This morning, AMD announced its acquisition of ZT Systems, a company known for constructing major hyper-scale AI systems for AMD. This move sheds light on various aspects of the AI industry and marks the beginning of a series in our Substack to delve deeper into this topic.
AMD Acquires ZT Systems in Unexpected Deal
The transaction is priced at $4.9 billion, with an additional $400 million payment pending later. Nevertheless, the actual value of the deal is anticipated to be lower as both companies have confirmed plans to offload the ZT Systems manufacturing division. Considering the scale of this manufacturing operation, it’s likely to attract interest from several large buyers.
Potential buyers could include major firms like MiTAC, QCT, HPE, Dell, Supermicro, Sanmina, Flex, or even private equity firms. Acquiring this business could particularly benefit HPE as it strives to enhance its position in the AI sector, competing with frontrunners like Supermicro and Dell. Following the acquisition, AMD will categorize ZT Systems’ earnings as discontinued operations and aims to divest the manufacturing sector swiftly due to the competitive dynamics with its client base, emphasizing the urgency of this divestiture.
A typical large AI cluster consists of several core systems. Top-tier system components include AI accelerators, GPUs, CPUs, RDMA NICs, and switch chips. On the other hand, lower-tier components involve liquid cooling systems for both data center and rack cooling, power supply solutions for data halls and buildings, rack designs, management networks, and physical security measures. It’s apparent that a small number of companies (around 2-3) dominate the production of high-end system components, while a larger group (often more than 10) handle the lower-tier components.
For AMD, possessing an 8-way GPU system is beneficial but only partially addresses challenges. Companies must also tackle the deployment of such systems as new resources for power and water become available, particularly in regions like the US and Europe.
AMD’s acquisition of ZT’s engineering capabilities propels this acceleration. When considering AMD’s potential $4-5 billion in AI server revenue this year, it’s important to recognize this as equivalent to only 1-2 GPU clusters. Supermicro, on the NVIDIA front, can now deliver a $4-5 billion cluster in under a month. For AMD to effectively compete with NVIDIA and justify its investments, it needs comparable technology, competitive pricing, and rapid availability. In the current market, where GPU clusters are expansive, delays of one year for a cluster to become operational are deemed unacceptable. A three-year timeline, like that for a HPE-Cray government Exascale supercomputer, is now considered exceedingly slow.
Final Words
Our assessment is that the total acquisition cost of ZT Systems by HPE will likely be significantly lower than the projected $4.9 billion after the disposal of the manufacturing division. This move would promptly position HPE as a noteworthy player in the AI domain, a field where it has previously been outpaced by competitors like Supermicro and Dell. Meanwhile, AMD is enhancing its capability to meet the AI Speed-to-Deploy (STD) demands of its clientele without competing with its allies. Failing to focus on the AI STD metric could jeopardize its future competitiveness. This scenario resonates profoundly for any AI startup operating outside the realms influenced by major players like NVIDIA, AMD, Cerebras, and possibly Intel.
ColoCrossing excels in providing enterprise Colocation Services, Dedicated Servers, VPS, and a variety of Managed Solutions, operating from 8 data center locations nationwide. We cater to the diverse needs of businesses of any size, offering tailored solutions for your unique requirements. With our unwavering commitment to reliability, security, and performance, we ensure a seamless hosting experience.
For Inquiries or to receive a personalized quote, please reach out to us through our contact form here or email us at sales@colocrossing.com.
