Intel: Time To Sink Or Swim
Summary:
- For Intel, 2022 has become a black page amid the turnaround. Intel now needs to turn the page.
- Its (internal) progress in 2023 will determine whether it will sink or swim in 2024+. Intel previously failed in these final stretches of development, so the stakes are high.
- There are stark parallels with how AMD remained very cheap for very long, even though everyone knew it had Zen in the pipeline.
JHVEPhoto/iStock Editorial via Getty Images
Investment thesis
Nearly 1.5 years ago, new Intel (NASDAQ:INTC) CEO Pat Gelsinger promised investors a turnaround with 5 nodes in 4 years to catch up. While the first one has launched, the four others are quite back-end loaded, with the second only arriving by the end of 2023 in what some are speculating will be quite a low volume.
As such, most of the progress Intel is making to leapfrog its competitors such as AMD (AMD) and TSMC (TSM) is still occurring mostly internally. Nevertheless, the R&D that is ongoing – and still will be in 2023 – will determine whether Intel will be able to execute on its fast-paced roadmap in 2024. This means that as Intel progresses further along the timeline of its roadmap, the stakes are increasing. As a reminder, Intel bulls have had to endure many, many comments from bears about Intel’s appalling execution track record, yet as of currently it is still on track.
Although the stock price hasn’t decreased much further, it has remained fairly stable since I recommended the shares earlier this fall due to the shares potentially bottoming.
Reviewing 2022
One can be quite brief in the review of 2022. Intel didn’t launch its major Sapphire Rapids Xeon or Ponte Vecchio GPU, neither did it launch the highly anticipated robotaxi business. Intel did ramp the successful Alder Lake CPUs and launched Raptor Lake this fall to overtake AMD in client core count (and has won at least one best-of-2022 award). Intel’s foray into discrete GPU can be called mediocre, which is what both Alchemist and Arctic Sound-M are. As consolation, there was the Blockscale bitcoin mining chip, as well as the passing of the CHIPS “fab subsidies” Act.
Behind the scenes, though, Intel continued its hiring spree, continued to execute its process roadmap “on or ahead of schedule”, and scored billions of dollars’ worth of foundry deals.
Financially, there was of course the mid-Q2 collapse, although this has since been replicated by others semiconductor giants such as Micron (MU), Nvidia (NVDA) and AMD (AMD).
Looking ahead to 2023
So with no or hardly any progress visible in 2023, the second year of Pat Gelsinger’s tenure, this brings us to 2023. Alas, it appears that most of the progress in 2023 will likewise remain only visible behind the scenes.
PC
On the client side, it now seems certain that Meteor Lake is a late instead of early 2023 product, as prior CEO Bob Swan once promised. Note that on the data center side, Intel has talked about it reshuffled roadmap as a “simplification”, so perhaps Intel simply decided that it didn’t make sense to launch Meteor Lake (14th Gen) pretty much simultaneously with Raptor Lake (13th Gen).
Nevertheless, as discussed earlier this year, there may be other reasons for the delay. The first possibility is that the Intel 4 progress perhaps hasn’t continued as quickly, although Intel recently said Intel 4 is now “manufacturing-ready” (technically this is on schedule, although it is in the latter part of the H2’22 six-month window).
Another, more likely, reason is the delay of TSMC’s 3nm process, as discussed a year ago. One piece of evidence for this hypothesis are the various trips Pat Gelsinger made to Taiwan, although this might be reversing cause and effect. The main evidence is the rumor that Meteor Lake will only have 128EUs (execution units), despite that there is unambiguous evidence from Intel itself that Meteor Lake was supposed to double the EU count (to 192), with another doubling expected for Arrow Lake in 2024.
As discussed earlier this year, the most logical explanation for this downgrade is that Meteor Lake was once supposed to use TSMC’s N3 process (although never confirmed), but Intel confirmed this summer that it will use N5. Since N5 has around 1.5x lower density than N3, this could explain the 1.5x downgraded specs. In addition, this change may also have contributed or have been directly responsible for its delay to late 2023. Nevertheless, intel has directly denied this hypothesis, saying that the node for GPU in Meteor Lake hasn’t changed. Of course, this may just be PR damage control.
To make things worse, Intel has inserted a Raptor Lake Refresh for 2023, at least for the desktop. Again, there could be various explanations. First, although Intel has claimed a 20% performance improvement for the Intel 4 process, it may nevertheless be difficult to reach the ultra-high frequencies that Intel is delivering with Raptor Lake (up to 6GHz out of the box). Secondly, there have/had been some rumors that Intel would use TSMC’s N3 process for at least some Arrow Lake CPU chiplets, so perhaps there may have been some roadmap changes there as well.
In any case, despite the barrage of new nodes Intel promised, somehow it still had to insert a Refresh into the roadmap.
Hence, current speculation is that Meteor Lake will only see a limited launch/ramp by the end of 2023, despite that Intel announced recently that the Intel 4 is “manufacturing-ready”.
To be clear, this means that the node is ready for HVM (high-volume manufacturing), not that it is going into risk production as some have suggested. (Intel doesn’t disclose risk production milestones, but for TSMC risk production occurs a year before HVM.) In the case of Intel 4, it seems that the node is ready, but there are apparently no products ready to actually use this process.
Data center
On the data center, 2023 should serve to catch up to the roadmap given the Sapphire Rapids delays. Nevertheless, if Intel fully intends to catch up, then it will be very busy with Sapphire Rapids launch in January, Emerald Rapids in the second half, and Granite Rapids in the first half of 2024.
Further, the latest information suggests that the high-end Sierra Rapids-AP has been cancelled (a leak someone sent me only contained Sierra Forest-SP). The lower-end Sierra Rapids-SP seems to have roughly the same core count as Granite Rapids, despite that the whole philosophy of the E-core roadmap was that Intel could insert a lot more cores given that it is a much smaller architecture.
While this suggests Sierra Forest might not take a meaningful performance leadership position, it should nevertheless be a rather small (and hence relatively cheap and low-power) CPU, competing on pricing and performance per watt.
Overall, my recent takeaway was that investors should be waiting for 2025 for signs of true data center leadership. To that end, Intel recently confirmed that it does not have any 20A Xeons in development, and instead intends to go straight to 18A, the fifth node on the roadmap that will bring process leadership back to Intel.
Other stuff
CFO Dave Zinsner recently said investors could expect more foundry announcements in 2023. The Tower acquisition should also close in Q1.
Financially, Intel isn’t expecting a quick recovery, with further downside going from Q4 to Q1. Note that Intel originally forecasted just a two-quarter inventory correction, making Q2-3 the bottom. Although the reorganization is offsetting some of this weakness, for the stock these near-term financial trends imply – as I had forecast after the Q3 results – that there could be more downside. In any case, there certainly aren’t any catalysts for upside on the near horizon, except if there would be a recovery in demand later in 2023.
Investor takeaway
For the most part, the progress that Intel should be making in 2023 will not yet be visible on the outside. In the past, I have likened Intel currently to AMD around 2013-2016, when it was working internally on its new Zen architecture that would ultimately bring AMD to where it is today. Similarly, by the end of 2023 Intel should be on the verge of starting commercial production of its highly anticipated 20A node, quickly followed by 18A later in 2024.
So in that regard, 2023 will be a crucial year as Intel enters the final stretches of its 20A development, which started as early as 2017. Note that it was in those same final stretches that 7nm (renamed to Intel 4) encountered its fatal yield issues. That only emphasizes the importance of this phase even more, as it in this phase that Intel starts to really ramp up its efforts to improve the yield to make commercial, high-volume production reality, in this case of the node that will re-establish Intel as the definite semiconductor giant. As such, although it may not be visible in products yet, the next year will crucially determine whether Intel will sink or swim going forward.
Disclosure: I/we have a beneficial long position in the shares of INTC either through stock ownership, options, or other derivatives. I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.