AMD's Lisa Su: AI production capacity has received sufficient assurance

Amanda Liang, Taipei; Jack Wu, DIGITIMES Asia 0

AMD CEO Lisa Su. Credit: AFP

AMD CEO Lisa Su's Asia tour in July to secure its supply chain became the focal point for analysts during the company's 2Q23 financial earnings call. In particular, discussions on the production capacity of AI chips and customer meetings allowed Lisa Su to talk freely about AI.

Regarding the analysts' questions about AI production capacity, Su admitted that the supply chain capacity is indeed very tight. She also confirmed that discussions with supply chain partners during the Asia tour included talks about working together to significantly expand production capacity.

Another question raised by analysts was whether AMD would encounter supply chain bottlenecks in 2024. Su emphasized that while the overall supply chain capacity is undoubtedly tight, AMD has already promised to take in the massive capacity of the related supply chains, including TSMC's CoWoS advanced packaging and high-bandwidth memory (HBM).

She stated that data centers, especially the generative AI application market, hold strategic significance for AMD. AMD has ensured sufficient production capacity supply from front-end wafer manufacturing to back-end packaging to the specific components required to execute MI300-level solutions. All this allows the company to massively expand its production capacity between 4Q23 to 2024 to meet customer demands.

Regarding the current visibility of AMD's generative AI solutions for data centers, Su stated that customers are highly interested in MI300X, with many hoping to deploy it as soon as possible. AMD is closely collaborating with customers to conduct necessary joint designs to promote relevant deployments.

Su highlighted that AMD plans to begin early deployments in the first half of 2024. As capacity ramps up and products continue to ship, the company anticipates a higher volume of MI300 deployments in the second half of 2024.

In the upcoming quarters, AMD will continue to promote the MI300 solution to several data center customers. Su emphasized that 2024 will be a very important year for AMD.

Additionally, analysts asked about AMD's specific advantages in the high-end AI GPU market that's dominated by Nvidia and contains plenty of other AISC chip suppliers. This includes traditional rivals like Intel and numerous startups, all aiming to grab a share of this new market.

In response, Su believed that multiple winners will coexist in this market. AMD's specific niche is that the MI300 is a product series designed to be highly flexible. It covers the training and inference workloads for segments such as supercomputing, AI, and large language models (LLMs). AMD's products can encompass all these segmented markets, giving them the edge to succeed in a multi-winner market.