The good news is that this surge in business and government spending happened just as hyperscalers and cloud builders — what Intel calls cloud service providers — were down 5%. They had already done much of their spending on SP Ice Lake Xeons earlier in 2021. By the way, Q4 2021 spending by hyperscalers and cloud builders for Data Center Group was still 19% lower, 3% to Q4 2019 spend. , and systems to communications service providers – that is, telcos and other smaller cloud builders and hosts – grew 22% in the fourth quarter of 2021, which also helped. For many consecutive quarters, companies have been relatively lukewarm on spending compared to their hyperscaler and cloud builder counterparts. But that wasn’t the case in the fourth quarter of 2021, when Intel’s data center group saw processor, chipset, motherboard and system revenue grow 53% year-over-year. To be fair, this was a reasonably easy comparison with a 25% drop in the fourth quarter of 2020, which was a very bad quarter for Data Center Group before the launch of the Xeon SP “Ice Lake” server in March 2021. What this really means is that business and government revenue for Data Center Group is only 14.8% higher in Q4 2021 than it was in Q4 2019, before the coronavirus pandemic hit. hit.
We suspect that some of the rise in business and government spending that Intel is seeing is tied to inventory hoarding through the IT channel, as supply chain shortages, for everything from cards to network interface to special power circuits on motherboards, were delayed in server shipments. They may be trying to get a head start, which is necessary when hyperscalers and cloud builders are ticking, enterprise and government customers are the tac, and communication service providers are somewhere in between. both.
And we know what you might be thinking: does this increase in corporate and government revenue have anything to do with the Intel-built “Aurora” supercomputer housed at Argonne National Laboratory? As far as we know that’s not the case, and that’s because Intel can only recognize revenue when a supercomputer is accepted and, again, as far as we know, the Aurora machine is only not yet built, let alone accepted. In fact, as we’ve previously reported, Intel Federal, which is the part of the chipmaker that’s the prime contractor for the Aurora machine, was to take a $300 million write-off in the quarter, which hurt Data Center Group’s earnings, and we believe this delisting is related to Aurora, but the company has been very stubborn in not directly answering the question. (The machine has been delayed several times, and we suspect there are penalties for this.)
During the quarter, the Data Center Group’s server chips, chipsets and motherboards accounted for $6.44 billion in sales, up 21.6% year-on-year, and adjacent businesses such as sales systems and networking accounted for $864 million, up 9.2%. Add them up and Data Center Group reported overall revenue of $7.31 billion, up 20%, but operating profit fell 16.9% to $1.73 billion. This impact on operating income includes the cost of increasing 10 nanometer manufacturing and its follow-up 10 nanometer and above process, called Intel 7, and its 7 nanometer processes, called Intel 4 (5 nanometers), and its developing a roadmap for the future. Intel 3, Intel 20A and Intel 18A process.
We’re still getting used to the new naming conventions, so here they are again as a reminder:
In a conference call with Wall Street analysts, CEO Pat Gelsinger said Intel shipped more Xeon chips in the month of December than the total number of server shipments “by a single competitor for the entire year 2021”. Gelsinger talks about you, AMD, and he added that in the fourth quarter of 2021, Intel shipped over one million units of the SP Ice Lake Xeons, which was equal to the shipments of the previous three quarters of 2021 (including Q1, when only hyperscalers and cloud builders were getting them before the official launch). That’s over 2 million SP Ice Lake Xeon, and that probably equates to just over a million servers. (SP Ice Lake Xeon are only available on single-socket or dual-socket machines, and some single-socket nodes are sold.)
Gelsinger reiterated that the company will begin shipping its first “Sapphire Rapids” Xeon SPs, based on the updated 10-nanometer SuperFIN process (now called Intel 7), in Q1 – again, most likely for hyperscalers and cloud builders ahead of the official Q2 launch of these chips. There were rumors that the Sapphire Rapids ramp had slipped in the third quarter, but Intel told us last week that the Sapphire Rapids chip was on track for a ramp starting in the second quarter.
“This unprecedented demand continues to be tempered by supply chain constraints as shortages of substrates, components and foundry silicon have limited our customers’ ability to ship finished systems,” Gelsinger explained. . “Industry-wide, this has been felt hardest in the consumer market, particularly in laptops, but the constraints have largely impacted other markets, including automotive, internet objects and the data center. As we predicted, these ecosystem stresses are expected to persist through 2022 and 2023, with gradual improvements over that time. The industry will continue to face challenges in various areas, including shortages of specialty and global foundries, substrates as well as third-party silicon.
We think at least part of Intel’s gain with the Ice Lake Xeon SPs is because AMD and Ampere Computing, which use Taiwan Semiconductor Manufacturing Co as a foundry, can’t get enough of their Epyc and Altra processors. respective to respond to the request. But that’s just a hunch. What we do know is that demand and supply are all sorts of struggles, and Gelsinger said as much as he envisions Intel’s next two years in data center computing.
We think there are constraints, and we also think chipmakers like when demand exceeds supply because they can charge a premium.
As for the competition between AMD and Intel, Gelsinger had this to say: “We expect there will be a bit of a back-and-forth with the competing alternatives, where they deliver a product and then we deliver the product. following. And as you’ve heard me say, we’re on the path to continued and undisputed leadership in this area. It will take us a few generations before we are indisputably in a leading position. But we believe that our product teams, our packaging teams and our process technologies, our factory capacity, all of this gives us the tools to create leading products. And then we combine that with our platform leadership and our software technologies, we have a path to undisputed leadership over time.
We will see. We’ve mapped out a Xeon SP roadmap to 2025 or 2026 that shows it can use architecture, process, and packaging to create very powerful server chips.
Data Center Group does not encapsulate all data center revenue generated by Intel, and we are trying to determine what the revenue and profit streams are across Intel. Here is the latest model including Q4 2021:
And for those of you who like raw data, here it is for the last two years:
Including sales of server chips, chipsets, motherboards and systems, as well as network switches and interface cards, server-based IoT edge systems, flash and Optane memory and FPGAs , we estimate Intel’s “actual” data center business was $9.12 billion, up 16%, and operating profit was $2.21, down by 3%.
- Companies fill hyperscaler gaps for Intel’s data center business
- View all news and articles from business news updates.