metamorworks - stock.adobe.com

AMD acquires ZT Systems for $5B to take on Nvidia in AI race

AMD plans to acquire AI systems designer and manufacturer ZT Systems for $5 billion. AMD CEO Lisa Su said hyperscalers want more than just chips when developing AI applications.

AMD aims to design entire systems, not just chips, to better compete with Nvidia in selling AI technology for the data centers of major cloud providers and large enterprises.

On Monday, AMD unveiled plans to spend $4.9 billion to acquire ZT Systems in the first half of next year. ZT is a designer and manufacturer of AI and cloud computing systems for hyperscale data centers.

AMD CEO Lisa Su told Wall Street analysts that interconnected server racks running tens of thousands of GPUs for model training and inferencing will grow even more complex in time. Therefore, customers will need a chip provider that can also help them design systems and get them into production faster.

Today, organizations take several quarters from the time of sampling GPUs and getting them into servers running production workloads, Su said.

"[The ZT] team will help accelerate AMD at large scale," Su said during the conference call with analysts. "We can actually do quite a bit of development in parallel."

ZT will help AMD's largest customers design their AI infrastructure. At the same time, the chipmaker will optimize its GPUs and CPUs for the systems, Su said. However, ZT will continue developing systems for organizations that want to use competitors' silicon.

"This is not about taking away customer choice," Su said. "Some hyperscalers are going to want different optimizations of those system designs, and we'll have the team to do that."

AMD is a much smaller provider of AI accelerators than Nvidia. Nvidia reported $22.6 billion in data center revenue for the quarter ended in April, with a significant portion comprising AI systems. AMD expects to have $4.5 billion in sales this year from AI data center GPUs.

ZT also designs and manufactures non-AI CPU-powered systems, so the acquisition could help AMD become a stronger competitor against Intel within the data centers of large organizations, said Jack Gold, an analyst at J.Gold Associates. AMD could use ZT to promote its EPYC CPU against Intel's Xeon chip.

Lisa Su, CEO, AMDLisa Su

"Since ZT also supplies non-AI solutions, this is a direct attack by AMD on Intel," Gold said on LinkedIn.

Analysts expect demand for AI GPUs to outpace CPUs in large data centers. AMD is launching AI accelerators as quickly as it can to increase its share of a market that Su said would rise from $45 billion last year to $400 billion in 2027.

In December, AMD released the MI300 Series, its first Instinct AI accelerator for hyperscale data centers. In 2026, the company plans to launch the MI400 series for large-scale AI training and inferencing. For programming GPUs running large language models, AMD offers its ROCm open software stack of tools, libraries and APIs.

AMD plans to sell ZT's hardware manufacturing unit after completing the deal, Su said. ZT's revenue was roughly $10 billion in the last 12 months, most of it from its manufacturing unit.

AMD expects to retain about a thousand of the privately held company's 2,500 employees at a cost of $150 million in operating expenses. The chipmaker expects ZT to contribute to AMD's gross revenues starting in 2026.

Following the acquisition, ZT will become a part of AMD's data center business group. ZT CEO Frank Zhang will lead AMD's manufacturing business, and ZT President Doug Huang will take over AMD's system design teams.

The ZT deal came a week after AMD closed the $665 million acquisition of Silo AI. The European lab develops AI services for autonomous vehicles, manufacturing and smart cities.

The ZT acquisition is one of AMD's largest. In 2022, AMD acquired Xilinx, a maker of programmable integrated circuits for customizing microprocessors, for $50 billion. That same year, AMD acquired Pensando for $1.9 billion. Pensando made programmable data processing units that run specialized tasks to reduce the workload on a server's CPU.

Antone Gonsalves is an editor at large for TechTarget Editorial, reporting on industry trends critical to enterprise tech buyers. He has worked in tech journalism for 25 years and is based in San Francisco. Have a news tip? Please drop him an email.

Next Steps

AMD layoffs follow AI job trend

Dig Deeper on Data center hardware and strategy