In-depth analysis of "Eternal Second" AMD AI computing power advantage: the market is suffering from Nvidia's monopoly!

Key Points

1. AMD has completed a comprehensive layout in the AI chip field for cloud and edge computing. The MI300 APU, which will be launched in the fourth quarter of this year, will compete head-on with Nvidia's products, which is highly anticipated by the market. This has also led to AMD's market performance being better than its traditional rival Intel, which is relatively weak in the business.

2. Customers are unhappy with Nvidia's monopoly, and the willingness to cultivate secondary suppliers is a favorable condition for AMD to catch up quickly. The AI chip market has a large enough capacity, which allows AMD to participate in the competition even in the presence of a disadvantageous software ecosystem. AMD has demonstrated its ability to find differentiation and catch up in research and development for a long time, and is expected to replicate this in the new AI chip field.

3. The visibility of the traditional PC and server business hitting bottom is still low, but AMD is confident in the recovery of server demand in the second half of the year. In the enterprise server market, AMD's products' higher cost-effectiveness compared to Intel will gradually become prominent, and it is possible to see AMD grab more market share in 2024.

Q&A Section

Q: Can you briefly talk about why we are paying attention to AMD?

A: Paying attention to AMD now is definitely to see its follow-up development in the AI GPU field. The market is focusing on the absolute leader Nvidia, and our judgment is that this market will definitely not be monopolized by one company. Therefore, as a follower, AMD also has a lot of potential and room for expectation to explore.

AMD is an interesting company, jokingly referred to as the "eternal second" in the industry. Whether it is competing with Intel in the CPU field or with Nvidia in the GPU field, it has never taken the dominant market position and has always been a secondary supplier. However, it can sit firmly in the second place on multiple fronts, demonstrating its research and development capabilities and ability to produce differentiated products.

At the end of last year, its market value surpassed its old rival Nvidia, and its market performance was also very impressive, even in the case of relatively low demand for PC and server. Everyone has high expectations for its development in the AI GPU field.

Q: Can you elaborate on AMD's layout in AI computing chips?

A: First of all, in terms of AI computing chips, the product that the market is most concerned about is the Instinct MI300, which will be launched in the fourth quarter of 2023. Strictly speaking, it is not a GPU product, but an APU accelerator processor, which can be simply understood as integrating CPU and GPU together and amplifying their respective advantages. It can be used to train large AI models.

This product should be a turning point for AMD's AI computing chips. Although it has had MI100 and MI200 products before, they have basically had very low shipment volumes and no outstanding product strengths. By constantly replacing products until a certain generation suddenly becomes very competitive, AMD has done this in the server market before. This time, we can see if it can replicate this in the AI chip field. Now we have limited information about the MI300 that is available to the public. AMD will have a data center and AI technology release on June 13th, which will increase visibility.

In addition, through the acquisition of Xilinx in 2022, AMD became the leader in the FPGA field and integrated Xilinx's AIveo market and technology. Future AI computing will not be limited to cloud-based AI data centers, but will also expand to the edge. Ultimately, our phones, smart home products, computers, and even future XR glasses should all have some AI inference computing capabilities. Xilinx's products are very competitive in this market.

Therefore, we can see that AMD has a comprehensive layout in AI computing hardware, from data centers to the edge. They also said in their external communication that Lisa attaches great importance to AI and has established an AI department, which is headed by former Xilinx CEO Victor. Basically, the company's resources will be tilted towards this area.

Q: The reason why everyone is optimistic about NVIDIA is not only because of its hardware leadership, but also because of the CUDA software ecosystem it has established over the years. How can AMD compete with NVIDIA in AI computing chips?

A: To be honest, we think that in the short term, it is difficult to break through the moat established by CUDA, so AMD may still play a secondary role in AI chips.

But we don't think we should underestimate AMD's potential in this market because AI computing chips will definitely be a huge market, and it will not be a market that blindly pursues performance without considering price.

Our feeling is that the parameters of large models will not continue to grow infinitely, but the focus of future considerations will be how to achieve similar effects on smaller models, so that after commercialization, it can really bring profits to investment enterprises. Burning money recklessly is definitely not a long-term solution.

The current situation is that NVIDIA's dominance in the AI training field has made many customers very uncomfortable and they are eagerly awaiting the emergence of a secondary supplier. Either put pressure on NVIDIA to reduce prices (which is currently difficult mainly due to production capacity constraints), or the secondary supplier's products are more cost-effective, as long as the performance can be basically equivalent and the software ecosystem can be compatible.

Previously, Microsoft had purchased its MI200, so if AMD's MI300 can be launched as scheduled in Q4 and achieve large-scale shipments quickly, it will not be a very surprising thing.

In the long run, in fact, large companies do not want to be too deeply bound by the CUDA ecosystem. Some capable companies hope to have their own ecosystem, so they may also choose to cooperate with AMD to customize some hardware products. Q: How is AMD's basic business besides AI-related new businesses? Their inventory levels are quite high in both their and Intel's latest financial reports. Does this mean that the bottom of traditional PC and server businesses is still not visible?

A: It is difficult to judge the bottom of demand now, which is a major characteristic of this year's investment. Uncertainty is greater than ever before, and even the market and the Fed are constantly revising their expectations for US inflation and interest rates. I think this is the market environment that we have to accept. Either we take a long-term perspective and "hold on", or we can choose to participate when the situation becomes clearer. It is difficult to say which strategy is wiser.

Regarding PC and server businesses, AMD is still relatively optimistic at their communication meeting. They believe that the PC business will improve sequentially in the future, and server demand will improve significantly in the second half of the year. Therefore, the second quarter should be the high point of their inventory level for the whole year. The current high inventory level is also for stocking up for the Genoa platform switch in the second half of the year.

Q: AMD has been trying to eat into Intel's enterprise server market for a long time. Is there any new progress in this regard?

A: In the enterprise server market, Intel has accumulated a lot, which is supported by a very large sales team and site engineer team. This is not AMD's expertise. We think that if AMD wants to grab this market, it can only do it step by step.

Actually, AMD has started to do differentiated marketing last year, directly targeting enterprise customers' CIOs and promoting them at events where some CIOs often attend. The cost-effectiveness advantage of AMD's products may be more prominent in the stage where the overall macro economy begins to decline. Even if Intel fully uses the Sapphire Rapids platform, this still applies.

At that time, it will be difficult to maintain market share purely based on customer relationship. Therefore, we think that by 2024, we will see AMD strive for more market share in the enterprise server market.