RobsonPL
Investment Thesis
Already since late 2020, even before Pat Gelsinger came on board, I had expected that 2022 and 2023 would be quite rough years for Intel Corporation (NASDAQ:INTC). Intel is in the midst of its delayed 7nm node (renamed since to Intel 4), is losing revenue due to shedding the NAND business, and its data center portfolio and roadmap continues to chase from behind. This all led me to downgrade Intel in January, as traditional rating systems only concern stock expectations for perhaps a year into the future, which is much shorter than the time Intel’s turnaround would take.
However, after the investor meeting I published a more bullish article, although with the caveat that obviously these developments would still take multiple years: Upgrading Intel Stock: Torrid Tech Turnaround. Readers should also note that by then the stock had declined to the mid-40s, which was not far from its multi-year low.
Even though since then the stock has continued to decline into the 30s, in this article I will detail several key points that to me show that Intel has damaged investors’ trust: management not hitting its goals is a major red flag, especially after how many times Pat Gelsinger has said that he wanted to reintroduce accountability at Intel. Pat Gelsinger said that Intel would do what (or even 1.1x of what) it said it would do, and when it said it would do it.
Hence, at this point in time Intel still seems stuck in limbo: while Pat Gelsinger never said he could fix Intel’s problems before 2024, the fact that some problems have become even worse since he took over leaves investors with valid questions marks if there will be any improvement by the time 2024 does arrive. As Stacy Rasgon has often said, this means that Intel for now remains “cheap hope” rather than a “cheap execution machine.”
Trifecta of Intel fails
There are several moving pieces behind my downgrade. Let’s evaluate them.
Ponte Vecchio and Alchemist
Ponte Vecchio was originally meant to be Intel’s flagship GPU to retake the HPC and AI crown from Nvidia (NVDA), as well as lead 7nm product, launching in late 2021. Even after the 7nm delay, Intel still said it expected to launch Ponte Vecchio in early 2022. Clearly, this hasn’t happened.
In addition, nearly a full five years after Intel said it would develop high-end consumer GPUs, there are still none available in the market. Notably, these GPU delays have obviously caused Intel to miss the whole COVID-19 boom and shortages that resulted in very inflated GPU prices. This could have been a strong opportunity to quickly take market share as a newcomer and get praise from consumers who don’t want to buy overpriced Nvidia GPUs.
Sapphire Rapids
Sapphire Rapids was one the earliest indications that Pat Gelsinger’s turnaround would not be a completely smooth ride, as in mid-2021 Intel announced the delay of Sapphire Rapids production from Q4’21 to Q1’22. Note that in late 2020, investors were still asking Intel how it would deal with Sapphire Rapids launching to closely after Ice Lake Xeon. To be fair, for a long time it looked like Intel would be able to hold itself to this slightly revised schedule, but more recently, new reports of delays have come out.
So, in line with my commentary about management shaking investors’ trust, Intel has not chosen to respond to my inquiries for clarification. As such, this means that Intel’s data center, which has been in shambles for half a decade now, remains a complete train wreck.
As case in point, note that Intel’s data center VP has emphasized that the Sapphire delay has nothing to do with the process technology (10nm) delays, as 10nm is in high volume production now. However, this is not a solace, as 10nm had been delayed for three years. So, if Intel’s Xeon product group couldn’t get a product ready by the time Intel finally got 10nm working, then what have those engineers been doing for over three years? This only makes the Sapphire Rapids clown show look even worse.
Meteor Lake
The last nail in the coffin for me was Meteor Lake. A new rumor has disappointed me in no less than two ways.
First, the rumor said that Meteor Lake would have up to 128EUs (graphics execution units). This is an outright downgrade from Intel’s own disclosure in 2021 which indicated that Meteor Lake would have up to 192EUs (2x more than its 10nm CPUs). Ironically, this one may actually not even be Intel’s fault; due to Taiwan Semiconductor Manufacturing Company Limited’s (TSM) 3nm delay (as Intel has said as recently as at its February investor meeting that the GPU tile would be manufactured on this node), Intel may have been required to change the node from 3nm to 5nm, which in turn may have required Intel to dumb down the GPU specifications given the inferior node it now has to use.
Secondly, and most importantly, the leaked Intel document points to a Meteor Lake launch in the second half of 2023, which (just like Ponte Vecchio) is later than the early 2023 Intel had said in mid-2021 after the 7nm delay. So the puzzle here is that since mid-2021 Intel has always maintained that its Intel 4 node would be on track for manufacturing in the second half of 2022 (in time for a Meteor Lake launch in the first half of 2023), yet then why will Meteor Lake only launch in the second half of 2023? Although it does take several months from manufacturing to launch of a products, normally this time lag is about half a year, not the full year that this timeline is indicating.
Hence, it seems that perhaps the TSMC 3nm delay has not only caused the GPU specs to get decreased, but by having to port the GPU tile to another node, this may have caused the whole CPU to get delayed. In any case this is a clown show and Intel is responsible for this mess.
(For a more technical argument, perhaps in order to save time and money the Intel 4 process node does not have a higher density transistor library, which is what GPUs are built on. In other words, since Intel expected to use TSMC for the GPU tile, Intel didn’t even bother to develop the library that it would use for a GPU chiplet, so Intel literally can’t manufacture the Meteor Lake GPU itself and hence had to rely on TSMC, who then turned out to have its own issues at 3nm, which although not officially confirmed is what I above speculated may be the reason for this apparent Meteor Lake delay.)
Bonus: Diamond Rapids
A last mention may be Diamond Rapids. Back when Bob Swan was still CEO, Intel had intended to launch this CPU in 2024. Intel at the time thought that this would be its definite return-to-leadership data center CPU that would be miles ahead of anything from either Arm or AMD (AMD), which indicates it would be manufactured on 5nm (now renamed to 20A).
However, the roadmap as presented in February has put Granite Rapids on Intel 3 in 2024, which hence pushed Diamond Rapids out to 2025. Now, investors may remember that Intel has pulled-in its 18A node to the second half of 2024, so perhaps Intel could take a similar approach with Diamond as it is doing with Granite, which is to skip the initial Intel 4 node in order to launch a superior product, albeit at a bit later timeline. So here, Intel may have pushed Diamond Rapids from 20A to 18A, and the timeline from H2’24 to H1’25.
Nevertheless, with the Sapphire Rapids delay, the question still remains whether Intel will be able to execute on this roadmap. Since Sapphire is now a H2’22 product at best (and perhaps even H1’23 since as mentioned Intel didn’t deny that rumor), then it would have to launch Emerald Rapids quite closely after Sapphire in 2023 in order to then launch Granite in H1’24.
Further thoughts
The main purpose of this article is to update investors on the product execution currently at Intel. However, how exactly investors decide to interpret these facts is open for some interpretation, which should be mentioned.
On one hand, Pat Gelsinger has literally said himself that the 2022-2023 roadmap is what it is, but expected unquestioned leadership in 2024-2025. This means he must have seen the roadmap, the terrible execution, and wasn’t even going to try to bother to fix the 2022-2023 roadmap. In other words, Pat Gelsinger wasn’t willing to take accountability for this pre-2024 roadmap, so is it fair for investors do so anyway?
On the other hand, as I have pointed out many times, Intel has hired over 12k net new employees since Pat Gelsinger became CEO. This incredible pace of hiring engineers should start to bear results at some point, shouldn’t it? Perhaps most of newly hired engineers have all been assigned on products that are targeting the 2024-2025 timeframe to deliver on Gelsinger’s promise of unquestioned leadership, but still these many delays will leave investors with a sour taste.
So, in summary, investors have to decide for themselves if either they see the current 2022-2023 execution as a sign that Intel was terribly broken and hence as proof that Pat Gelsinger was needed to fix Intel, which he is (and the 12k new engineers are) working on for 2024-2025. Or rather as a proof that despite Pat Gelsinger and the >12k engineers Intel has added, Intel has not been fixed.
As a reminder, the evidence to the downside was presented above, while the evidence to the upside is that Intel’s “5 nodes in 4 years” (which has actually become “4 nodes in 2-3 years” depending on how you count) remain on track.
Valuation
With a market cap around $150B, a P/E close to 10 and a 4% dividend, Intel is slowly becoming cheaper than dirt. (Since it is Pat Gelsinger’s goals to double Intel’s earnings, the long-term forward P/E may even be closer to 5.) Hence, in this section I reiterate that the downgrade in this article is solely based on management not doing what it said it would do, which as discussed was balanced out by Pat Gelsinger, who basically said he didn’t even want to take responsibility for the 2022-2023 roadmap anyway. This led to the conclusion that it is up to investors themselves if either they hold Pat Gelsinger accountable, or still believe in the turnaround in 2024+.
In the latter case this would be an opportune time to buy the dip. For two more arguments why now may actually be the time to buy the dip are: (1) that the fundamental process tech remains on track despite the shaky product execution; and (2) although the statement that Intel should ship a 20A product first (and on schedule) before Intel becomes investible, the issue that by that time the stock may already have appreciated meaningfully. The best time to invest is before others join the bandwagon.
Investor Takeaway
Most of my critics have told me that the issue is that I always believe Intel’s PowerPoints, but in reality the products always get delayed. I must admit that those people continue to be right, while I continue to be wrong. Quite literally all of Intel’s 2022-2023 products seem to be suffering ugly delays and setbacks.
In my downgrade in January I said something along the lines of “let Intel ship a 20A product first,” and given the continued reports of delays we’ve received about Intel so far this year, it seems that may indeed have been the right mentality, although the real reason for the Intel stock performance is most likely the overall macro environment.
Ironically, though, at this point it seems that Intel’s process technology development group has actually been executing the best among all of Intel’s groups, as this group hasn’t encountered any further delays for the past two full years now. Since process technology remains foundational for the products Intel sells, this improved process execution (with Intel’s goal to regain leadership in 2025 still on track, especially given TSMC’s own delays) means that once Intel’s product groups start waking up, Intel should actually be fine.
As such, and given that Pat Gelsinger himself never wanted to take responsibility for the 2022-2023 roadmap, in addition to the stock being at a multi-year low, I am unwilling to downgrade Intel two notches to “sell.” I do remain “neutral” until I see signs that Intel is able to do what it said it would do when it said it would do it. In other words, while Intel is still down, the fundamental turnaround thesis of regaining process tech leadership remains unshaken.
To me, the most crucial milestone is to launch Granite Rapids and Sierra Forest in H1’24 and then their 18A successors in H1’25. If the process tech remains on track, there should be no fundamental reason why this couldn’t happen, but Sapphire Rapids (and to some extent Meteor Lake) has put questions marks behind this thesis, which was the reason for this update.
Be the first to comment