On October 10th local time, AMD launched a series of new chips, including Instinct MI325X GPU, fifth generation EPYC server CPU, DPU Pensando series, AI PC processor Ryzen AI PRO 300 series, etc. The release of these products demonstrates AMD's competitive ambitions in the AI computing and data center markets.
Despite the positive impact of the press conference, AMD's stock price performance was relatively flat. After the press conference, the stock price initially rose, but quickly fell back, closing at $164.18 per share, a decrease of 4%, and a slight increase of 0.26% after hours.
Why has the stock price fallen? On the one hand, due to the influence of the US stock market, all three major indexes closed slightly lower, and technology stocks also had mixed ups and downs; On the other hand, there is noise in the market regarding whether AMD can effectively compete with Nvidia, especially in terms of ecosystem and AI software integration.
Analysts have pointed out to 21st Century Business Herald reporters that despite AMD's outstanding hardware performance, Nvidia's advantage in AI software remains its core competitiveness. This complex market reaction indicates that while investors are optimistic about AMD's AI prospects, they remain cautious about the competitive landscape.
On the other hand, Nvidia founder and CEO Huang Renxun recently conducted a three-day non trading roadshow, continuing to convey growth confidence. Next, the market will continue to pay attention to whether AMD's new products can be adopted by major data centers and AI companies. The company's performance in the coming quarters, especially the third quarter financial report, will be the key to evaluating sustained growth and stock price momentum.
Update GPU, CPU, DPU
As the focus of this press conference, AMD's new version of Instinct MI325X AI chip has attracted much attention. MI325X adopts AMD CDNA 3 architecture, which has improved performance and functionality compared to its predecessor MI300X.
Meanwhile, the benchmark is Nvidia's previous generation H200. It uses 256GB of HBM3e memory, with a maximum memory bandwidth of 6TB/second, and MI325X exceeds H200 in both values.
AMD stated that the MI325X outperforms the Nvidia H200 by 20% -40% in inference performance when running large models such as Llama 3.1 and Mixtral. Compared with H200, MI325X can provide a 1.3-fold performance advantage in FP16 (16 bit floating-point number) and FP8 peak theoretical computing performance.
In addition to MI325X, AMD has also revealed information about the next generation MI350 series GPU accelerator. The MI350 series will adopt AMD's CDNA 4 architecture and use advanced 3nm process technology, equipped with up to 288GB of HBM3E high bandwidth memory, benchmarking against Nvidia's latest Blackwell series.
According to AMD's roadmap, the MI350 series will be launched next year. The 8-card MI355X can achieve a peak computing power of 2.3PF in FP16 data format, which is 1.8 times higher than the MI325X and on par with the computing power of the Nvidia B200. In FP6 and FP4 formats, its computing power can reach 9.2PF, which is nearly twice the computing power of B200 in FP6 format.
This data indicates that AMD intends to further enhance its position in the server GPU market, and data centers are a promising growth market in the industry. Previously, AMD Chairman and CEO Su Zifeng stated that AMD expects data center GPU revenue to exceed $4.5 billion this year. In addition to GPUs, AMD has launched the fifth generation EPYC "Turing" series of server CPUs. This CPU adopts TSMC's 3/4nm process and Zen 5 architecture, with a maximum configuration of 192 cores and 384 threads. AMD stated that the EPYC 9965 performs several times better than Intel's flagship server CPU Xeon 8592+in various aspects. Over the years, AMD's market share in the server CPU market has been steadily increasing, reaching 34% in the first half of this year.
In addition, on PC processors, the Ryzen AI PRO 300 series targets the AI PC market, with over 100 Ryzen AI PRO PCs expected to be launched by 2025; In terms of network solutions, AMD has also released AI network cards Pensando Pollara 400 and Pensando Salina 400 DPU. DPU, as a key component of data centers, is of great significance for improving data processing efficiency and security.
Overall, the new products presented throughout the press conference mainly revolve around the data center market, with server chips as the core. They not only include GPU, which is considered an AI chip, but also core products such as CPU, and involve key components such as DPU. At present, AMD and Nvidia are accelerating the launch of new products to increase the related revenue of data centers.
Continue to challenge Nvidia
At the same time, the competitive landscape between AMD and Nvidia has become even more intense. For a long time, Nvidia has been a leader in the data center and AI fields, with its GPU products accounting for about 90% of the AI computing market share. In recent years, competitors such as AMD have also accelerated their rise, providing more AI options.
It is reported that many generative AI solutions from companies such as Microsoft, OpenAI, Meta, Here, and Stability AI have adopted the MI300 series AI chip. With the new MI325X entering mass production in 2025, AMD is expected to further enhance its influence in the data center and AI markets.
Analysts generally believe that AMD is taking active measures to compete directly with Nvidia in the data center and AI fields, but the market position gap between AMD and Nvidia still exists, especially in terms of customer adoption speed and ecosystem maturity.
Several industry insiders told reporters that AMD has been struggling to catch up with Nvidia, especially since Nvidia has also developed and integrated CPUs, the competition has further intensified. Now it is not just a competition of single chips, but a competition of systematic solutions.
For example, NVIDIA has a well-established ecosystem in the GPU field, and over the past decade, it has honed its CUDA software ecosystem empire, making its products widely used in fields such as deep learning. However, AMD is also constantly improving its software ROCm so that AI developers can easily switch more AI models to AMD's chips. This has gradually given AMD the strength to compete with Nvidia in terms of technology and ecosystem.
While AMD is making frequent efforts, Nvidia has also been amplifying its voice recently. Previously, Blackwell's delayed delivery raised concerns in the market about Nvidia's new products. Recently, Huang Renxun's three-day speech swept away the fog and released strong growth information to the market.
According to the communication points outlined in the Morgan Stanley report, the deployment of Blackwell is progressing smoothly in the short term, and the products have been sold out in the next 12 months, demonstrating sustained strong market demand. The initial issues with Blackwell's silicon wafers have now been fully resolved, and it is expected that Blackwell will achieve billions of dollars in revenue in the January quarter.
At the same time, the report also pointed out that Nvidia's management emphasized the new opportunities brought by the complexity of AI inference, which is exponentially expanding with "long-term thinking". And more complex reasoning requirements require more computing power, which also means a richer hardware combination - Nvidia stated that the upcoming rack level product is the best solution.
Overall, 2025 will be an exceptionally strong year, defined by Nvidia management as the early stages of the long-term investment cycle. Huang Renxun also clearly stated that it is expected to continue to grow in 2026, although quantitative data has not been disclosed.
Looking ahead, 2025 will be a crucial year for intense competition between both parties. Nvidia's Blackwell architecture is about to enter the market on a large scale, and AMD will also focus on the new generation of chips, striving to occupy a larger share in the data center market. AMD is continuously challenging Nvidia, while Nvidia is also strengthening its long-term leadership position in the market. In the coming years, this competition between tech giants will not only be about hardware performance, but also about the maturity of overall solutions and long-term market demand.