The next – and more profitable – 10% server share for AMD


When all of this is said and done, Intel will deserve some sort of reward for keeping its 14 nanometer processes sufficiently advanced as it gets its 10 nanometer and 7 nanometer processes together to succeed, one way or another. , to retain a dominant market share in the server. space.

Or, maybe Intel just got lucky that AMD’s supply couldn’t even meet the potential demand it might otherwise have if there were no capacity limits. Taiwan Semiconductor Manufacturing Co, which etches the base complexes into its Epyc server chips. (Globalfoundries, AMD’s own foundry spin-out mixed with IBM Microelectronics and Chartered Semiconductor, still manufactures the memory and I/O hub parts of AMD’s “Rome” Epyc 7002 and “Milan” Epyc 7003 processors .)

This time last year, when we looked at AMD’s increase in share of processor shipments and revenue since re-entering the server space in 2017, the company had just broken the 10 barrier. % shipping and seemed to be on a fast path to 20% or maybe 25% Opteron class. And then the 10 nanometer “Ice Lake” SP Xeons were launched, and say what you will about how the Rome chips beat them and the Milan chips hammered them, you go to the data center with the server chip you have, paraphrasing former secretary of Declare Donald Rumsfeld, who when under fire for the sorry state of the US military at the start of the gulf war joked with some spite saying that “you go to war with the army you have”.

If you look at the data from Mercury Research, which does a fabulous job looking at the competition between AMD and Intel, you’ll see that AMD has made a big jump in share of server processor shipments and then it levels off a bit or even decreases some and then once it has gained a foothold, it explodes a few points to a new level and repeats that mesmerizing jagged shape over and over again.

So at the end of the fourth quarter, according to Mercury Research, AMD shipped 1.13 million server processors, an 82.9% increase over the previous year, which is tremendous. But the overall market for server processors – pushed into the channel, not consumed by customers on the other side of the channel – grew 29.9% to 8.85 million units. And so, Intel was able to grow almost at the same rate as the market as a whole, growing 24.6% to 7.71 million units. And so, Intel was able to have a record quarter for server processor shipments, which also propelled server revenue to new highs despite a 5% decline in sales to hyperscalers and cloud builders and thanks in large part of a return to spending by telecommunications and communications companies and service providers. We suspect there’s been a little channel stuffing from Intel, and a lot of component bundling (i.e. a price cut that doesn’t affect the Data Center revenue line Group, but reduce data center revenue and profit Core group adjacencies such as ASIC switches, network interface cards, silicon photonics (primarily Ethernet optical transceivers), and Optane storage. Pricing will all help – unless server demand continues to grow and Intel does a good job with the upcoming 10-nanometer “Sapphire Rapids” SP Xeon expected in Q2.

To stay with the war metaphor, Intel’s 14nm infantry with a few 10nm tanks were able to hold the line against AMD’s tactical assault teams and snipers. And there are about 10 nanoscale combat ships on the way as AMD brings a new class of weaponry to wear with the Epyc 7004 “Genoa”. If AMD had more capacity, it would eat more of a share. there is no doubt. But we think Intel and AMD are happy to run capacity close enough to demand that they can still have shortages that cause prices to hold or even rise, especially with higher-end SKUs in their processor lines. And they’ll stretch product launches to breaking point, and still be able to let hyperscalers and cloud builders dispose of those parts on the sly and charge a premium for that as well.

In other words, server buyers, none of you are in control. TSMC and Intel Foundry Services are, and they set the tone on pricing and set the pace for shipments, and if you need a server, you go to the data center with whatever processors you have.

Anyway, back to the stats. When you do the math, AMD had a 12.9% shipment share in Q4 2021, according to Mercury Research, and that’s just three-tenths of a point higher than the share it had in Q3 and mimicking the same tepid growth in share it had in the previous year at the same time.

It’s hard to say what will happen when the Genoa Epyc 7004s hit the market and perhaps more revenue is recognized for the “Trento” custom Epyc processors used in the Oak Ridge National Laboratory “Frontier” supercomputer. (we believe much of this was done in the second, third and fourth quarters of 2021, but AMD hasn’t elaborated on how it works).

What we will observe is that AMD’s revenue share has exceeded its shipment share since Q2 2021, when this revenue recognition for Frontier might have started. And given that major supercomputing centers typically pay less than half the list price for CPUs and GPUs, based on the estimates we’ve made, this revenue recognition would have actually hurt the revenue share of AMD. And so, if AMD’s share of X86 server chip market revenue is higher, it means that Intel’s average revenue per chip tends to grow slowly compared to AMD’s. (Both tend to increase as more and more customers purchase the stack.) We will be watching this rate of revenue expansion carefully. In the fourth quarter, for example, AMD captured 14.4% of $7.48 billion in X86 server processor sales, and it held the same share of $6.65 billion in X86 server processor sales in third quarter 2021, according to Mercury Research.

It is even harder to say what will happen towards the end of 2022, until 2023 and until 2024. At that time, the server market could consume almost 10 million X86 server processors per quarter , and maybe even 500,000 Arm or more. server processors (why not?), with X86 servers generating somewhere near $9 billion in revenue per quarter. We still have a long way to go to get to that point, but when it does, if current trends continue, it’s not hard to see AMD having somewhere north of 20% shipment share and close 25% revenue share of the X86 market. , which will continue to grow albeit at a much slower rate than it currently does.

It will be fun to watch unfold, quarter by quarter, blow by blow.


Comments are closed.