The restructuring of the chip industry chain is staged: NVIDIA and its opponents

Economic Observer reporter Zhou Yingmei reported that NVIDIA has refreshed its previous record. On February 17, NVIDIA (NASDAQ: NVDA) released its fiscal 2022 Q4 financial report and full-year financial results, which showed that its single-quarter revenue and full-year revenue hit new highs.

NVIDIA's Q4 fiscal 2022 revenue reached $7.64 billion, up 53% year-over-year. For the 12 months ended January 30, 2022, revenue was $26.91 billion, an increase of 61% year-over-year. At the same time, NVIDIA showed super earning power, with a net profit of $9.752 billion for the whole year.

The AI-powered data center business is NVIDIA's fastest-growing business in recent years. This also means that Nvidia's "chip + AI" capabilities are being released. Wu Quan, an investment research expert in the semiconductor industry with about 14 years of industry experience, told the Economic Observer that NVIDIA's AI business makes money because of the GPU foundation. In the future, only a few giants with basic chip capabilities will be able to compete with NVIDIA for the capabilities and markets of "chip + AI".

At present, the global semiconductor market is undergoing a major reshuffle. Under the trend of large-scale semiconductor companies, giant mergers and acquisitions will continue. The challenges for Chinese semiconductor companies are also getting bigger.

AI capabilities are unleashed

The data center business is NVIDIA's second largest source of revenue and the fastest growing business in recent years. Revenue reached $3.26 billion in the fourth quarter of fiscal 2022, up 71% from the same period last fiscal year and 11% from the previous quarter. Revenue for the full year of this business increased by 58% to $10.61 billion.

The main service provided by NVIDIA is AI systems. Customers include Meta, Microsoft, etc., for search and training, and the open-source NVIDIA FLAR, which provides a common AI model for companies in the medical, manufacturing and financial sectors.

The further deepening of the digital economy presents a huge opportunity for NVIDIA. "The ubiquity, cloudization, and edge computing capabilities of data capabilities and algorithmic capabilities all require corresponding computing power. The processing of computing power has corresponding requirements for timeliness, depth and degree of intelligence, which is exactly the strength of NVIDIA. Wu Quan told reporters.

Wu Quan said that the original data form is some structured data, the most common is text. Now it is more traffic-based content, voice, video is widely used, these are partly unstructured data. This places increasing demands on the resiliency, storage, and other capabilities of data centers. "This also means that for the entire chip, the entire processing system or operating system, the entire architecture system has relatively high requirements, and this is precisely the GPU strength of NVIDIA's long-term layout."

"The strong companies that deal with text are AMD or Intel, which are traditional companies that are known for their CPU capabilities." Wu Quan said.

NVIDIA's AI business has achieved strong profitability, which is quite different from some domestic AI companies. Wu Quan believes that this is because NVIDIA's AI is enhanced based on GPUs, which have a foundation and an ontology. AI is a value-added process. However, Nvidia's profitability does not rely on AI capabilities, but through the GPU to empower and enhance AI, which is relatively easy. The implementation of AI algorithms or computing power can also be landed, which is a coupling or matching.

Gartner analyst Sheng Linghai told the Economic Observer that ai (including deep learning and machine learning) this road is excavated by NVIDIA's artificial intelligence scientists with self-developed chips, and NVIDIA drives AI change mainly in chip power.

Many AI companies are modeled after NVIDIA. "Now some domestic companies that do artificial intelligence want to make artificial intelligence into a product, the problem is that it can't make a product, because its core is deep learning, and there is no general artificial intelligence." Every project to do will definitely lose money, this is still very expensive and resourceful. Sheng Linghai said.

Wu Quan said that the above situation still has some wake-up call for domestic AI companies. They need to clearly understand that AI is only an enhanced capability, it needs a carrier, to have basic capabilities, and then to use AI as an enhanced capability. For example, Inspur and Sugon are also doing AI, but it is based on the server base.

"Domestic companies that do pure AI turn algorithms into electronic chips, and the value of this demand may also need to be questioned." Wu Quan said.

Sheng Linghai mentioned that from the results, many companies that do artificial intelligence chips, including Intel, Ali and Baidu, currently have relatively small shipments. "The main reason is that the competition is not NVIDIA, and it is impossible to make a profit without volume." It's a loop. ”

moat

In the past, Nvidia has always had the title of "King of Graphics Cards". In 1999, Nvidia released its first GPU (graphic processor). Since then, GPUs have become the core products of NVIDIA.

Today, Nvidia's main source of revenue is gaming graphics. In the fourth quarter of fiscal 2022, gaming revenue reached $3.42 billion, up 37% from the same period last fiscal year, breaking the previous record. For the full fiscal year 2022, revenue was $12.46 billion, up 61% year-over-year.

"The GPU that Nvidia makes now is not exactly the same concept as the GPU of that year." Wu Quan explained that what may have been just a graphic processing in the past is now based on its entirely new architecture. "NVIDIA has built a new era of computing power architecture, based on GPUs, forward can enhance AI, backward to a certain extent fused with THE CPU, it has made a downward extension of the ability." NVIDIA firmly grasps the market share and market share of the GPU, and also firmly grasps the architecture and system of the GPU.

NVIDIA's chip+AI capabilities are bringing some changes to the industry. Wu Quan said that for example, some large-scale data processing is similar to supercomputing to a certain extent, and the computing power based on the algorithm is already very strong, and the growth rate is also very fast. The degree to which market or application needs are met is also increasing.

Who can compete with NVIDIA? On February 17, 2022, Nvidia's market capitalization was higher than Intel+AMD+Xilinx combined. From the Q4 financial reports recently released by Intel and Nvidia respectively, Nvidia's revenue and net profit are actually not as good as Intel. Wu Quan analyzed that such a high valuation reflects the market's optimism about its future growth.

Wu Quan believes that in the future, in terms of chip + AI capabilities, NVIDIA is still likely to be caught up, but it is still traditional semiconductor giants that can compete with NVIDIA in this market, such as Intel and AMD.

"First of all, we have to look at the chip capability. This is a basic ability, and there are actually only a few players who have this basic ability. Wu Quan believes that AI cannot be defined as an industry alone, it is just a tool and capability, and it will become a standard in the future, "I have AI and others do not have the possibility of getting lower and lower." AI will only be a basic skill, but it will be enhanced on the basis of the existing chip capabilities and existing scene application capabilities. ”

Nvidia also has a long-term layout in the automotive field. However, due to the impact of Tesla, which it works with, the automotive business revenue has continued to decline last year to this year. On February 16, Nvidia announced a partnership with Jaguar Land Rover. Sheng Linghai believes that NVIDIA's advantage in the automotive field is still autonomous driving, which is still in the early stages, mainly for high-end cars and some manufacturers to cooperate.

"Autonomous driving is divided into level 1, 2, 3, 4, level 1 and 2 may also use Intel's solution, but more than 3 is basically to use NVIDIA's car chip solution, which is a very real problem." Sheng Linghai said that unless it is like Tesla to make its own chips, it has its own data, generally high-end cars have this function, the cost is relatively high, whether other companies can afford it is also a problem.

Many domestic brands are using NVIDIA's platform, including NIO's ET5 sedan, Xiaopeng Motors' G9 SUV and Xiaoma Zhixing's robot taxi fleet, all of which are using NVIDIA DRIVE Orin. Nvidia's NVIDIA DRIVE Hyperion platform provides AV systems to automakers, including Desai, Flextronics, Quanta, Valeo and ZF.

In terms of professional visualization, NVIDIA open platform Omniverse mainly provides 3D design collaboration and virtual simulation for creators, which is also the hottest meta-universe entrance at present. The biological field also supports the construction of 3D cell simulations. This business contributed $643 million in revenue to Nvidia in the fourth quarter of fiscal 2022, an increase of 109% over the same period last fiscal year, and full-year revenue of $2.11 billion, an increase of 100% year-on-year.

Big changes in the semiconductor industry

Over the past year or two, the semiconductor industry has seen a series of acquisitions.

Previously, Nvidia also planned to spend a sky-high price on the acquisition of Arm, so it worked for a year and a half, but eventually abandoned the acquisition because of regulatory challenges. Acquisitions are not unique, though. On February 14, AMD completed the transaction with Xilinx, and on February 15, Intel announced the acquisition of tall towers.

Sheng Linghai said that the acquisition is generally going in the direction of integration. After Moore's Law comes to an end, players in the entire semiconductor market also have to find a direction for their future and strengthen their position competitiveness - this is the most critical.

Wu Quan further explained that the deepening of the digital society has more and more demand for semiconductors, and the dependence on semiconductor capabilities is also deepening. The deepening of the degree of intelligence of the digital economy and the entire society has in turn made the degree of semiconductorization of the global industry more clear. These semiconductor companies will move towards scale, large companies and large companies will merge, and next, large companies will carry out a series of mergers and acquisitions.

In the specific acquisition case, Wu Quan believes that the success of NVIDIA's acquisition of Arm has little impact on NVIDIA. Necessity or strategy is not particularly strong. It is only said that after its acquisition, it will enhance its volume, supplement a piece of business, and enhance its mobile presence.

"Most people in the industry still believe that this acquisition should not be passed, and no one wants to see it succeed." Sheng Linghai said that once the acquisition is successful, Arm's current model is easy to form a monopoly, "chip manufacturers with Arm may have a more difficult life, depending on its eyes." ”

For Intel's acquisition of Gaota and AMD's acquisition of Xilinx, Wu Quan said, "Their play style is relatively sharp, and what I mean by sharpness should be more strategic." Intel actually has an absolute monopoly position in the CPU space, accounting for at least 85% of the market share. Of course, Intel went to acquire the tower to make up for its IFS (foundry service) capabilities. It will compete with TSMC for oem services. ”

"The top two fights are often injured in the third place, Intel's acquisition of the tower, although it competes with TSMC, but the most affected companies may be SMIC." Wu Quan analysis.

AMD is smaller than Intel, and its strong rise in the CPU field in recent years has also begun to need some tools to support it. Compared with Intel, AMD has taken a differentiated route, and the acquisition of Xilinx is mainly to supplement FPGAs and enhance its chip intelligence capabilities, or customization capabilities. This will form a differentiated competition with Intel.

Wu Quan believes that these two acquisitions have not had much impact on NVIDIA. However, for China, there are very few FPGA capabilities, and they are relatively weak in technology, IP accumulation and market capabilities.

Therefore, Wu Quan reminded that China needs to be highly vigilant against a new round of mergers and acquisitions of semiconductor giants and the reshuffle in the semiconductor field. Because our industrial landscape is often a commercially separate structure for design, manufacturing, and packaging and testing, there is a great deal of dependence on these manufacturers. "Domestic companies need to look at other people's paths, not to benchmark the present, but to benchmark the paths and stages that others have walked 10 years ago, 20 years ago, when the same volume."

Recently, two semiconductor-related documents were released, one is the U.S. Competition Act of 2022 passed by the U.S. House of Representatives, which plans to invest $54 billion to support the local semiconductor industry. The other is the European Union's "Chip Act", which plans to invest 43 billion euros, also around local semiconductors. "Both documents focus on local giants, local manufacturing, or simply around manufacturing." Wu Quan believes that this will reconstruct the structure of the global industrial chain, especially the division of labor in the regional industrial chain. Previously, the supply chain circulation will involve various countries, and the intermediate links will disappear after localization. For domestic enterprises, the convenience of purchasing chips and the controllability of the supply chain will also be much worse.