AMD to Buy ZT Systems to Accelerate System Delivery

Advanced Micro Devices (Nasdaq: AMD) today announced its intent to acquire AI infrastructure solutions designer ZT Systems, a company with which AMD already shares multiple customers, for $4.9 billion. It's a move that AMD says will enable it to deliver complex AI systems—including networking solutions—to hyperscaler cloud customers faster and with greater customization.

The deal, set to close in the first half of 2025, is aimed squarely at a couple of problems AMD sees as key to competing against archrival NVIDIA (Nasdaq: NVDA). First, it typically takes “several quarters” to deliver a full AI cluster from start to production. With help from 1,000 newly acquired ZT Systems’ design engineers, that timeframe will shrink significantly, management says. And accelerated deployment should help with AMD’s rollout of its Instinct MI350 and MI400 GPUs to the hyperscaler datacenter market sometime next year. Those components in particular are meant to advance AMD’s position relative to NVIDIA.

AMD also plans for ZT Systems, founded in 1994 and headquartered in Secaucus, New Jersey, to help customize solutions for hyperscalers and large enterprise customers. AI clusters are complex, and “There’s not a one size fits all,” said AMD CEO Lisa Su on a conference call with financial analysts this morning. Largescale systems customers are asking for bespoke solutions that fit specific requirements.

This puts AMD’s OEMs and ODMs in the spotlight. CEO Su stated in the press release: “Combining our high-performance Instinct AI accelerator, EPYC CPU, and networking product portfolios with ZT Systems’ industry-leading data center systems expertise will enable AMD to deliver end-to-end data center AI infrastructure at scale with our ecosystem of OEM and ODM partners.”

Analysts Applaud but Question the Numbers

Analysts on today’s conference call congratulated AMD management on the deal but questioned the level of accretion and margin improvement that could result from the acquisition. CEO Su was vague on specifics. She said the way to think about the situation was to consider that it rounds out a portfolio of acquisitions that are helping to boost AMD’s silicon and software solutions and deliver them as quickly as possible. Only last week, for example, AMD closed the $665 million acquisition of Silo AI, which creates language models for use with AMD GPUs. This is just one of several recent small acquisitions aimed at improving AMD's AI stack.

When pressed, Su said that AMD’s purchase of ZT won’t be accretive until the end of 2025 and the start of 2026, when the numbers should show increased sales volume due to an acceleration of customer deployments resulting from the ZT Systems buy. Margins, management noted, will be reasonable.

Interestingly, AMD said accretion from the buy won’t reflect any input from AMD’s planned sale of ZT Systems’ manufacturing business, the largest part of ZT, which accounted for over $10 billion in revenue over the past 12 months. AMD is apparently already peddling the division to prospective buyers.

More and Better Networking

There were several questions on today's call about whether ZT Systems will help AMD with AI networking, an area in which NVIDIA faces stiff competition from Arista Networks and other companies, including AMD, that are part of the Ultra Ethernet Consortium’s (UAC’s) efforts to displace NVIDIA’s InfiniBand with Ethernet.

CEO Su said that yes, ZT will help AMD’s networking profile, as will ongoing input from the star networking team acquired with Pensando in 2022. She also noted that in addition to work with the UAC, AMD will continue to spearhead the develop of a standard Ultra Accelerator Link (UALink) between GPUs, an effort announced earlier this year with partners Broadcom, Cisco, Google, HPE, Intel, Meta, and Microsoft.

The networking piece is key: NVIDIA itself has responded to Ethernet demand with its own streamlined Spectrum-X Ethernet networking technology, while its InfiniBand networking technology remains firmly entrenched across many AI clusters, and its full-stack approach has gained momentum.

Changing the Competitive Landscape

Now AMD is seeking to offer its own full stack. If AMD’s acquisition of ZT Systems works out as planned, it could put AMD in a better position against NVIDIA. That market leader also faces a growing roster of competitors, including Huawei in China, a large market in which NVIDIA’s sales efforts are hampered by U.S. regulations. A more fragmented market could present opportunities for AMD, a long-established player, to take more share of a market populated by many startups. AMD estimates the market will be worth $400 billion by 2027.

Still, whether AMD can realize sufficient momentum from this deal to justify its price remains to be seen. There could be another specter at the banquet: Some observers are questioning whether the hyperscalers, a large portion of AMD's target market for this deal, are spending too much on AI infrastructure. If that trend persists, AMD could lose out on its projections.

So far, though, even incremental improvements in datacenter system delivery could help AMD in its ability to meet customer requirements with speed and precision.

AMD will pay for ZT Systems 75% in cash, 25% in common stock. The purchase price includes a payment of $400 million contingent on certain milestones ZT Systems must reach after closing. This morning, AMD’s shares were trading at $152.84, +4.28 (up 2.88%).

Futuriom Take: AMD's purchase of ZT Systems has little apparent downside. At the very least, the addition of 1,000 AI infrastructure experts could help grow AMD's volume of hyperscaler and enterprise wins by boosting the speed of delivery and the customization of solutions. Still, whether the datacenter enterprise AI market will actually grow to $400 billion by 2027 remains to be seen.