At the Morgan Stanley Technology, Media, and Telecommunications Conference, AMD CEO Lisa Su engaged in an energetic dialogue with Morgan Stanley analysts, addressing several hot topics in the AI infrastructure boom. Su specifically highlighted that the current market focus is on the collaboration between AI chip manufacturers and large tech companies, the diversification of AI infrastructure, and its impact on future computing architectures.
AI Infrastructure Tends Towards Heterogeneity, Cannot Rely on a "Universal Chip"
Discussing AI computing demands, Su stated that the types of computing needed for AI are becoming increasingly diverse, and the construction of AI infrastructure is entering a more complex stage. She emphasized that AI computing no longer relies on a single "universal chip," but rather has entered an era of heterogeneous computing. Su pointed out that the diversity of AI workloads means different application scenarios require different types of computing resources, from training to inference, and from large models to medium models.
"In the next phase of AI infrastructure, no single chip can do everything best; it is already a heterogeneous world," Su said. The diversity of computing needs requires more flexible architectures and chip solutions. She believes that ASICs (Application-Specific Integrated Circuits) will have a place in certain computing tasks, but AMD will strive to balance between flexibility and customization for specific workloads.
Su also specifically mentioned that AMD's collaboration with OpenAI is an excellent example showing the potential for investment in AI infrastructure. Last year, AMD reached a large-scale computing power supply agreement with OpenAI, which also purchased about 10% of AMD's shares at a reduced price. Recently, AMD also reached a similar cooperation agreement with Meta, with a transaction scale of 6 gigawatts, potentially valued at hundreds of billions of dollars.
Collaboration with Meta: Stock Warrants Provide Incentive for Deepening Partnership
Discussing the cooperation with Meta, Su revealed that AMD has further strengthened the partnership by issuing performance-based stock warrants to Meta. She emphasized that the value of the stock warrants lies in their ability to accelerate purchases in the transactions and support AMD's ecosystem development. "We see broad opportunities and are charting a path for the future of AI infrastructure," she said. The parties will work together to achieve performance goals, and this collaborative model could be replicated with other clients in the future, but for OpenAI and Meta, the cooperation is unique and deep.
Impact of CPU Supply Constraints and Memory Price Increases on the Market
Aside from the diversity of AI infrastructure, Su also discussed two major issues impacting the supply chain recently: CPU supply constraints and rising memory prices. She stated that the surge in AI computing demand has led to greater than anticipated demand for CPUs, especially among large tech companies, many of whom underestimated their AI computing needs, with the market size far exceeding expectations from three to six months ago.
Su revealed that AMD is in a favorable position in collaboration with the supply chain to meet most market demands and plans to further expand supply capacity in 2026 and 2027. However, she also noted that the increase in DDR4 and DDR5 memory prices is affecting consumer product pricing, and this trend is expected to continue to exert significant cost pressure on the PC market in the second half of this year.
Fierce Competition in the Chinese Market, AMD Continues to Focus on Global Chip Innovation
Speaking about the Chinese market, Su noted that despite fierce competition, AMD remains optimistic about the market and acknowledged the achievements of Chinese chip suppliers. She stated that AMD will continue to focus on chip innovation worldwide and strive to provide customers with the best options through a diverse product lineup and flexible computing solutions.
