Intel’s second quarter is pretty much a carbon copy of the first three months of 2024 when it comes to revenues across its newly constituted groups, and with an operating loss that is twice as big. And an outlook that basically says it is going to be tough slogging in 2025, with the situation improving as both foundry processes and better chip products hit the market in 2026.
As we exit 2024, there is a non-zero chance that AMD’s official Data Center group will have slightly more revenues than Intel official Data Center and AI group, with both on track to hit somewhere around $12 billion or so. If you add in the edge and datacenter portions of Intel’s Network and Edge group (NEX) as well as the Altera FPGA business, Intel’s “real” datacenter business, as we estimate it, will still be on the order of 25 percent larger than the “real” datacenter business at AMD.
And in 2025, unless current trends do not change, there is no reason to believe that AMD’s datacenter business could not be significantly larger than that of Intel, and this crossover point could happen even earlier if an expected recovery in general purpose server spending does not materialize as expected in the second half of this year.
So when Pat Gelsinger, the prodigal and presumed savior of the company that he loved first and loves best, tells Wall Street that this is the most significant resurrection of Intel since it had to exit the memory business for the first time – we’re not talking about 3D XPoint and flash here, people, but DRAM back in 1985 – he ain’t kidding. This might qualify as a miracle when it is all said and done.
Intel’s numbers for Q2 are bad, to be sure. But the announcement that it was seeking another $10 billion in cost savings, that it would be halting its dividend payments to investors immediately, and that it would be laying off 15 percent of its workforce – it has 116,500 people, plus another 8,800 people in subsidiaries including Mobileye, Altera, and some flash employees that have not been spun out yet to SK Hynix – has sent its stock into a tailspin and its people into panic.
Intel stock is down 26 percent as we go to press. And 15 percent of the core workforce is 17,500 people who are going to get pink slips – and that may not be the end of it, we think. The dividend was already chopped by two thirds earlier this year, so most of this decline in Intel’s market capitalization will be worry about the layoffs and the mad dash to cut the bloat and get leaner and meaner. There is always worry in these situations that the cutting has to go to deep, and it will not just remove fat, but muscle and gouge some bone.
There are going to be some big accounting actions and writeoffs coming as a result of these actions, too, so brace yourself, Wall Street. Intel did not talk about these, but said most of the layoffs will be done by the end of 2024.
Sometimes, such harsh actions work. IBM, which had its self-described “near death experience” back in the mid-1990s, had tens of billions of dollars of writeoffs – just absolutely unheard of, unimaginable, inconceivable for the original blue chip stock – and laid off 200,000 of its 400,000 workforce as it pivoted to software and services and trimmed its independent systems fiefdoms to get its own costs in line with revenues. IBM eventually got back to 400,000 employees, but the way, and has subsequently sold off a lot of systems and services businesses to focus on being a hybrid platform provider, with Red Hat at the center of that strategy. This is IBM’s fifth rebirth in its Herman Hollerith built punch card machines to do the 1890 census in 1890, which is the true kernel of the company we know as Big Blue.
This rebirth of Intel is more like a company reminding itself of what it learned to do in the mid-1980s: design good chips and make good chips, and be paranoid enough to survive. (Andy Grove was right about “only the paranoid survive.”) It is hard to be paranoid, in the right way, when you are filthy rich with very little competition, as Intel was during the 2010s.
And as Nvidia is doing today, Intel had complete hegemony over datacenter compute and dominance over desktop and laptop compute, and it was able to extract and keep most of the profits of the ecosystem built around its compute engines. Let this be a warning: Intel sowed the seeds of its own near-death experience, as companies always do. It is a wise company that lets its partners make some vig out of their enormous efforts to put your products in the field. Jensen Huang’s chanting “Dell, Dell, Dell” back at Computex in May could be a little too little money, a little too little GPU allocations, a little too late.
But that is a separate, and parallel, allegory for another day. Intel has even rougher road ahead, but it is, in fact, getting its foundry act together. It is getting its chip designs together and it will be relying less on Taiwan Semiconductor Manufacturing Co for the chiplets in its most advanced CPUs and bring them home to its fabs with its 18A, 14A, and 10A processes. It remains to be seen if others will use its foundry for the 14A and 10A processes, but Intel is clearly getting the third party tooling together from Cadence, Synopsys, Siemens, and Ansys together so they can, and we think it will get some business from chip makers looking for alternative etching and packaging. We think it will have a profitable foundry operation and that it will make good chips that people want to buy as well. But we also think the X86 market is going to decline as Arm rises at the hyperscale and cloud builders, and Intel not only knows this, but accepted this years ago and that is why it must have an open foundry business. If you can’t beat Arm CPUs, you have to make Arm CPUs.
“We firmly believe in the IDM 2.0 strategy,” Gelsinger told analysts on the call with Wall Street. “We are building two world-class companies. The forensics that we have done this year – this clean sheet exercise as we could describe it – is building a world-class Intel Foundry and building a world-class Intel Products Group. These efforts, we believe, have identified many opportunities for us to have financial savings. We have launched those aggressive steps today, and we believe that with the new products, a better financial position that we have done for a more efficient operation, that we see the long-term opportunity for significant value creation for all of our stakeholders.”
That is a lot calmer than the situation really is. This is going to be hard to watch over the next couple of years, and even harder to do. We want Intel to succeed, and we have faith that the IT sector will be better when it does. For whatever that is worth.
With that, let’s go through the numbers for the second quarter. Intel rejiggered its financial reporting starting this year, and this is what it looks like now with Q2 added into he mix:
In the second quarter, Intel had $12.83 billion in revenues, down nine-tenths of a point, and posted an operating loss of $1.96 billion, almost twice that from the year ago period and from the first quarter of 2021. Because it has subsidiaries and so many intersegment eliminations, we are only getting an operating income from Intel these days, not a net income.
Intel exited the quarter with $11.3 billion in cash and just a tad under $18 billion in short term investments. Which is enough capital for maybe one and a half fabs these days. In other words, not much if you are trying to be the world’s second largest foundry and compete against TSMC.
If the US government hadn’t given Intel $8.5 billion in cash and incentives as part of the CHIPS Act and if it had not been able to get Apollo Global Management to invest in its Ireland fabs and Brookfield Asset Management to invest in its Arizona fabs, the company would have had to burn that investment fund for cash. And, given where Intel may be going, the company may have to do that anyway.
The DCAI group saw sales decline by 3.5 percent to $3.05 billion, and operating profit was hit by a 41.2 percent decline to $276 million, which represents a mere 9.1 percent of revenues.
Remember when Intel’s datacenter business was consistently delivering operating income above 45 percent of revenues, and often kissed 50 percent? Remember when it was around $7 billion a quarter?
The NEX group had a 1.5 percent decline in Q2 to $1.34 billion, but operating income more than doubled to $139 million. Both DCAI and NEX are dealing with the fact that Intel’s OEM and ODM customers bought a lot of chippery in 2022 and 2023 and they have to burn down inventory. Intel still believes a server recovery is underway in the second half outside of AI servers, and that it still has an advantage as the host processor for AI servers. We shall see.
For the past decade, we have been trying to extract the “real” Intel datacenter business out of its numbers, which was a bit harder when Intel had flash, 3D XPoint memory, various kinds of networking, custom systems, and edge server and telco server stuff spread across various divisions. It is getting easier to figure this, but it is not fun to do so, as you can see in the chart above.
By our estimating, this real Intel datacenter business is down 11.8 percent to $4.62 billion, and flat sequentially from Q1, and operating income is down 55.6 percent to $390 million, representing an 8.4 percent share of revenues.
While the existing Xeon 6 “Sierra Forest” and forthcoming “Granite Rapids” server CPUs will help, AMD has its “Turin” Zen 5 Epyc server CPUs right around the corner, and all the hyperscalers and cloud builders have or will have custom Arm server CPUs. The pie is just not as big for Intel as it used to be – and never will be again.
Inside Facebook’s Future Rack And Microserver Iron
The hyperscalers and cloud builders have been setting the pace for innovation in the server arena for the past decade or so, particularly and publicly since Facebook set up the Open Compute Project in April 2011 and ramping up as Microsoft joined up in early 2014 and basically created a …
Combining AI With HPC To Find Better Battery Designs
The melding of low and high precision mathematics to accelerate the pace of scientific discovery has been a topic of discussion for some time now. During her keynote at ISSC last year, AMD chief executive officer Lisa Su mused that the combo could vastly reduce the energy requirements to scale …
The Year Ahead In Datacenter Compute
For more than a decade, the pace of the server market was set by the rollout of Intel’s Xeon processors each year. To be sure, Intel did not always roll out new chips like clockwork, on a predictable and more or less annual cadence as the big datacenter operators like. …
I suggested one love initiative, all USA and world IT guys to give to intel managerial an alms to get better decisions and not mass dimissal.
“. Let this be a warning: Intel sowed the seeds of its own near-death experience, as companies always do. It is a wise company that lets its partners make some vig out of their enormous efforts to put your products in the field.”
This seems farfetched. There’s always death by a thousand cuts in these situations but the primary cut so to speak was when INTC made Brian Krzanich CEO and he brought in a bunch of bozos in key positions who all ended up screwing the company internally. Brian meanwhile was busy disinvesting(?) in foundry to keep up the earnings and lining his pockets. I wish Pat the best. The man is doing a thankless job.
The reason Arm got its camel nose under the datacenter tent is because Intel was greedy. And none of the OEMs or ODMs really made money in servers. I disagree.
There is a thousand paper cuts, and then there is compelling the eight customers who comprise half of server shipments to do their own thing because you charge too much and then raking enterprises over the profit coals to make up for the lower prices the hyperscalers and cloud builders get compared to them. This is a self-inflicted wound. Not to the heart or the spleen, but in the guts for sure.
Hi. Well researched and written article. One correction. I believe “Intel existed the quarter” should be “Intel exited the quarter”, correct?
Yes, thanks for the catch. Was thinking fast and writing faster.
Having read this I looked up market cap for AMD & Intel. I know market cap in chip stocks is somewhat of a BS metric. Still… $235B to $91B. Who’d have thunk it 10 years ago. Is it time to load up on Intc stock yet, Timothy? I bet Intc hits $10 before the recovery.
I don’t trade in stocks I write about — I actually have a money manager who does things in various funds and I only look once a year — but at some point, this is like AMD in 2015 or IBM in 1992. I don’t think you can count Intel out. Not at all. But damn, this looks hard.
I don’t understand any of it. The company loses 30 billion of capitalization in one day due to an effort to save 10 billion? How did that help?
My main concern is the phrase “focus on our core business” predicates the disruptive innovation that happens when what’s currently profitable prevents development of what will be profitable in the future.
Said another way, Optane memory and Omnipath networking are exactly the types of technologies experiencing rapid growth needed to support training large neural networks. Could anyone other than an accountant have missed that?
In other news Zen5 has full width 512 bit vector units while Intel removed that feature in order to make their P cores compatible with the E cores. Even if the 13th and 14th generations didn’t crash, reducing math capabilities goes in the opposite direction of what’s needed for AI inference at the edge.
From what I can tell, the series of bad decisions made in the name of turning things around have hastened a cascading failure.
I think the Raptor Lake (13th-14th gen Core) chip-damaging overvoltage microcode bug is playing into Intel’s stock dive here, possibly more than the financials that are similar to those of last quarter. It would have been better if the outfit could have performed as in Q4 2023 (or better) to be sure, especially the Foundry and Client Computing Group, which would have validated the turnaround started in Q1/Q2-2023, and generated consistent overall profit IMHO.
The 128+ P-core Granite Rapids Dwayne Johnson Xeon 6 CPUs are just around the corner to help produce good news (though they’ll face 128+ Zen 5 core Turins … a face-off that should be quite the spectacle!). For more good news, Aurora is most likely getting tuned as we speak to perform closer to its 1.98 EF/s theoretical Rpeak, and the machine is already #1 in HPL-MxP with 10.6 EF/s of mixed precision oomph ( https://hpl-mxp.org/results.md ). Add to that its #1 position in io500 for Western democracies (China otherwise leads this list). So, overall, I think there’s demonstrably good tech coming from the company, and expect it to rebound by Q1 2025 at the latest (even if the Foundry group is not yet profitable by then).
$21 for Intel stock is a bargain IMHO (non expert) … but I’d bet against emerth’s bet of $10 before recovery … it won’t go below $19.68 (eh-eh-eh!).
Fair enough, and props for the arms length. I agree it looks like AMD in 2015.
I wanted to write this and you wrote better than me, so thanks.
I dont agree with TPM statement “We want Intel to succeed, and we have faith that the IT sector will be better when it does.”
Intel has made the history, but as you wrote for many years has planned and executed illegal activities to prevent customers benefiting from better alternatives.
Now there are better alternatives such as Nvidia and AMD + others in the ARM realm + other companies innovating in the accelerated compute.
I whish AMD can get at least 50% of the CPU and GPU market to balance the odds.
Intel can just become the new IBM, and thanks for all the fish.
I don’t want to return to Intel hegemony. I just want two foundries and ten chip designers instead of one and one and a half.
“We think it will have a profitable foundry operation…” When? And at what revenue and margin assumption?
The article and much of the stock world doesn’t even know that a lawsuit is pending for Intel because it’s just been revealed that all of their 13-14 gen processors have a 100% fail rate and legal action is necessary. This will drive the stock failure down along with the brand name for a while.
INTEL is tightening it’s belt and strengthening it’s muscles. It has more up it’s sleeve than yoyu might may invasion. I myself doubled my investment as soon as I heard the dreaded news. The doors are not shutting. The wind shall open them so wide you will be astonished from the result. HAIL almighty Intel. I shall never dessert a sure thing.
Once Intel gets their new plants online, I think we will see a new motivated Intel. I think Pat has a tough road ahead, but I believe Intel can indeed become an enviable company again.
I won’t be praying for Chipzilla.
For as long as the x86 market has existed, Intel has played games to punish competitors, charge for performance and try to pull the wool over our eyes. They know about vulnerabilities that they don’t address in a timely manner because addressing them would lose performance, so in essence they’re tricking us with false performance comparisons. Heck, they’re doing that now by saying the motherboard makers are responsible in part for 13th and 14th gen issues, when it’s Intel that can’t give clear guidance, because if they did, they’d lose all the important benchmarks when compared with AMD.
They just don’t seem like the kind of company worth trusting, either in the past or now. They’re a one trick pony, and now that they’ve abused this pony too much, that one trick isn’t even very good any more.
It’s time to move on.
Intel pointed out that the inflation for TSM processing would be hitting the margins for Lunar Lake. The same could be said for any of TSM’s customers on advanced nodes or using their advanced packaging.
The inflation should continue to rise as TSM customers will need to pay for more advanced EUV processing, backside power delivery, GAA transistors, hybrid bonding and co-packaged optics if developing chips for three years out.
Xeon competitors will soon need PCIE6/ PAM4 transceivers, CXL 3.0, MCR DIMM support, Ultra Ethernet, AMX tiled matrix processing.
Lunar Lake competitors will need a comparable integrated gaming GPU, WIFI7, h.266 hardware decode, in-package memory. Intel’s integrated Xe2 GPU also contributes to platform AI processing sum of 120 TOPS.
I suspect that this problem has been going on for some time. I got an inside look at Intel some 15 or more years ago when I became an accidental Intel employee (they’d purchased a small company I’d just joined). What I witnessed was a lot of waste, misplaced effort, overpayment for technologies and so on which at the time I figured was just Big Company stuff (I’m more used to lean/starving startups.) However the thing that really attracted my attention was that there was basically two semiconductor fabrication processes, Intel’s and “everyone else’s” (i.e. TSMC etc.) Intel’s process worked, its had been and continued to be successful but you could see as process technology developed they’d have to invest more and more for less and less return.
This highlights the problem with a company with an extremely successful product line. For a time they’re swimming in cash so it looks like they’ve truly got the golden touch. Unfortunately, like with just about everything else, unless you’ve figured out how to enforce a de facto monopoly you’re only as good as you last big success/win. Especially when you’re on top — you become the target for everyone else to beat.
“AMD’s official Data Center group will have more slightly revenues than Intel official Data Center and AI group”
“slightly more” perhaps?
Intel reportedly has EUV processing in Fab 52 coming on line this year and Fab 62 next year. Their CHIPS act document shows they intend to update Fab 42 for 18A. That’s a lot of EUV capacity coming online that is currently not contributing. Ohio fabs are probably three years out, and the same for Israel now that they are on the back burner.
Anyone thinking Granite Rapids doesn’t have a place should take a second look at the 8800MT/sec MCR DIMM support and the per core AMX tiled matrix acceleration. They’re also ready for the CXL 2.0 memory pools that have been talked up for the last few years.
Granite Rapids will sell, no question about it. But Turin will be here shortly, too.