Although Intel (INTC) delivered a nice beat with a surprise all-time record data center revenue, my investment thesis – which recently has been neutral/hold – hasn’t changed. This is because the beat was solely due to 5G and enterprise growth, which had to offset a disastrous fifth consecutive quarter of severe double-digit cloud declines. (In stark contrast to Microsoft’s (MSFT) Azure results just the day before.) It seems CSPs are either getting their CPUs from AMD (AMD) going forward, or are waiting for Intel’s 10nm Xeons, but either way Intel is to blame here for its sloppy data center execution.
So as I have warned investors recently, the worst may still be to come for Intel. Meanwhile, Pat Gelsinger has been brought in to solve these issues, so he isn’t to blame. Indeed, he has bet on the long term, and Intel added an absolutely massive ~11K employees in 2021 (of which ~4K in Q4 alone), over half an NVIDIA (NVDA), and R&D has surged to what is now a $16B run rate. The execution machine is being rebuilt and should deliver strong results in the years ahead. Just ignore the bloodbath in the near term as Intel has handed its cloud business to AMD on a dinner plate.
If some of the statements seem a bit contradictory, it’s likely because my opinion about the quarter’s results changed a few times. Initially, I was lyrical about the big data center beat – which going by the stock reaction went completely unappreciated. Then, however, I turned bearish as I saw the cloud results. (Intel’s enterprise hasn’t grown for the last nearly a decade, so it is unlikely that the current growth is sustainable, contrary to the cloud where Intel had a 20-40% CAGR over the last decade.)
Thirdly, after analyzing the results a bit more, my thesis about market share losses to AMD has tempered a bit in favor of CSPs perhaps waiting for Ice Lake and Sapphire Rapids to place their next orders. As such, one could be very bullish: if Intel is delivering all-time record data center sales despite CSPs still being in “digestion”, then what if those start to buy again? In the bullish case, DGC could then significantly outperform. (In the bearish case, CSPs don’t return and enterprise returns to a decline next year on through comps.)
Fourth, despite the DGC and revenue beat, however, ultimately Intel posted just 4% YoY growth, so one should not become too bullish yet. On the other hand, the sell-off does provide an opportunity for those looking to accumulate some shares for a long-term investment.
So contrary to some analysts who may like to downplay Intel’s results (and ignore the big DGC beat), the stock price isn’t (and wasn’t) priced to expect any beats in the first place. So overall, the possible for upside below $50 outweighs any risks, with even more upside possible in the long term if Pat Gelsinger’s strategy materializes.
Data center: anemic cloud results and 10nm ramp
Reviewing my most recent track record, it may seem correct that investors should perhaps do the opposite of what I said: last quarter I predicted a decent Q3 beat, but Intel even (slightly) missed its own guidance. More recently, I downgraded Intel to neutral/hold, leading some to suggest this might be the bottom. For Q4, I suggested investors should not expect a beat. But that’s exactly what Intel did. However, I would argue the numbers aren’t as rosy as they may seem, so hear me out.
To be fair, like most bulls undoubtedly, I was initially in awe upon seeing that the data center had pulled an all-time record quarter out of its hat, basically out of nowhere (given the downside Q3 last quarter). This is something AMD (AMD) bulls likely would have deemed impossible anymore at this point. Indeed, after last quarter’s data center disappointment, which Intel blamed on China, I asked why, if there was a shortage, Intel wasn’t able to redirect sales from China to elsewhere. Well, it seems Intel proved the doubters wrong in Q4 by returning with a vengeance.
However, it is at this point that the more curious investors will look for the segment breakdown results. Here’s what they show for YoY growth: cloud -5%, enterprise +53%, comms +22%.
What this shows is that the data center was carried big time by enterprise, and comms further saving the day. This is significant because Intel for over half a decade has struggled with mixed enterprise performance given the rise of the cloud, and now it turns out enterprises are still buying (their own) servers. In other words, enterprise is the segment that Intel for years has assumed to be a more or less flat business.
This, in turn, means that one should be cautious about extrapolating the current growth, as the more prudent assumption would be that this is rather a short-term surge (for whatever reason, but likely due to enterprises catching up from underspending in 2020) rather than a structural return to growth in the enterprise. Although Intel did seem to expect this demand to continue at least for the next quarter.
Okay, so let’s play devil’s advocate and praise Intel for at least not letting AMD (solely) fill in this surge in enterprise demand. Then, in addition, there is even more to like about the results: Intel continues to capitalize on its quasi-monopoly in software-defined networking (SDN), which for years I have called one of Intel’s most important growth businesses, and yet again networking delivered with 22% growth. Note that this business delivered (over) $6B revenue in 2021, so the strong growth at scale continues as SDN gains ever more momentum with the buildout of 5G.
However, this is unfortunately where the praise must end. Because for all those great results (one of which could be rather temporary, the enterprise, and another one due to structural growth, networking), we now have to turn to the cloud.
It was just the day before that Microsoft delivered another truly stunning quarter with yet another re-accelerating quarter for Azure, which has remained growing at nearly 50% for as long as anyone can remember. I remember some analysts saying Zoom’s (ZM) 300% (or whatever the exact value) growth in the midst of the pandemic was one of the most impressive quarters ever, but re-accelerating a ~$40B run rate business to 47% growth is arguably on a whole other level still. So if this was 2018 with Intel’s 97% market share in the data center, one might have wondered how much Intel may have grown its revenue.
Instead, cloud delivered a 5% decline. And it gets even worse. Out of curiosity, I checked back to the Q4’20 results, and those showed another 15% YoY decline. To be fair, in 2019 cloud grew 48%, but this just goes to show that Intel’s cloud business seems to have run into a brick wall after a decade of double-digit growth. Note that cloud has been weak all year, so at this point it seems the only explanation is that this brick wall is called AMD Epyc.
Indeed, during the Q&A, an analyst noted that CSP has now been declining for no less than 5 consecutive quarters, with the minimum YoY decline being 15%. What I actually disliked was Pat Gelsinger’s response, which was perhaps the first time I disagreed with what he said. Instead of addressing the question, he turned his attention to the edge, which he argued would see “explosive” growth in a few years. In my opinion, here Pat Gelsinger is just too easily dismissing what for many years has been Intel’s most important business. As indicated, Microsoft’s results show that cloud is experiencing “explosive” growth right now. Cloud isn’t some future possibility like robotaxis, but something that should drive Intel’s results at present.
There is even more I disagree with. Intel’s GM for Xeon went to Twitter to celebrate the Ice Lake ramp, reiterating Intel’s previous comments about the 1M Ice Lake units shipped in Q4, and that Intel shipped more Xeons in December alone than any other competitor (read: AMD) during the whole of 2021.
While this is neat, in fact it doesn’t match the ramps of Skylake and Cascade Lake, both of which Intel called its fastest ramping Xeons ever. For comparison, three quarters after launch, Skylake already approached 50% of total units, while I estimate Ice Lake is still only at around 12.5%. This means Ice Lake actually seems to be ramping ~4x slower.
This comparison clearly speaks for itself. Nevertheless, unfortunately it is quite logical that Intel is taking its time with ramping Ice Lake given that Intel also reported a 30% YoY wafer cost reduction for 10nm.
Indeed, there is another possibility that may have contributed to the cloud decline over the last year: Intel isn’t able to supply CSPs with the 10nm parts they are waiting for. Further, it is obviously known that cloud service providers are quick to adopt new technology, so they may have been quick to abandon Intel in favor of AMD. (Intel indeed confirmed it has lower market share in cloud.)
Given the slow 10nm ramp, by Intel’s standards, it seems 14nm will continue to be responsible for most of DGC revenue into 2022. As Pat Gelsinger said, DGC hasn’t ramped a new process node for 5 years. So at this point, the main question is if data center will get the same “5 nodes in 4 years” treatment as DGC.
For example, Pat Gelsinger has said he has to undo a decade of mistakes, and one of the last (and also worst) mistakes Intel made before his appointment was making Meteor Lake the Intel 4 lead product instead of Granite Rapids (as was the plan before the delay).
Lastly, the difference in execution compared to PCG (which has to use exactly the same 10nm node) remains stark, since Intel just said Tiger Lake has been its fastest ramping notebook CPU ever, and Alder Lake may become Intel’s overall fastest ramping CPU ever.
Intel posted all-time record revenue of $20.5B, exceeding guidance by $1.3B and up 3% YoY. Non-GAAP gross margin was 55%, driven by the higher revenue. Remember that gross margin for the next two years is expected around 51-53%, driven by (1) increased 10nm volume in desktop and data center, (2) fab buildouts, (3) start-up costs of new nodes, and (4) lower gross margin of new nodes at early stages.
The business results are shown above. CCG was lower driven by the Apple (NASDAQ:AAPL) modem and Mac transitions, as well as due to a tough comp after the strong Q4’20 (driven by low-end consumer then). In Q4’21, revenue shifted back to “big core”, leading to higher ASPs. The earnings presentation with the full breakdown can be found here.
Going on a bit of a tangent, during the Q&A, one topic was with regards to how to square the circle of Intel reporting quite severe unit declines in PCG despite the supposedly high demand. Although I don’t subscribe to the view that Intel magically lost a bunch of market share in one quarter, I don’t have an explanation besides repeating what Intel said, that some of the comps are misleading due to the loss of the Apple business. It seems some people assume that shortages mean there is infinite demand, which obviously isn’t true either; Intel also reiterated that there remain many issues across the industry with shortages regarding all sorts of components.
As discussed above, DGC had a record $7.3B revenue, up 20% YoY. Looking further, IoTG and Mobileye both had a quite soft finish of 2021, with basically flat revenue QoQ. My bullish prediction in early 2021 was for >$400M revenue in Q4 for Mobileye. NSG also didn’t continue growing, but this business has been sold now.
Lastly, PSG was up 15% and up a bit QoQ. Intel has been saying for a few quarters that this business has been heavily affected by third-party shortages, and said that it expected PSG could have delivered $500M revenue more in 2021.
Still, one could note Xilinx (XLNX) is operating in the same environment, and since Intel’s acquisition, the revenue gap has actually widened instead of decreased. I have reviewed FPGAs a while ago, and concluded Intel has actually achieved technology leadership. So from that view, the acquisition has been successful. Another thing to note is that Intel “lost” quite a bit of FPGA revenue due customers moving to Intel 5G ASICs, which are reported as part of the data center adjacency segment. I would actually estimate that combining this ASIC (likely on the order of $300-400M) and FPGA revenue, PSG is about similar in size as Xilinx.
Digging a bit deeper into DGC, interestingly, when I looked back at my prediction at the beginning of 2021, Intel’s data center revenue in Q4 ended up pretty much exactly as I had predicted.
Nevertheless, there are many question marks going into 2022. Cloud has now been in “digestion” (and/or market share loss) for 5 quarters. Will this continue indefinitely, or will 10nm for example start to drive growth? In the latter case, one could wonder how much higher Intel’s DGC revenue could grow if Intel already achieves a new ATH during a digestion phase. In that case, Intel could easily deliver another record year in 2022.
On the other hand, if as suggested AMD is eating Intel’s cloud lunch big time, and the Q4 enterprise weakness indeed turns out to (indeed) be anomaly, then growth may be limited or even negative. Given these uncertainties, I would caution investors about forecasts for Intel in 2022 at least until the Investor Meeting.
Below are DGC’s segment results in their full glory. Again, note how cloud hasn’t returned to growth, contrary to 2020. Again, in the most optimistic scenario where one continues to ignore AMD, then one might assume CSPs are waiting for Ice Lake to upgrade their servers. But this would prove Ice Lake should have launched in 2020 already, and should have exited 2021 being responsible for the majority of DGC revenue.
Lastly, the graphs below shows the trend in revenue. Unfortunately, since AMD doesn’t break out its data center revenue, it isn’t possible to make a graph detailing revenue share trends for both. The graph makes a simple interpolation for how Intel could deliver DGC growth in 2022. Perhaps the truth really is somewhere in the middle, and CSPs have been digesting their CPUs for the last five quarters after the exceptional 2020:
For some final thoughts, I have to mention one more Pat Gelsinger, as Intel claimed that it had shipped more Xeons in December alone than AMD shipped Epycs in all of 2021. Going by these comments, perhaps I have to come back to some earlier suggestions that Intel is giving its cloud business to AMD.
However, the issue here is that as discussed Intel has a very successful 5G networking business, where AMD still has literally 0% market share*. These 5G parts tend to have lower ASP, so Intel may be selling a ton of those, which could mean that while AMD may indeed have about ~12% market share overall, AMD’s market share in cloud may already be quite a bit higher. That AMD growth, then, could be the difference between a 5% decline and double-digit growth.
*Update on the 0% market share: of course Nokia just announced that it uses Epyc for its new servers.
Full year results
For the full year, Intel eked out its sixth consecutive record year. One could note that at the start of 2021, not a single analyst (except this author) thought that could or would happen.
Capping off that same streak is also CCG, despite the modem and Mac headwinds. IoTG recovered with 33% growth, although if it had followed its long-term double-digit growth rate uninterrupted, it should have been a bit higher by now. Mobileye was up 43%. Although, as discussed, its sequential growth has not continued, Mobileye nevertheless had a strong year in terms of design wins, and started 2022 by announcing some L2(+) partnerships that should drive ASPs up. PSG was up a bit, although overall flat for half a decade or more.
Intel guided for more or less a flat Q1 YoY, with gross margin starting to drop to the 51-53% range that Intel disclosed last quarter, driven by Intel 4 start-up costs. I would expect there to be some PC downside, offset by data center strength given the exceptionally low DGC in Q1’20. Intel will disclose more details at the Investor Meeting.
In addition, Intel announced a new reporting structure, which aligns with what Pat Gelsinger already said last year he saw as his six businesses:
- Data center and AI (including FPGA)
- Networking and edge
- Intel Foundry Services
Execution machine is back
What I have actually been focusing on most for the last year has not been the financials, but the spending. Pat Gelsinger has been undoing the BK 2016-2017 reorg (which affected well over 10K employees) and the Bob Swan cost cutting with great and swift force, and this has only been accelerating through the year.
This continued in Q4. Intel added another ~4K employees (surely mostly engineers, although likely also some in manufacturing) in just one quarter, akin to adding a quarter of an AMD or Nvidia in one single quarter, bringing the total to ~11K added in 2021. In addition, R&D has been growing at a ~10% rate since Pat Gelsinger joined, and crossed the $4B mark in Q4.
Given this pace of growth in R&D and employees, it seems these numbers may continue to increase even further going forward. This all means the execution machine has been restored, Intel is indeed back. Nevertheless, this also provides evidence that it will take years for the results from all the newly added teams to become visible. (As discussed, the products currently driving Intel’s revenue are still from the old Intel.)
But obviously, nothing is more exciting than the prospect of Intel exercising its sheer size to destroy the competition. This is for example exactly why Raja Koduri joined Intel from AMD in 2017 to take on Nvidia. But it seems he actually had to wait four years for the CEO to finally greenlight his investment plans. Note that while it is not visible with regards to market cap and market share, in terms of revenue, Nvidia is still small compared to Intel.
Still, some investors may be tempted to do some napkin math: R&D may top out at perhaps $20B, and capex should be around $30B going forward. Intel is (and will be) investing about as much as Nvidia’s and AMD’s revenue combined. Clearly, there won’t be a lot of money for stock buybacks left.
A summary of the earnings call can be found in the tweet below.
Intel highlighted that third-party shortages will persist into 2023, although Intel is mostly shielded from foundry price increases as an IDM. All process nodes remain on track. As noted, Intel shipped 1M Ice Lakes, but also said it shipped a record number of servers. Intel’s previously disclosed record was 8M in 2018, confirming my statement that the Ice Lake ramp remains negligible. The reason for this is the low 10nm yield, with Intel saying it drove a 30% wafer cost reduction over the last year. Intel had further records for substrates and PDK releases, and signed long-term supply agreements across a variety of vendors.
With regards to IFS, Intel expects Intel 16 customer test chips this year, with revenue following in 2023. Intel 18A PDK has also been shipped to three unnamed customers, proving that IFS is not experiencing the same issues as, for example, the data center.
We’ll give you the proof points you should expect to see in 2022 that show that we are on track for our long-term plan, all backed up by the transparency and accountability that our new reportable segments will provide.
Stock and predictions
It’s hard to forecast Intel’s 2022. This is due to the unpredictability of the data center: what will enterprise do (Intel says strong demand will continue due to strong backlogs), what will cloud do (Intel expects it to pick up, which does seem reasonable after a 5-quarter digestion)?
Overall, though, despite being so bearish about Intel’s present execution and data center portfolio, Intel’s results are decent, so it isn’t like Intel’s financials will collapse. As stated, if Intel can continue the momentum with which it ended 2021 in the data center, then Intel’s results should be solid, and the concerns about gross margin declines are hugely overblown given that this is a year of factory start-up costs as Intel 4 starts to ramp by the end of the year.
Hence, it seems the market had already forgotten about Intel’s guidance it gave in Q3 (to which the stock reacted exactly the same: down) since nothing has really changed, and in terms of revenue, Q1 should be a solid start to the year (excluding NAND).
For the rest, investors will be looking forward to the investor meeting, and any detail Intel may give about 10nm yield and its expectations for Intel 4 (and perhaps even 20A). The main issue that is impacting gross margin is primarily 10nm yield. So as Intel gets to new nodes, it will have the double headwind of what should be more economical nodes with higher yield (due to the use of EUV) and an improved cost structure due to increased product competitiveness.
While Intel did post a nice beat, investors will now be looking forward to AMD’s results to see how much higher they could have been still. In particular, Intel’s cloud results look outright appalling the day after Microsoft’s Azure reported another incredible quarter of high growth at scale. Intel will surely be looking to break its streak of five consecutive double digit cloud declines next quarter, but obviously Intel’s issue is that it doesn’t really have anything to sell to CSPs, given the rather slow pace of the Ice Lake ramp. It is bleak to realize that most of Intel’s revenue in 2022 in the data center will still come from 14nm.
Looking at the bigger picture, though, and a more encouraging outlook emerges. Intel added a monstrous (nearly) 11K employees in 2021, with the rate of hires accelerating throughout the year, which suggests Intel isn’t done yet. R&D is seeing similar ramp in spending. Pat Gelsinger realized the only way he can save Intel from its demise is by hiring more people to build better technology and better products.
So in conclusion, people don’t call it turnaround for nothing. In that regard, while the process roadmap provides the best indication for when improvements should be expected, the question at this point that has not been answered yet is if the data center will actually get the same “5 nodes in 4 years” roadmap as the PC. Investor Meeting in February should provide further information.