Bottom line: The strategic rationale behind AMD acquiring ZT makes sense, to a point, but $5 billion is a lot to pay, especially when the true value of the transaction relies heavily on the price AMD will get when they spin off ZT's manufacturing unit.
Last week, AMD announced it would pay $4.9 billion for privately-held ZT Systems. This deal seems to have caught many off guard, in no small part due to the fact that ZT is not well-known. Put simply, we understand the strategic merits of the deal, but we question the price AMD paid.
ZT designs and builds data center systems. They exist on the border between an ODM (building others' designs) and an OEM (designing their own systems). In particular, they are best known for building and installing server racks for the hyperscalers. Of course, this means that they are suppliers to many of AMD's competitors. AMD has said they plan to spin off the manufacturing side of this business sometime next year, after the deal closes. They will then hold onto the design side of ZT.
Editor's Note:
Guest author Jonathan Goldberg is the founder of D2D Advisory, a multi-functional consulting firm. Jonathan has developed growth strategies and alliances for companies in the mobile, networking, gaming, and software industries.
The Street liked the deal, with AMD stock up ~5% on the news. The logic here seems to be that this deal makes AMD more competitive with Nvidia. Nvidia has long provided more than just discrete chips to its customers. Their business model is built around supplying complete systems, which once meant graphics cards and now means entire server racks and even complete data centers. ZT's design assets will let AMD compete more fully with Nvidia.
AMD has been riding high this year, or at least holding its own, on the back of its MI300 GPU system, which is positioned to compete directly with Nvidia's H100 systems. The trouble is that Nvidia delivers those complete systems, while AMD was essentially just delivering individual servers. Our understanding is that AMD's big MI300 win at Microsoft was hampered by months of delays stemming from problems getting the cluster up and running. In theory, ZT would have solved this problem prior to the MI300 launch by building and testing a working system.
All of this makes a fair amount of sense. AMD needs system design capabilities, and ZT provides them with those. But our sense is this logic only goes so far.
First, we are not entirely convinced as to just how much of the market is going to move in this direction. As much as Nvidia is rewriting the playbook for data center architectures, there are limits to how large the customer base is for such systems.
Dell and HP still sell a lot of catalog systems. The hyperscalers like to use ZT to assemble and install their servers, but they are increasingly designing those servers themselves. So while there is definitely going to be demand for ZT's design services, it is not clear just how big that demand will be. How much revenue will those services generate for AMD, and how much will just become a cost of doing business?
This, of course, leads to the question of price. We know very little about ZT's financials. Reuters' Max Cherney puts their revenue last year at $10 billion, "most of which is derived from their manufacturing unit." AMD also noted the deal will only become accretive by the end of 2025, which implies ZT's operating margins are fairly skinny.
Overall, this is a bit of a strange deal. AMD needed a sizable server design team, and there are not many of those out there. Building their own organically would take years, so buying ZT is a sensible option. That being said, $4.9 billion is a lot of money, especially when they do not want to keep the majority of the business. Ultimately, the deal's value hinges heavily on what they get for the manufacturing unit.