amid the rapid development in the field of artificial intelligence, NVIDIA was well aware of the intensified competition. The introduction of the H200 this year aimed to optimize inference costs. Although the H200 inherited the core architecture of the H100, it features a more powerful 141GB memory (compared to the H100's 80GB) and improved memory bandwidth, further enhancing inference performance.
However, as AI applications continue to expand with a surge in demand for computing power, the supply of the H200 is struggling to meet the urgent needs of the market. In this context, the MI325X showcases its competitive advantage. The MI325X not only boasts up to 256GB of memory, but it is also more affordable and has a more abundant supply than the H200. Therefore, large corporations like Microsoft and OpenAI are opting to support AMD, not just to reduce reliance on NVIDIA, but also to ensure an early and ample supply of the MI325X to advance the proliferation of AI services.
Prior to the large-scale shipment of NVIDIA's B200, the MI325X is undoubtedly the preferred solution for data centers. How the industry will respond after the release of the B200 remains to be seen. Considering the current state of technological development, AMD is likely to struggle to produce a GPU that can match the performance of the B200 before 2025. According to analyses by foreign media, it is anticipated that by 2026, AMD may develop products nearing the performance of the B200. By then, NVIDIA's B200 or even more advanced GB200 (a single GPU comprised of two B200s) might already be widely available in the market, placing AMD in a predicament similar to this spring's.
Despite this, AMD still has a chance to secure a place in this competition. Even if it cannot unseat NVIDIA's leadership, it is likely to capture a significant market share. AI has deeply integrated into our daily lives, and whether we admit it or not, many sectors have embedded AI technology. As artificial intelligence continues to proliferate, AI will more directly participate in the operation of the livelihood economy and require massive computational resources, exceeding the capability of existing data centers.
Thus, AMD still has time to catch up, while NVIDIA needs to pay more attention to this formidable competitor. After all, considering the lessons learned from the market battle between Intel and AMD, if NVIDIA shows even the slightest complacency, it may be surpassed by AMD at some point.