NVIDIA: Unrivaled Leader

 

To catch up with the AI chip leader NVIDIA (NVDA.US), AMD (AMD.US) under the leadership of "Mother Su" is unambiguous.

On August 12, 2024, AMD announced the acquisition of Silo AI, the largest private AI lab in Europe, for $665 million in cash. That's not all, AMD is full steam ahead, and just eight days later, it announced the acquisition of ZT Systems, a hyperscale solutions provider, for a total of $4.9 billion in cash and stock.

AMD expects the transaction to be completed in the first half of 2025, but it is still subject to regulatory approval and may be consolidated before the end of 2025.

 

What is the purpose of acquiring ZT?

ZT Systems, headquartered in New Jersey, has over 15 years of professional experience, focusing on designing and deploying AI computing and storage infrastructure for large-scale data centers for the world's leading cloud service companies. The company has shown its unique competitiveness in the field of AI training and inference infrastructure solutions, with its excellent capabilities in design, integration, manufacturing, and deployment.

ZT Systems provides customized solutions for the world's largest hyperscale cloud service providers, and its expertise in the field of artificial intelligence systems can complement AMD's chip and software technologies.

Advertisement

AMD believes that ZT brings professional benefits in design processes, mechanics, liquid cooling, networking, product validation, software design, AC/DC power, and scale testing.

AI requires a large amount of computing power, prompting technology companies to connect thousands of chips together to achieve the necessary data processing capabilities. Connecting a large number of chips together means that the entire server system structure needs to be optimized, and ZT is good at system design and rack-level solutions, which helps to enhance AMD's data center AI systems and customer support capabilities, which is the reason for AMD's acquisition of ZT.

Mother Su revealed to the media that ZT's engineers can enable AMD to test and launch its latest GPU chips more quickly to meet the needs of cloud computing giants such as Microsoft (MSFT.US).

She pointed out: "Combining our high-performance Instinct AI accelerators, EPYC CPUs, and networking product portfolio with ZT Systems' industry-leading data center system expertise will enable AMD to provide end-to-end data center AI infrastructure on a large scale with our OEM and ODM partner ecosystem."

After the transaction is completed, ZT will join AMD's data center solutions business unit, and ZT's CEO and founder Frank Zhang will lead the manufacturing business that is about to be sold, while ZT's President Doug Huang will lead the design and customer empowerment team, both reporting to AMD's Executive Vice President and General Manager Forrest Norrod.

AMD plans to spin off and sell ZT's server manufacturing business and does not intend to compete with companies such as Super Micro Computer (SMCI.US). For ZT's manufacturing business, it will be held for sale and classified as a discontinued operation, expected to take 12-18 months to sell, but according to media reports, AMD has not yet negotiated with potential buyers.

 

The impact on AMD's performance

AMD expects the data center AI accelerator market capacity to expand from $45 billion in 2023 to $400 billion in 2027, with a compound annual growth rate of over 70%.

AMD expects ZT to bring about 1,000 engineers, for which AMD's annual operating expenses will increase by about $150 million.

According to CNBC, Mother Su is expected to bring in about $10 billion in revenue per year from ZT.

Caihuashe estimates that this is equivalent to 42.96% of AMD's revenue of $23.276 billion for the 12 months ending June 29, 2024, of which AMD's original data center revenue for the 12 months was $9.051 billion, which may mean that the merger with ZT will double its data center revenue scale.

In comparison, NVIDIA's revenue from data center business for the 12 months ending April 28, 2024, reached $65.804 billion. See the red part in the figure below, after the delivery of AI chips such as H100, its data center business revenue continues to rise sharply. For AMD to catch up, mergers and acquisitions may be the fastest way.

 

NVIDIA's competitive advantage

However, NVIDIA's AI chips have been delivered in bulk and monetized, and it has accelerated the launch of newer AI products and services, continuously building its own AI moat.

For example, in March of this year, NVIDIA released the "Blackwell" model, which may be delivered later this year or in 2025, but before delivery, the company has released a new generation of artificial intelligence platform "Rubin" to replace the Blackwell platform.

According to technology media wccftech, the first generation of Blackwell GPU (B100/B200) will enter the data center later this year, and NVIDIA also plans to release a boosted version.

The new generation of Rubin R100 GPU will form the R series of products, expected to be mass-produced in the fourth quarter of 2025, and systems such as DGX and HGX solutions are expected to be mass-produced in the first half of 2026.

According to NVIDIA, the Rubin GPU and corresponding platforms will be launched in the first half of 2026, and the Ultra version will be launched in 2027.

The launch of this series of new products and the iteration and upgrade of products undoubtedly demonstrate NVIDIA's accelerated development in the field of artificial intelligence.

Huang Renxun announced plans to launch new Spectrum-X products every year to meet the growing demand for high-performance Ethernet network platforms for AI. In addition, NVIDIA also launched the AI model inference microservice NVIDIA NIM, providing models in the form of optimized containers, deployed in the cloud, data centers, or workstations.

It can be seen that NVIDIA still occupies an absolute leading position in the innovation of AI chips and solutions, and AMD needs to step up its efforts to catch up.