Upgrading Intel Stock: Torrid Tech Turnaround (NASDAQ:INTC)
A few days ago, Nvidia (NASDAQ:NVDA) disappointed investors by reporting awful automotive results – for years touted to be its next growth segment. Nevertheless, both analysts and the market didn’t even blink. By contrast, the next day Intel (INTC) raised its guidance at Investor Meeting 2022 – on track for its seventh consecutive record year – and this was greeted with the stock dropping to a 52-week low. Overall, Intel Investor Meeting 2022 was Pat Gelsinger’s latest, but underappreciated, master act.
The market clearly isn’t measuring all companies with the same yardstick. However, as I will discuss, proof is already mounting that Intel’s comeback remains on track and will be successful. As Pat Gelsinger said, the turnaround train is leaving the station. Hence, I am upgrading Intel and would suggest investors to take advantage of the dip to get on board.
1. Intel’s fundamentals are very strong
Although I will discuss the tech a bit more below, skipping right to the financials, the first thing to observe is that Intel operates in strong and mostly growing businesses.
First, PC has grown a lot in the last two years. Although this growth is expected to moderate going forward (obviously, the bears will expect a decline to pre-COVID-19 levels), Intel nevertheless has a strong roadmap (which starts with Alder Lake this year) to regain some share and improve ASPs. Overall, a solid business that has provided Intel with scale for decades.
Secondly, Intel’s demise in the data center has been greatly exaggerated. Intel repeated its January statement that it shipped more severs in December than AMD in all of 2021. Overall, given that Intel is a bit further behind in its roadmap (compared to its PC roadmap) and that AMD’s share in data center is lower than in the PC, surely AMD will continue to grab some more share over time. Nevertheless, data center is a strong growth market, so Intel expects at least mid-single growth going forward, and accelerating to double digit further out. Simply put, the market is growing faster than whatever market share Intel may be losing.
As an exercise for the reader, assume that the TAM is going to double by 2026 (which is Intel’s forecast) and that AMD currently has 10% market share, then calculate how much market share Intel has to lose in order not to be able to grow at an average of high (or even mid) single digit growth.
Thirdly, Intel has built an incredible moat in the 5G network, which is a double digit growth segment. Combined with IoTG, the newly formed NEX (network and edge) group is already nearly $10B in size and as mentioned has a double digit outlook over time.
Overall, the sum of these businesses implies a continued mid single-digit growth over time, perhaps even up to high single digit.
2. Entering strong growth businesses
For those who read the tweet thread above, one of the main points that Intel made at Investor Meeting was that it has added three strong growth businesses. Although Intel didn’t explicitly say it as such, common sense indicates that all three of these businesses could become 11 figures in size over the next five to ten years, meaning they could add a combined $30B (or more) to Intel’s revenue, up from about $3B currently.
Or as I tweeted, this is akin to three AMDs/Nvidias emerging inside Intel out of nowhere. Combined, these will accelerate Intel’s growth to double digits as they scale in size, although this will still take a few years. Hence, this is why Intel is expecting growth to accelerate to double digits by the end of its five year investment plan.
First, Intel has been investing in a portfolio of high-end discrete GPU products for gaming and AI since late 2017 – when Raja Koduri joined from AMD. Obviously, these investments have taken several years, but Arc is launching over the next few quarters from laptops to desktops to workstations, while Ponte Vecchio is launching in the second half of the year. Additionally, investors may also have seen that Intel is also starting to ship its Bitcoin ASIC this year. Although Intel didn’t provide more granular details, Raja Koduri put forward the goal for (nearly) $10B revenue by 2026. This means Intel may already surpass AMD in the next several years despite being a brand new entrant in this space.
In gaming, Intel is leveraging its existing position in CPUs to cross-sell its new Arc GPU as well, resulting in over 50 design wins. In the data center, Ponte Vecchio is an unquestioned leadership product right out of the gates, with Intel showing several benchmarks with 1.7x to 2.6x higher performance than Nvidia’s A100. Essentially, Ponte Vecchio is a 5nm-class GPU, and as such will remain competitive (on par) as Nvidia launches its next-gen H100 around the same time. Intel will finally provide choice in the AI market where Nvidia has been uncontested for years. Anyone who has seen Nvidia’s obscene ~68% gross margins (on par with Intel’s all-time highest margins despite Nvidia not even having its own fabs) knows that customers will take Intel seriously to drive competition for their own benefit.
Secondly, Mobileye is finally starting to bring its higher-value products to market in the next few years, moving up from ADAS to autonomous driving for both consumers and commercial applications such as robotaxis. Given the pending IPO, Mobileye didn’t provide any revenue targets, and obviously, there will likely be quite some errors bars, given the uncertainty of the robotaxi ramp. So to be fair, it isn’t certain if Mobileye could reach an 11-figure size by 2026 yet.
Thirdly, Intel Foundry Services. Intel had already disclosed Qualcomm (QCOM), and further added Cisco (CSCO) to its list of key customers. Intel said it was also engaging with several other “whale customers”, and those might be announced later this year. Intel said that customers would have to decide in 2022 to commit to 18A for a ramp in 2025. Since 18A is IFS’ first leading edge node, this means IFS will gain a lot more clarity in size. Intel further said it was “almost overwhelmed” by the customer reaction to the Tower acquisition, which will add trailing edge/specialty foundry capabilities as well as critical knowhow in running a foundry business to Intel.
In terms of revenue, Intel expects modest growth from its current ~$800M size over time. In essence, IFS’ ability to scale fast will depend on its ability to capture and expand in wallet size/share among its “whale customers”. Given that Intel is aiming to recapture process leadership just as IFS is starting to scale in 2025, this is strong value proposition. Nevertheless, although IFS could clearly become an a 11-figure business over time, as I have tweeted its judging IFS by its 2026 results akin to judging AMD’s EPYC by its 2018 results.
Overall, by definition, these are three disruptive businesses and hence there are many errors bars, although investors will likely appreciate Raja Koduri’s $10B target. Nevertheless, I would argue that everything begins and ends with leadership technology, and that is exactly what Intel is bringing to the market in each of these three businesses.
1. Leadership process technology
Investors remain skeptical about its ability to regain process leadership. However, Intel provided a large list of proof points to instill confidence that it will indeed be able to accomplish its goal. As a reminder, since process technology is foundational, obviously a return to process leadership will once again position and put Intel forward as the definitive force that anyone has to reckon with.
Intel summarized the gist of it in the slide above. Meteor Lake will go into production in the second half of the year, and Intel has previously described it as its best lead product in 4 process nodes. Intel further announced that it is going all-in on Intel 3 in the data center. Intel 3 will leverage Intel 4 but with various enhancements. For 20A, Intel already has test wafers running in the fab, and plans the same for 18A in the second half of the year. In fact, given that 18A is announced to ramp in H2’24 in this slide, it seems Intel has pulled-in its timing, although Intel didn’t explicitly mention this. Lastly, Pat Gelsinger showed a wafer with 18A SRAM test chips.
As a reminder, what enables this cadence are several things:
- Intel is standardizing on industry-standards, such as EDA tools, and is also more closely collaborating with equipment vendors to leverage learnings from the ecosystems.
- Intel is moving forward with the latest lithography tools from ASML (ASML), compared to being two or more years behind to EUV currently, and will in fact be the first to leverage the ~$450M high-NA tool in volume production with 18A. Investors should know (be aware) that the absence of EUV at 10nm was one the key reasons for its prolonged delay (with Intel going as far as saying it wouldn’t have had as much issues if it EUV has been available for 10nm).
- Intel has modularized its process development methodology and has also planned in contingencies, both done in order to prevent issues with certain modules from impacting the rest of the process flow.
- Intel is following a Tick-Tock-like cadence where every second node will be an enhancement of the previous node. So overall, Moore’s Law remains more or less on a two-year cadence despite the “5 nodes in 4 years” target, with 4 and 20A being the only real “Ticks”.
- Intel has parallel/overlapping design teams (each working on a Tick+Tock), and Pat Gelsinger has dramatically increased the Technology Development budget to an “unlimited” budget (although quantified as $1.5B).
- The 10nm and 7nm (Intel 4) delays were due to issues with yield. While Intel was solving these defects, Intel’s Components Research team kept busily working on (researching) next-gen technologies such as RibbonFET and PowerVia, which are now being implemented in 20A.
2. Strong PC roadmap
In my Investor Meeting preview, I had stated that Intel’s previous roadmap from 2018 had been completed and executed more or less on schedule with the launch of Alder Lake. Hence, Intel needed to provide a new roadmap, which it did.
The roadmap shows client will obviously continue to use Intel’s latest process tech, as well as the latest foundry tech with N3 and beyond. Hence, at no point going forward will AMD have an advantage. As Pat said, AMD is in the rear-view mirror, and they will be so forever going forward. In addition, the PC roadmap will also use Intel’s most advantage packaging technology. This is quite significant since many had doubted if Intel’s advanced packaging would be economical for such high-volume segments. However, Intel disproved these doubts by confirming it plans to ship “hundreds of millions” of Meteor Lakes, all with Foveros-packaged chiplets.
Lastly, one reminder that Alder Lake is indeed a leadership product on every single metric, including performance per watt. In the benchmark below, Alder Lake’s power has been measured to be noticeably lower than AMD (54W vs. 63W) despite delivering vastly higher performance.
3. Catching up in data center
As mentioned above, Intel’s demise in the data center has been greatly exaggerated. The pattern I observe is that people like to talk down Intel’s data center portfolio and roadmap without this being based on any tangible data. In any case, as Pat Gelsinger said, Sapphire Rapids is “a” leadership product, but just not as much as Intel would have wanted. As shown below, real-world benchmarks prove that Sapphire Rapids can indeed deliver significant performance, with Sapphire Rapids + HBM being over 2x faster than Milan-X despite the lower core count.
That Sapphire Rapids is, in fact, a leadership product can also be seen when studying its feature set. Of the features listed below, Milan-X only has Total Memory Encryption, and Genoa (likely) only adds DDR5 and PCIe 5.0.
Moving to the future, Intel actually delivered exactly what I had predicted, although to be fair not when I had predicted: already in late 2021, I had informed investors that Intel would leverage its differentiated P/E-core strategy in the data center as well, with Sierra Forest leveraging the E-core in order to leapfrog AMD and anyone else in terms of core count. It turns out that this is indeed exactly what Intel will be doing.
However, in terms of timing, Intel is going from bad to worse, as Granite Rapids has now been pushed out by two years since 2019, and Sierra Forest has also slipped from 2023 to 2024. Okay, so let’s just try to explain Intel’s (likely) reasoning behind this. From these disclosures, it seems like Intel is not really planning to use TSMC (NYSE:TSM) at all in the data center. This means Intel has to rely on what it does have internally. Obviously, given that Intel is currently two years behind TSMC, one shouldn’t expect any miracles at the 4/3 nodes.
So, understandably, given the Intel 4 delay, Intel decided to plan a new CPU on the 7 node, which is Emerald Rapids (the equivalent of Raptor Lake in the PC). This will be a straightforward refresh on an old process node. However, since Intel is targeting an annual cadence, this means Granite Rapids is getting pushed even further out to 2024. The upside, as Intel announced, is that this allows Intel to “upgrade” Granite Rapids to the superior 3 node, which will have a higher transistor density, so it could perhaps have a (slightly) higher core count. In addition, Pat Gelsinger also indicated Granite Rapids would have a new architecture compared to the previous Granite Rapids plan.
As such, while the CPU will launch another year later, this isn’t the same Granite Rapids anymore. So, while technically it is a two-year delay, Intel might just as well have canceled Granite Rapids and given this new Xeon on Intel 3 different name altogether.
In any case, none of this really matters with regards to the core count race that the internet prefers to talk about, since this is the whole point of Sierra Forest, which in theory could have many hundreds of cores for all we know. As such, the only issue that remains is the timing, as Intel 3 isn’t going into production until the second half of 2023. Hence, as Pat Gelsinger said product leadership will go “back and fro” over the next few years.
For the regular roadmap:
- Intel takes product leadership with Sapphire Rapids (56C) over Milan-X (64C) in Q2’22
- AMD takes product leadership with Genoa (96C) over Sapphire Rapids (56C) in Q4’22
- Intel fails to overtake AMD with Emerald Rapids (64C) in H1’23
- Intel takes product leadership with Granite Rapids (112-144C?) over Genoa (96C) in H1’24
- AMD takes product leadership with Turin (196C) in H2’24
- Intel takes product leadership with Diamond Rapids over Turin in 2025
For the cloud roadmaps:
- AMD enhances product leadership with Bergamo (128C) in H1’23
- Intel takes product leadership with Sierra Forest (256C?) over Bergamo (128C) in H1’24
- AMD may or may not take product leadership with Turin Cloud (256C) in H2’24
- Intel takes product leadership with next-gen Forest in 2025
The Intel roadmap did confirm my suspicion (in light of comments by Gregory Bryant at CES) that the data center would be slower than the PC to move to the leading edge, as the roadmap suggests Intel will only move to 20A (or 18A) in 2025. Nevertheless, if Intel adheres to an annual cadence, then an H1’25 launch for Diamond Rapids wouldn’t be much later than Arrow Lake (H2’24).
Overall, the data center leadership discussion is much more nuanced than what the internet generally suggests it is. For example, for certain workloads such as crypto or AI or compression, neither Genoa nor Turin will come anywhere close to “leadership” since Intel has dedicated accelerators for these workloads in its Xeons. (Although AMD recently announced it would launch products with Xilinx AI IP in 2023.) Pat Gelsinger’s statement that Intel and AMD will leapfrog each other over the coming years seems quite likely to be true. Obviously, the whole point of the “Intel Accelerated” process technology cadence as discussed above was to move on from 4/3 as fast as possible in favor of 20A/18A.
As one last point, investors may consider that from mid-2017 to mid-2021, for four years, Intel’s roadmap in the data center had been completely stagnant. Intel wasn’t able to offer any performance increase whatsoever during that whole period. By contrast, from mid-2021 to mid-2025, Intel will have launched four CPUs, at an annual cadence, each with significant performance increases (Sapphire, Emerald, Granite, Diamond). For example, core count alone will have increased by at least 4x. So even if Intel will still be busy with catching up and battling for leadership, at least Intel now will have something new to sell to its customers each year.
Financials: Incredible Proposition
The stock market is currently heavily busy with selling off Intel due to its gross margin erosion. However, Intel is forecasting gross margin to rebound back to 54-58% by 2025/2026. In fact, Intel was being conservative, since Intel detailed at least 4 areas which each could unlock 1-2 points of margin expansion, which means in the best-case Intel could get back to its historical ~60% range, although investors should be aware that there will be one headwind for gross margin, which is the IFS business.
Things get even better when looking at revenue. Assuming the graph above is to scale, Intel is currently forecasting 2026 revenue of $125B. This implies investors will get significant revenue, gross margin, free cash flow and earnings expansion over time. In other words, the stock is being myopic about the near term without considering the long term. In numbers, Intel trades at a measly ~1.5x 2026 P/S and perhaps just 1x ~2030 P/S if Intel can continue to grow at double digits beyond 2026. (Of course, bears would scream from the rooftop about how optimistic Intel’s revenue forecast supposedly would be.)
Indeed, Pat Gelsinger made the case for a 4x investor return (“double double”): twice the earnings at twice the multiple. This means Intel may finally be vindicating my $210 price target from 2020.
A last financial topic worth highlighting is Intel’s Smart Capital strategy. Intel’s official guidance is for a 10% offset due to incentives, but this seems incredibly unrealistic, with management indeed readily admitting that it is aiming much higher. The E.U. and U.S. CHIPS acts should allow Intel to aim for as much as 20-30% subsidies. Needless to say, this vastly improves the already quite compelling economics of being an IDM. Just imagine if TSMC was giving its customers 20-30% rebates (which clearly TSMC doesn’t, as it actually hiking prices by that amount, reportedly).
Of course, the financial (and hence stock) upside detailed above is for a large part contingent on Intel executing to its technological roadmap (and also for the supposed market TAM growth to materialize). However, as discussed above, Intel is actually reasonably well progressing on its plan to catch up, and could already deliver its second-gen RibbonFET (18A) by the time TSMC has yet to ramp its first-gen N2. In fact, more and more people in the tech community are speculating that TSMC is actually having major issues.
As another reminder, something the bears haven’t understood is that it really doesn’t matter how Intel’s capex compares to TSMC, for two reasons. First, Intel’s capex is 100% leading edge, whereas TSMC’s capex also contains up to 30% trailing edge spending (although this is changing a bit given Intel’s Tower acquisition). So for example, in 2022 TSMC will likely spend less than $35B on capex, which is just negligibly more than Intel’s guidance for $30B gross capex. Secondly, even then all this really doesn’t matter, since the only thing that matters for the long-term demand is which process technology is actually being manufactured in these fabs.
In any case, Intel has provided investors with a quite comprehensive roadmap with many milestones across technology development, client, data center and foundry.
I have summarized Investor Meeting 2022 in following series of tweets (although some may be exaggerating a bit).
In summary, the simple case for Intel is that its return to process leadership will once again transform Intel into the definitive semiconductor force to be reckoned with – from CPUs to GPUs, FPGAs, IPUs, NPUs/AI. This will impact every single major semiconductor company in the market – from AMD (AMD) to Nvidia (NVDA), Qualcomm (QCOM) and TSMC.
Intel’s pitch at Investor Meeting 2022 – the culmination of Pat Gelsinger’s first year back at Intel – was that Intel is the next great growth story. The turnaround train is leaving the station at a torrid pace.
As such, there is a lot to like. Arguably, the bottom for Intel may actually already be behind and the future is getting incrementally brighter over the next few (but long) years, as Intel has now rolled out Alder Lake and is starting to roll out Sapphire Rapids in 2022 (the first time Intel is catching up, albeit briefly, in the data center since 2019). Intel already aims for double digit data center growth in 2023.
In terms of financials, obviously there will still be a lot of pressure in the next few years as Intel’s investments are geared not just to catch up, but serve to actually overtake TSMC and all of Intel’s key competitors, as well as to enable Intel’s growth targets by building out a ton of fabs. So, while Intel still has a lot of work to do, key nodes such as 20A are already getting to the stage where Intel is now running IP shuttles through its pilot lines. As a nice bonus, Intel has apparently pulled-in 18A from early 2025 to perhaps late 2024 as a further sign of confidence. (Intel has said it could still do 18A without high-NA in the worst-case.)
As a reminder, the amount of capex Intel/TSMC spend is irrelevant, the only thing that is relevant is which process nodes are running in those fabs. For example, if 2025 will mark the start of a decade of Intel industry leadership, then demand for Intel’s fabs (whether due to in-house products or IFS customers) should continue to grow even beyond the timeline laid out at Investor Meeting, which should fuel further fab investments – which may or may not be at the expense of TSMC.
Finally, with regards to the stock, long-term Intel bears such as Stacy Rasgon have questioned why one would buy the stock now, if Intel for example isn’t even expecting to be FCF positive by 2025? The simple answer is because I have no clue what the stock price will be in 2025. Maybe investors will see that Intel has by then successfully ramped 18A, and perhaps TSMC has delayed 2nm, and hence investors may already have missed a lot of INTC upside. In addition, at no point in the next half a decade is Intel expecting a revenue decline. Still, maybe investors will ignore the revenue growth and instead continue to by myopic about the 51-53% gross margin range Intel has forecast to last through 2024, which may push the stock even lower. I don’t know.
Hence, Investors should ask themselves if they are convinced by the proof points Intel has shown from technology development (18A wafer), client (Meteor Lake disaggregated chiplet CPU powered-on in 30 minutes), data center (Sierra Forest many E-core CPU to win the core count race), graphics (Ponte Vecchio over 2x faster than Nvidia A100, 4Mu Arc shipments in first three quarters of launch), foundry (already endorsed by the likes of Qualcomm and Cisco (CSCO), expects key customer announcement in 2022, Tower acquisition) and financial outlook (expects accelerating growth as new businesses such as Mobileye and graphics scale, and existing ones get more competitive).
If they are convinced by these proof points, then why not take advantage of the current stock price, which sits at the low-end of a multi-year range? As a small bonus, Intel already pays out a >3% dividend at this price. There is nothing scary about that.