AMD’s Planned Acquisition Of ZT Systems Is Strategic And Sensible

Advanced Micro Devices (AMD) announced on Monday that it reached a definitive agreement to acquire ZT Systems. ZT System is a company that not many have heard of before this deal was announced but the company is a premier server manufacturer and currently serves Microsoft (MSFT) Azure, Amazon (AMZN) AWS, along with other major customers. The disclosed $10B annual revenue run rate is a testament to the company’s customer traction.

Despite sizable revenues, AMD is not buying ZT Systems for its revenues. AMD plans to keep the systems design arm of about 1000 engineers and plans to sell off the manufacturing side of ZT with about 1500 employees and which generates substantially all of ZT’s revenues. The core competencies that AMD gets from such an acquisition can be seen from the bottom half of the image below.

The acquired group of engineers will develop optimized solutions using AMD CPUs, GPUs. GPUs, FPGAs, and other components. Given ZT’s background, these engineers can enable world-class AI infrastructure which AMD can deliver to an ecosystem of hyperscalers, OEM and ODM partners.

Investors may wonder why AMD needs this now given it has already been successful with launching MI300 to considerable success. The reason is that without systems design competency, AMD is at the mercy of ODMs in Taiwan to enable the systems. WIth the market leader Nvidia (NVDA) exerting considerable power on the supply chain, AMD may not get the attention it needs from its partners. With this acquisition, AMD is taking its destiny more into its own hands. With AMD enabling designs, ODM’s will also be incentivized to move fast given that moving slow means that other ODMs can use AMD’s base design and enable systems. Given the explosive nature of AI demand, ODMs would be unwise to delay AMD MI family enablement.

Post acquisition, AMD would have an AI design footprint similar to Nvidia's except that AMD would not have a captive manufacturing operation (image below). This full portfolio approach cuts down the validation time for AMD designs meaningfully as customers will have a hyperscale caliber validated reference designs right off the bat.

It would not be easy for AMD to build or buy this kind of talent, especially in the Americas. Building a team new, in particular, is a non-starter because it would take a long time and AMD needs these competencies immediately to better compete against Nvidia.

The reason for divesting the manufacturing arm is that ZT, in a sense, is a direct competitor to Super MIcro (SMCI), Dell (DELL), and other companies which sell servers - especially to large ISPs and hyperscalers. AMD, unlike Nvidia does not want to compete with its customers by being in the system business. Some analysts see Nvidia’s system business as its strength, but Beyond The Hype sees it as an opportunistic play and a long term negative. Competing with customers rarely ends up well in the long term.

AMD is making the right choice here by selling off the systems manufacturing business. Furthermore, given the heated AI hardware space and geopolitical tensions, ZT’s North American and European manufacturing assets are likely highly desirable. It would not be surprising if many players including SMCI, HP (HPE), and Chinese/Taiwanese server manufacturers bid up the assets. ZT manufacturing arm sale could fetch a nice return to AMD. There is an excellent chance that such a sale will be announced almost simultaneously with the close of the AMD/ZT deal.

Given the above dynamic, it is unknown what AMD is paying for the design arm of ZT. But, even at the announced full price, AMD is paying less than $4.5M per person for a talented design team in a super-hot market where the opportunity costs are off-the-charts.

The cost of this acquisition, net of whatever AMD needs to fork-out after selling off the manufacturing business, is primarily the increased opex for the approximately 1000 ZT engineers. AMD estimates annual opex to be $150M at the time of deal closing in mid-2025.

Most importantly, what seems to be lost on much of the investor and analyst community is how quickly this deal will pay off for AMD with some analysts estimating that a mid-2025 close implies that new solutions will not be enabled until 2027. This is a complete mis-reading of the situation. Note that ZT is already supplying to Microsoft and it is unlikely that this deal occurred without Microsoft feedback at some level - especially given that Microsoft is AMD’s top MI300 customer.