What company can make Nvidia's Lao Huang stand on the platform in person? Huang Xiaogui's generative AI heat wave made Nvidia popular: last Friday, the market value exceeded $2 trillion. However, there is one company that is even more fierce than Nvidia. It is the ultra-microcomputer known as NVIDIA's "own son". From last year to the present, its stock price has risen by as much as 13 times, surpassing Nvidia; The market capitalization rose from $5 billion to $506. Fiscal year 2023 revenue of 71$200 million, with fiscal 2024 revenue expected to be $14.3 billion to $14.7 billion.
Gelonghui.
When many people hear about ultra-microcomputers, their first reaction is: Is it AMD? AMD is called "Chaowei Semiconductor", and Supermicro Computer is called Supermicro, abbreviated as SMCI. AMD designs chips, and SuperMicro Computers sells server solutions. For example, NVIDIA's GPU is not simply connected by wires, but is a group of several pieces, plugged into the rack of a supermicro computer, equipped with other parts to make a server, and sold to customers such as OpenAI, Google, Microsoft, and Meta for large language model training or cloud computing. Or Microsoft has tens of thousands of GPUs on hand, and ask a supermicro computer to come to the door to build a room of GPUs and make rows of servers.
The rack of the supermicro computer can be made into servers for different scenarios, such as chips from NVIDIA, Intel and AMD. The performance of the server made with the ultra-micro computer rack is stable, and the combination runs smoothly. The ultra-micro computer also has a unique liquid cooling technology, which reduces the power consumption of the GPU and allows enterprises to greatly save costs. Some analysts have pointed out that with the large number of GPUs required for generative AI and the high level of heat generation, next-generation data centers must be cooled by liquid cooling instead of air cooling. The reason why Supermicro Computer is called Nvidia's "pro-son" is because the relationship between the founders of the two companies, Jensen Huang and Liang Jianhou, is extraordinary, and Supermicro Computer can give priority to buying Nvidia's GPU, so that Supermicrocomputer's products are bought off the production line. In the second fiscal quarter of 2024, Supermicro's revenue doubled year-over-year to 36$600 million – half of that for AI. Born in Chiayi County, Taiwan Province in 1958, Liang graduated from Taipei Institute of Technology in 1978 with a degree in Electrical Engineering (a five-year junior college) and later received a bachelor's degree in electrical engineering from the Taiwan Institute of Technology and Technology. In 1984, he went to the United States to study and received a master's degree in electrical engineering from the University of Texas at Arlington. After that, Liang Jianhou worked as a senior engineer and project leader at C&T, a major display chip manufacturer, and Suntek, a major software manufacturer. In 1991, he served as president and chief engineer of Micro Center Computer, a motherboard design and manufacturing company. In 1993, he and his wife Liu Qiuzhu founded Supermicro Computer in San Jose, California, USA. In 1995, the world's first server motherboard that could be plugged into two CPUs, later launched a motherboard that could plug in four CPUs, and later launched a motherboard that could operate with CPUs and GPUs at the same time. In 2004, Liang cut more of the company's business and focused on more energy-efficient server systems. It is said that after watching the disaster movie "The Day After Tomorrow" with his children, he suddenly thought: What can he do about global warming? He and his technical team developed an energy-saving power supply device. Liang Jianhou often said: There is only one Mother Earth. In 2007, Supermicro Computer was listed on the NASDAQ. In 2009, Supermicro launched an epoch-making product: a super server rack with multiple GPUs. Supermicrocomputers can take a few bites of cream in the big cake of today's generative AI, and the foundation was laid back in 2004 and 2009. Today's supermicro computer server racks can be spliced together on a large scale like Lego bricks, equipped with power supplies, heat dissipation modules, motherboards, and chassis. As many chips as customers have, the supermicro computer can customize the corresponding server racks to carry them, and the samples can be submitted in one month. In 2018, Supermicro won the order to build Intel's data center. Supermicro Computer estimates that Intel can save $13 million a year on electricity bills by purchasing their products. During this period, the supercomputer thoroughly understood the principle of the liquid cooling system used in the supercomputer and applied it to the server, gradually replacing the air cooling. Liquid cooling has a better cooling effect, it is not easy to have heat dissipation dead ends, and it can also save the space for installing a fan in the case and the electricity cost of turning on the fan. Liquid cooling is the ability of ultra-micro computers, and some analysts of ** companies believe that the performance of ultra-micro computers has soared, and many customers are attracted by liquid cooling - after all, server Dell, HP, IBM can also do - customers do not want to be in the critical stage of large language model competition, hardware forks. Liang and Huang have a deep relationship, and the two started their own businesses in 1993, and the two companies are only 10 miles away. The motherboard of the supermicro computer has been equipped with Nvidia's GPU for a long time, and no matter what Nvidia does, the supermicro computer will follow. Later, Nvidia made AI GPUs, and no one was optimistic, so Liang helped to help produce servers that could combine NVIDIA GPUs for AI training. Huang has appeared at Supermicro's events on several occasions and acted as a "salesman" for Supermicro's server solutions. Liang Jianhou has three younger brothers, all of whom have founded businesses in Taiwan and worked as a ** business for supermicro computers. The eldest brother Liang Jianfa's Daxun Technology mainly makes chassis and cooling systems, the second brother Liang Jianda's Kenwei Technology makes power supplies, and the younger brother Liang Jianguo's Haoran Technology proofs Daxun's chassis. Liang saw that he was very strict with his younger brothers. According to local ** reports, in the early days of entrepreneurship, the chassis of Daxun Technology had just been designed, and Liang Jianhou asked Liang Jianfa to fly to the United States immediately with the chassis, and at most he flew three times a month. Liang Jianfa went to pick up a cup of coffee before the meeting, and was 30 seconds late when he entered the conference room, and was scolded by Liang Jianhou in front of everyone, and he had to be punished for standing. Once, when he wanted to change the product design, Liang Jianhou ignored Liang Jianfa's business trip in Shanghai and asked him to fly back the next day to discuss the plan. It is the demanding requirements for products that allow Supermicro Computer to survive the competition with giants such as Dell, HP, and IBM, and make arrangements in advance to catch up with this round of generative AI opportunities. This article does not constitute any investment advice. The cover image is from the WeChat gallery. This article has participated in reports such as Gelonghui and Business Today, and I would like to thank you.