The growth momentum of NVIDIA, a large US semiconductor company, continues. The company has received high ratings in all evaluation projects of artificial intelligence (AI) computing semiconductors, with a total market value exceeding Intel and competing with South Korea’s Samsung Electronics, which is the second largest in the semiconductor industry by market value. The difficult period for Nvidia to clean up the virtual currency bubble has passed. Including Britain’s ARM’s acquisition negotiations, Nvidia’s ambitious semiconductor leader replacement plan is gradually taking shape.
At present, there is a semiconductor in the semiconductor industry that AI researchers all over the world want to try and are evaluated as a “monster”. That is the GPU (graphics processing unit) “A100” released by Nvidia in May for data centers. Google and Microsoft, which can use the product via cloud services, have successively filed applications.
What are the capabilities of “monsters”? At the end of July, the organization “MLPerf”, an organization that evaluates the performance of semiconductors for AI computing, announced evaluation results in eight areas including object detection, translation, and recommendation display (Recommendation). In all projects, A100 has achieved the highest performance as a commercial semiconductor. A Japanese AI engineer said in amazement, “not only suppressed the competition, but also thrown off the opponent.”
To talk about the current Nvidia, AI must be mentioned. From May to July 2020, the sales of “products for data centers”, which are mainly AI computing semiconductors, surpassed Nvidia’s main business “game semiconductors” for the first time. Although the sales of Nvidia gaming semiconductors are also growing, the sales of semiconductors for data centers have soared to 24 times in 5 years.
The stock price keeps rising
Nvidia’s stock price also continues to rise. On July 8, the total market value surpassed the powerful competitor Intel, and began to compete with Samsung in August. On August 27, the total market value reached US$311.6 billion, ranking third after TSMC and Samsung among global semiconductor companies.
Nvidia’s rapid development in the AI field is inseparable from Jensen Huang, who is known as “CEO who always wears a leather jacket”. Huang Renxun founded Nvidia in 1993, starting from the GPU that enables smooth display of game images, creating a semiconductor company with a unique business model.
NVIDIA not only provides semiconductor chips, but also establishes a basic system including software. If AI semiconductors are delivered directly, only a part of researchers can use them proficiently. Nvidia has also developed software that allows more people to use semiconductor performance.
Taking the application of computed tomography (CT) image analysis as an example, Nvidia has developed software that can visualize scan results, thereby lowering the barriers to the introduction of semiconductors. As a rival semiconductor company, even if it develops software to drive semiconductors, it will not develop software for customers.
In mid-April when the new crown epidemic broke out and people on the street disappeared. A block away from Nvidia’s Silicon Valley headquarters, the head of Nvidia quietly established a data center for AI computing. Together with the company’s A100 GPU and AMD’s CPU (central processing unit), the internal supercomputer is an electronic component of Mellanox Technologies (Israel) that Nvidia acquired in April with an investment of US$6.9 billion.
Mellanox Technologies’ strengths are the components that support internal communications in the data center. Nvidia decided to make an acquisition in March 2019 when the company’s stock price and performance were sluggish, so it did not attract much attention outside the semiconductor industry. To regard the data center as the basic unit of calculation, such components are indispensable. Nvidia builds its own facilities to demonstrate its value to customers. In the latest world supercomputer rankings, this supercomputer ranks 7th and is also used for research on new crown treatment drugs.
Nvidia is currently negotiating to acquire British ARM, and the result will determine the future direction. If ARM’s design capabilities can be obtained, it will not only strengthen the GPU, but also give NVIDIA the opportunity to get involved in the CPU field. Customers using huge data centers-Google and Amazon are all trying to produce semiconductors independently, so the acquisition of ARM is of great significance to NVIDIA’s continued leadership.
Nvidia has not publicly acknowledged the acquisition negotiations for ARM. However, at the earnings conference on August 19, Huang Renxun praised ARM’s power-saving performance, full of ambitions: “Because of the excellent energy efficiency, there is a lot of room for performance improvement. In short, we like to work with ARM.”