A quarter of data center CPUs are AMD Epycs nowadays, gaining ground from Intel’s Xeon lineup. AMD’s one-two punch of CPUs and GPUs is beating Intel’s revenue from the data center market for the very first time. Nevertheless, it’s a fraction of Nvidia’s mammoth gains.
Note that this is total revenue brought in by AMD’s business unit focused on data centers, not the number of chips sold. We understand that Epycs typically bring in a higher price per SKU sold than Intel does, but the biggest growth is in the MI300X GPUs which are eclipsing even AMD’s own sales expectations.
Nvidia on another planet
That AMD has been doing well in recent years is obvious. Unlike ailing Intel, it is a real Nvidia alternative in the GPU field with comparable performance and a workable software stack. In addition, its own Epyc chips are an attractive rival to Intel’s Xeon range.
AMD’s data center side pulled in $3.5 billion last quarter, compared with $3.3 billion for Intel. However, no one can ignore superpower Nvidia, which collected $26.2 billion from data center compute alone in the most recent quarter. Over all of 2024, AMD CEO Lisa Su expects Instinct GPUs to pull in $5.5 billion. In essence, Nvidia’s on another planet here.
SemiAnalysis research shows even more clearly why Nvidia is in no way threatened. Nvidia’s networking revenue alone is competitive with the entire (!) data center offerings of AMD and Intel, a side business which, while fundamental to its scalability with InfiniBand, is not its headline feature.
Room for success
For AMD, the broader numbers are positive, but the annual growth of the data center segment steals the show at 122 percent. These are similar growth rates to Nvidia, which shows that with the AI chip boom, there’s room for multiple success stories. Perhaps even for Intel, should it start to make promises that do have a basis in reality and up its game with Gaudi GPUs.
Either way, it is unsurprising. AMD recently indicated at its own Advancing AI event that it has thought about its newfound position in the market for some time. For example, it is working with third parties on the Ultra Ethernet Consortium for clear infrastructure standards in the data center that competitors can also benefit from. Furthermore, AMD’s AI stack of GPUs is a lot less closed off than Nvidia’s CUDA, although adoption will ultimately be decisive.