It is the nature of big tech companies with near monopolies to start looking a bit like Rome in its Golden Age – the Pax Romana that held from when Augustus Caesar became emperor in 27 BC until Marcus Aurelius died in 180 AD.
During these peaceful times, all things seem possible and all manner of things are funded from a seemingly bottomless fountain of molten gilt that flows like the blood in the opening of House of the Dragon. But eventually, as always, the barbarians smell the party and they come hoarding to the gates – and then everything gets incredibly difficult.
It happened to IBM in the late 1980s and early 1990s, it happened to Microsoft in the late 1990s and early 2000s, it happened to Sun Microsystems in the early to late 2000s, and it started happening Intel three years ago and it intensifying before our very eyes in the third quarter. It is quite possibly happening with the Facebook collective known as Meta Platforms. Twitter has met its Barbarian In Chief and his big checkbook. Mark our words: One day it will be the turn of Google and Amazon, too. It only takes one upstart with a better algorithm or a different idea about how to locate information or how to shop and warehouse products to AltaVista Google and Pets.com Amazon. Apple has risen from the ashes at least twice.
Everything always happens faster in the future. . . . including, and maybe especially, decline. And Intel has many rivals, all coming for it on all fronts.
Intel co-founder and long-time chief executive officer Andy Grove was right: “Only the paranoid survive.” But being paranoid is not enough. Being diligent about product and process and promises matters more; paranoia is a motivator, not the work product. Perhaps nothing is worse than when a dominant company has hegemony over markets and gets sloppy.
IBM was a notoriously bad user of computing, and so was Microsoft until it got serious and built a cloud. We are learning from the conference call going over the third quarter 2022 Intel financial results with chief executive officer Pat Gelsinger and chief financial officer David Zinsner that Intel’s vaunted foundry has not only been making some wrong moves in chip manufacturing process technology development, but has also been letting its chip design groups get away with what sounds like undisciplined designs that result in many more steppings than is usual – and therefore yielding much higher costs – to get chips out the door.
And so, now come the inevitable job cuts at Intel, which will cheer Wall Street no doubt, and if things get really crazy in the coming quarters, perhaps the dividend will go out the door and the massive capital budgets Intel needs to get both its chip architecture and its chip manufacturing operations back on track will have to be curtailed.
High Stepping It
It is Gelsinger’s job to be optimistic and but also to be scientific about analyzing Intel’s many problems and then fixing them, many at a time. Lou Gerstner did this brilliantly with IBM, Steve Jobs with several iterations of Apple, and Satya Nadella with Microsoft.
Let’s start with how bad it is for Intel right now in the datacenter, which is all we really care about at The Next Platform. (We do appreciate that PC processors help ramp chip processes and also give Intel the profits to invest in future chip processes, and therefore the synergies between foundry and design and PC and server are important.) As we go through the numbers, we will talk about what Intel is doing to get its foundry operations back in order and where it thinks it is at with its server chip roadmap as context.
In a word, it’s bad.
The Data Center and AI group that sells server CPUs, chipsets, and motherboards as well as specialized chips for AI processing, saw a 27.3 percent decline to $4.21 billion and posted an extremely anemic $17 million in operating profit, which might as well be $0 when compared up against the $2.29 billion in operating profit that Intel posted against $5.79 billion in DCAI sales in the year ago period. This was with untold how many hundreds of thousands of units of the still impending “Sapphire Rapids” Xeon SP processors shipping to the select, chosen hyperscalers and cloud builders who are always in front of the line and who always get CPUs ahead of launch. We think that they will consume all of the Sapphire Rapids chips that Intel can make using its SuperFIN 10 nanometer process (which has its issues) until early next year, which is why we probably won’t see a Sapphire Rapids launch until early next year.
And even if Sapphire Rapids is the fastest Xeon in history to get to 1 million units, AMD is getting set to leap ahead on November 10 with its “Genoa” Epyc 9004 series, which is probably going to leave Intel mostly with “supply wins” instead of “design wins” at a lot of customers. Intel may be able to improve its average selling prices with Sapphire Rapids, but supply wins is the only reason why. If AMD had unconstrained supply, Intel would not just be laying off people in sales and marketing, as it has announced, but it would be cutting even deeper into the organization.
To a certain extent, we believe that Sapphire Rapids, which was supposed to compete against the prior “Milan” Epyc 7003 processors, has already done battle and got its supply wins against the Genoa Epycs, which no doubt have been shipping to the same select, chosen hyperscalers and cloud builders for some time. Genoa was expecting to come up against Intel’s “Granite Rapids” with Genoa, and the compares are gonna be a whole lot easier with Sapphire Rapids.
Which is why we think Intel got its server CPU clocks cleaned in Q3. It already happened. And the preliminary results from AMD for its Q3, which we covered three weeks ago, show this. With its datacenter business up 48.5 percent to $1.64 billion in our model, that’s an incremental $537 million for AMD against a $1.58 billion decline for Intel. Some of that decline is that people don’t want to buy older Xeon SPs anymore when they know new server CPUs are coming from Intel and AMD, some of that might be due to macroeconomic issues, and some of that is the price/performance advantage that AMD Milan and Genoa Epycs have over Intel Ice Lake and Sapphire Rapids Xeon SPs. It is hard to carve that up without having the inside data from both companies.
What we can clearly see is that Intel is slashing prices to win business at the same time that SuperFIN process costs are higher than Intel would like. But, Sapphire Rapids is in “production release qualification” with the hyperscalers and cloud builders, so there is that at least. Gelsinger said on the call that “Emerald Rapids” Xeon SPs, which will be socket compatible with the Sapphire Rapids chips and which uses the same Intel 7 (10 nanometer SuperFIN) process as Sapphire Rapids, is “showing good progress and is on track for calendar year 2023.” Gelsinger added that the first stepping – called A0 in the chip lingo – for the Granite Rapids Xeon SPs is “out of the fab and yielding well” on its Intel 3 process, which is what Intel calls its 5 nanometer EUV process. As we reported back in February, this process was yielding so well that Granite Rapids was jumped from the 7 nanometer Intel 4 process to Intel 3. In any event, Granite Rapids and its companion with low-powered, energy efficient cores called “Sierra Forrest,” are on track for 2024.
“So the next three generation products are all making very good milestones,” Gelsinger said of the Xeon SPs mentioned above. “I really feel like the worst of our execution is behind us. And we’re really starting to see some enthusiasm, momentum, excitement building in those teams as they turn the corner on these products. We do think that the market is softer on the enterprise side and somewhat on the cloud side as you reinforced by some of the other comments from others. That said, as we ramp these products, it’s all about having the best product to gain share, to gain ASP, to improve the margins of the business. And we now feel like our portfolio is taking share or taking shape to accomplish exactly that. And we’re going to be aggressive. We’re going to fight for every socket. This is a game where we have to reestablish ourselves in the marketplace.”
We eagerly await the architecture advances in these three chips to take on what AMD will bring to bear with Genoa and Zen4 cores and its successor, “Turin” with Zen5 cores. If Intel wants to play beefier Zeon P cores against more numerous Zen4 and Zen5 cores, then it had better be doing some pretty substantial architectural changes in those future generations of Xeon SPs.
The Rest Of The Intel Datacenter Business
The DCAI group is not the only datacenter game Intel has, of course. A very large part of its sales in its Network and Edge group (NEX) because of custom Xeon D processors, FPGAs, and network ASICs and network interface cards, ends up in datacenter. And edge is just a massively distributed datacenter as far as we are concerned, and that is how we look at it. And at this stage of development, a considerable portion of sales and costs for the Accelerated Computing Systems & Graphics (AXG) group will be for datacenter and edge products.
NEX posted $2.27 billion in sales, up 14.1 percent as a bright spot. But operating income fell by 85.3 percent to $75 million as demand for Xeon D and other custom Xeon chips aimed at networking slowed. Demand for Ethernet products (we suspect this is mostly NICs, and mostly due to shortages of Nvidia’s ConnectX NICs) was up and so were 5G and edge circuitry. Macroeconomic concerns and inventory issues (as with PC chips) were an issue, and that pulled down profits as Intel cut prices or customers went to lower-priced SKUs at the same time Intel is investing in its NEX roadmap.
The AXG unit is a big part of Intel’s future, and right now, it is all pretty much investment with relatively larger losses. The Blockscale blockchain chips are apparently doing well, and are a datacenter product. There is heavy investment here for the video encoding, streaming, and compute variants of the Xe GPU architecture, which is piling up in operating losses.
It is very hard to get a sense of all of this over time from the Intel quarterly presentations, so we put together a table summarizing the data we have on the new groups thus far. We won’t be able to add Q4 2021 data until we see Q4 2022 data and Intel does the compares. Take a gander:
Here is how the legacy Data Center Group from 2009 plus the combination of DCAI, NEX, and half of AXG look as a proxy for the datacenter business:
But as you know, DCG was never all of Intel’s true datacenter business, and so we have converged out two models in the old and new presentations to create a unified “real” Intel datacenter business profile with its revenue stream and operating profits between 2015 and 2022:
Depending on how you allocate those AXG losses, you can change the profitability of that “real” datacenter business at Intel. We took our best stab at it above, and reckon that this whole business is in the hole to the tune of $105 million, down from just a tad under $2.6 billion in the year ago period, against revenues of $6.37 billion, down 22.6 percent.
Which leads us to the cost cutting that Intel announced. Intel took a $664 million restructuring charge in Q3, and will take another one of similar size in Q4 of this year, to reduce costs. This is part of a program to cut costs by $3 billion in 2023, with one third of that cost coming from sales and two thirds coming from operating expenses. Between now and 2025, Intel will be cutting somewhere between $8 billion and $10 billion in annual costs, with one third of that coming from operating expenses and two thirds coming from sales.
An observation: If your products have the right speeds and feeds, use the right processes, and come out on time, you don’t need to push them with sales and marketing, they get pulled by customer demand. And that might mean the Intel co-marketing money that has propped up a lot of the OEMs for many years is going the way of all flesh. We shall see.
These are huge cuts, obviously. When Intel says operational costs, it seems to be aiming at the foundry business, which means Intel will be more rigorous about designs and steppings with its own chips. The PC, server, FPGA, and networking chip design teams inside Intel will play by the same rules as Intel Foundry Services is expecting its external customers to play by. The free-for-all is over here, and within the other groups too.
“These savings will be realized through multiple initiatives to optimize the business, including portfolio cuts, rightsizing of our support organizations, more stringent cost controls in all aspects of our spending, and improved sales and marketing efficiency,” Zinsner explained.
We wonder where the product cuts will be, as you probably are wondering, too.
I think it’s disingenuous to talk about revenue and profits without discussing R&D expenditure.
Hi. I have said this in the past. Intel won’t stay on 10nm forever. By 2025 they will be on an equivalent process to AMD, which will allow them to close the gap with Epyc in terms of number of cores and power per watt. Meaning that AMD’s design wins are on borrowed time. However, Intel’s supply wins aren’t going anywhere. Intel owns their own fabs – even if they did sign a financing deal where third parties will get 50% of the fab revenue for quite awhile – and AMD is limited by how much capacity they can finagle from a TSMC that also services Apple, Qualcomm, MediaTek and Nvidia customers that give them higher volumes and more money, plus TSMC is raising prices.
Unless – that is – AMD chooses another fab. Samsung technically reached 3nm before Intel, but the issues referred to here https://www.semianalysis.com/p/samsung-electronics-cultural-issues may prevent AMD from relying on them. So will AMD need to choose Intel Foundry Services to close their supply gap? You should not rule it out.
Finally, you did not prominently mention ARM in the datacenter this time around. It is relevant because my belief is that once both AMD and Intel are on equivalent process nodes in 2025, ARM servers will take more market share from AMD than Intel. Also, please note that the primary competition for Intel AXG will be Nvidia. The reason: where Intel has viable competitors to Nvidia’s data center and cloud HPC and ML/AI tools, AMD does not.
See nine years of Intel Xeon and five years of AMD Epyc full line supply here at WW open market share here September 2013 through October 8, 2022;
https://seekingalpha.com/instablog/5030701-mike-bruzzone/5661542-intel-antitrust-litigation-today-november-10-2021-united-state-congress-and-states-ag
Highlight of the Xeon nine-year supply wave is Intel working with the channel to get the glut of Haswell v3 and Broadwell v4 supply sold off before ramping Sapphire Rapids preserving channel accumulated capital for new production procurements from whatever primary vendor.
Pursuant component vendor product volume classically determined $1K average weighed price less cost of sales to determine gross per unit, any category volume share on gross per unit divided into gross revenue, and channel supply share specific inventory holdings, channel supply share is the only truly independent and verifiable market share standard in relation such a determination from any design producer 10-Q or component design producer’s constituent ‘OEM’ said quarterly finished goods volume.
As Mr. Morgan aptly points out “supply wins” instead of “design wins”. That is the best tech does not always win?
Mike Bruzzone, Camp Marketing
“If Intel wants to play beefier Zeon P cores against more numerous Zen4 and Zen5 cores, then it had better be doing some pretty substantial architectural changes in those future generations of Xeon SPs.”
Zactly – the debate (not just here) makes it sound Intel just has to get their node act together & they are back in the game.
BS. Their key problem is architecture. Their 14nm would compete w/ amd 7nm if they had chiplets etc. Like amd, but they dont & they wont. Its a very long haul project they seem to have little stomach for.
The above advances wont provide them with the ~seamless expandability of large teams of low latency chiplets, & all the other benefits of zen – economics, efficiency….
I see little cause for optimism for them.