be finalized.
Ultraman’s financing this time undoubtedly highlights his profound concerns about the future development of artificial intelligence. In his view, with the increasing popularity of artificial intelligence technology and the continuous expansion of application scenarios, existing chip production will be unable to meet the growing market demand. In fact, the current production forecast for artificial intelligence-related chips has been lower than actual demand expectations.
It is worth mentioning that the capital investment and technical difficulty required to build and maintain wafer fabs for manufacturing semiconductors far exceed the asset-light model favored by many of OpenAI's peers. Technology giants such as Amazon, Alphabet's Google, and Microsoft (also the largest investor in OpenAI) usually choose to design their own chips and then outsource the manufacturing process to professional foundries.
In the semiconductor field, building a cutting-edge manufacturing plant is a huge investment, often costing tens of billions of dollars and often taking several years.
According to previous reports from people familiar with the matter, Ultraman hopes to raise as much as US$8 to US$10 billion in negotiations with the G42 Group alone. Although the progress of the negotiations between the two parties is still unclear, it is worth mentioning that openai and g42 established a close cooperative relationship in October last year.
Ultraman has profound insights into the future development of the artificial intelligence industry. He believes that in order to ensure sufficient supply of chips globally by 2030, the industry must take immediate action. This is also an important reason why he spares no effort to promote the chip project. Even when he briefly stepped down as CEO of openai in November last year, he did not give up this effort, and immediately restarted the project after his reinstatement.
According to internal sources, Altman has also had in-depth discussions with Microsoft about this grand plan, and Microsoft has shown strong interest in it. Since openai launched chatgpt, artificial intelligence applications have triggered an unprecedented wave of attention around the world. The desire of enterprises and consumers for artificial intelligence continues to heat up, which in turn has given rise to huge demand for computing power and processors. However, Altman has admitted many times that the current chip supply simply cannot meet the strong market demand.
In the field of chip manufacturing, Intel, TSMC and Samsung Electronics are undoubtedly the three giants. They are all potential targets for openai when seeking partners. At present, the location of these chip factories has not been finalized, but if the factories can be located in the United States, it will undoubtedly greatly enhance the manufacturing strength of the United States. After all, the United States currently produces only 12% of the world's chips domestically and relies heavily on outsourcing to reduce costs.
S&P Global noted that Ultraman's project is expected to become a reality in the near to medium term as chip demand surges as semiconductors become more widely used in the automotive sector and geopolitical trade risks persist. . The main bottleneck in running artificial intelligence models in
is the lack of enough chips to handle the huge computing power requirements behind robots such as chatgpt or dall-e.These bots are able to respond to various prompts and generate text or image content. Part of the reason why Nvidia's market value exceeded the $1 trillion mark for the first time last year is that its h100 GPU occupies a near-monopoly position in models such as gpt-4, gemini, and llama 2.
With the continuous evolution of artificial intelligence technology, the demand for higher-performance chips is becoming increasingly urgent. In this race to manufacture high-performance chips to support complex artificial intelligence systems, the competition among all parties has become increasingly fierce. However, the number of fabs capable of producing high-end chips is limited, forcing industry leaders such as Altman to bid for capacity years in advance to ensure the smooth production of new chips. Competing with technology giants such as Apple,
undoubtedly requires strong financial strength as a backing. Because in this process, investors will bear high costs that the non-profit organization OpenAI cannot yet bear.
Meanwhile, other companies developing artificial intelligence models are also trying to make their own custom chips.
Microsoft was the first to announce the creation of the first custom artificial intelligence chip for training large models, and Amazon followed suit by launching a new version of the trainium chip. Google uses its DeepMind artificial intelligence to design artificial intelligence processors such as tensor processing units (TPUs), which will run on Google Cloud servers.
In the cloud service market, AWS, Azure and Google are also using Nvidia's h100 processor. This week, Meta CEO Mark Zuckerberg revealed in an interview that Meta will have more than 340,000 Nvidia H100 GPUs by the end of the year to support the company's development of artificial general intelligence (AGI). ambitions.
Facing fierce competition in the market, Nvidia has released the next generation gh200 grace hopper chip, aiming to further expand its dominance in this field. Not to be outdone, its competitors AMD, Qualcomm and Intel have launched processors designed for running artificial intelligence models on laptops, mobile phones and other devices. (Compiled/Golden Deer)
