For the first time ever, AMD outsells Intel in the datacenter space

News
By
published

But both fall far behind sales of Nvidia's AI GPUs.

AMD EPYC CPU
(Image credit: AMD)

For well more than two decades, Intel has been the undisputed leader in the market for datacenter CPUs. Intel's Xeon processors powered the vast majority of servers, whereas AMD's processors commanded a single-digit market share just some seven or eight years ago. However, the situation has changed drastically. While Intel's Xeon CPUs still power the majority of servers, the most expensive machines now use AMD's EPYC processors. This is why AMD's datacenter business unit now outsells Intel's datacenter and AI business group, as observed by SemiAnalysis.

Indeed, AMD's datacenter segment revenue reached $3.549 billion in the third quarter, whereas Intel's datacenter and AI group's earnings were $3.3 billion in Q3 2024. Just two years ago, Intel's DCAI group earned $5 billion - $6 billion per quarter. But as AMD's EPYC processors have gained competitive advantages over Intel's Xeon CPUs, Intel has had to sell its server chips at significant discounts, which has reduced the company's revenue and profit margins.

It is noteworthy that Intel's flagship 128-core Xeon 6980P 'Granite Rapids' processor costs $17,800, making it the company's most expensive standard CPU ever. By contrast, AMD's most expensive 96-core EPYC 6979P processor costs $11,805. If demand for Intel's Xeon 6900-series processors remains high and the company can supply these CPUs in decent volumes, then Intel's datacenter revenue will likely get back on track and surpass AMD's datacenter sales. However, Intel still has to ramp up production of its Granite Rapids products. 

While both Intel and AMD now earn around $3-3.5 billion per quarter selling datacenter CPUs, Nvidia earns much more from its datacenter GPUs and networking chips, which are required to make AI processors work in concert in datacenters. In fact, sales of Nvidia's networking products totaled $3.668 billion in the company's second quarter of fiscal 2025. Meanwhile, compute GPU sales reached $22.604 billion in Q2 FY2025, which far surpasses the combined sales of Intel and AMD datacenter hardware. Altogether, Nvidia sold nearly $42 billion worth of AI and HPC GPUs in the first half of this year, and it is likely that the company will sell even more datacenter processors in the second half.

See more CPUs News
Anton Shilov
Anton Shilov
Contributing Writer

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

22 Comments Comment from the forums
  • prtskg
    Since data is for both datacentre gpu and cpus, it'll still take some time for AMD to reach 50% share in datacentre cpus.
    Reply
  • P1nky
    At this rate Intel, as we know it, will cease to exist in 2-3 years.
    Reply
  • jeremyj_83
    "By contrast, AMD's most expensive 96-core EPYC 6979P processor costs $11,805."

    The 6979P is an Intel part. For the Epyc 9654 (Zen 4) cost that much on launch in 2022/23, however, that isn't the top SKU anymore with Zen 5 Epycs now out.
    Reply
  • ottonis
    I am glad that all the great work at AMD is now paying for itself and generating as much or more revenue in the datacenter market as Intel does with its history of morally doubtful business practices.
    Of course, as the article mentions, not all is roses, though, as nVidia is clearly showing where the really big revenues are coming from: the (AI)-compute GPU market. To my knowledge, AMD does have decent AI-GPU accelerator cards that may or may not (I think the latter is true) outperform nVidias latest and greatest but that still will find many pockets willing to pay for them as nVidia is either too expensive or just cannot entirely deliver the huge numbers of cards the AI-market demands.
    Now, AMD only needs to actively and massively support developers with impeccable documentation, technical support, and easy to use developer tools for their AI-compute GPUs. Then, and only then, it may sorten the gap to nVidia.
    Reply
  • Nikolay Mihaylov
    jeremyj_83 said:
    "By contrast, AMD's most expensive 96-core EPYC 6979P processor costs $11,805."

    The 6979P is an Intel part. For the Epyc 9654 (Zen 4) cost that much on launch in 2022/23, however, that isn't the top SKU anymore with Zen 5 Epycs now out.
    So, you are saying, things are actually even worse for intel than the article sugegsts? ;)
    Reply
  • jeremyj_83
    Nikolay Mihaylov said:
    So, you are saying, things are actually even worse for intel than the article sugegsts? ;)
    Yeah. Servethehome did a review of the newest Zen 5 CPUs when they were released. The 128c Zen 5 was overall faster than the 128c Xeon. The 192c Zen 5c was even better and they found out that PCIe 4 SSD and 100GbE wasn't fast enough to keep these (192c) CPUs fed with data. When they went with PCIe 5 SSD and dual port 200GbE, performance increased by up to 15% (IIRC).
    Reply
  • Pierce2623
    prtskg said:
    Since data is for both datacentre gpu and cpus, it'll still take some time for AMD to reach 50% share in datacentre cpus.
    Pretty sure instinct is still in Radeon and Ponte Vecchio is still in XCG. Neither one is counted in their respective Data Center group because companies always try to keep strongly performing and poor performing units separate. That way they can say “look how great Data Center is doing and our only poor performance is in graphics…blah blah blah”.
    Reply
  • YSCCC
    If true, intel is having serious trouble of not only losing market in the DIY market where fans always say it's unimportant, if the brand image get damaged so badly it will be difficult to gain back momentum, just like Nokia in the phone world
    Reply
  • FrostByteVA
    While Intel's Xeon CPUs still power the majority of servers, the most expensive machines now use AMD's EPYC processors.

    The title of the article is AMD Outsells Intel in the datacenter space. The article says that "Intel still powers the majority of servers" but that "the most expensive machines now use AMD". So what specific space is AMD outselling Intel? GPU enabled servers but not standard servers? Don't get me wrong, we have AMD Gen3 and Gen4 and are looking at 5 for the next order. However, the article does not state in what space AMD is outselling Intel.
    Reply
  • TTech002
    I don't understand how you could write an article about servers, and not mention the,
    Epic 5 9755 128 Cores, and the Epic 5c 9965 192 Cores.
    Reply