Intel Stock: The Bad Can Get Better (NASDAQ:INTC)

Investment Thesis

As discussed extensively in previous articles, Intel Corporation (NASDAQ:INTC) earnings are down for a variety of reasons, both macro- and micro-economic. This has resulted in a perfect bear storm.

In the near-term, as this article previews Intel’s Q3 report, due Thursday, October 27, the “big” question is if Intel’s forecast for a recovery in Q4 will materialize. On the one hand, Intel claimed it has been vastly under-shipping the market due to the inventory correction. On the other hand, the results and guidance of other companies such as Micron (MU) aren’t exactly inspiring either.

On the balance, my view remains that Intel has become undervalued under $30 since investors are willing to pay only for immediate profits, despite that there is much leverage built into Intel’s multi-year plan (which is impossible to predict exactly and therefore not present in any analyst “gospel” estimates). This means Intel’s earnings have the potential to multiply over time, and with that the stock.

Earnings preview

The model below isn’t so much a model, as it is just extrapolating the Q2 results going forward. When I saw that this resulted in just below $65B revenue for the year (the low-end of the guidance), I was satisfied with the exercise.

	DCAI	NEX	AXG	IFS	MBLY	CCG	CCG Adj	Sum
22Q1	6,034	2,213	219	283	394	8600	694	18437
22Q2	4649	2333	186	122	460	7040	625	15415
22Q3	4649	2333	186	122	460	7040	625	15415
22Q4	4649	2333	186	122	460	7040	625	15415
	19981	9212	777	649	1774	29720	2569	64682

Click to enlarge

Looking over these results, if this is indeed what Intel will report, then this would be just horrible:

• DCAI reported its lowest revenue since (at least) Q1’21 (given the new segment reporting, results from earlier quarters aren’t available). For reference, the highest DCAI revenue yet was in Q4’21, just below $6.5B. Although there have been periods of weakness in the past (2019 and early 2021), those periods (“cloud digestion”) were after periods of high growth.
• CCG (excluding the adjacencies) reported its worst result since Q2’16.

While the other segments are a bit more resilient, they are still small, and even if they improve from the Q2 results, aren’t going to move the needle:

• NEX is the highlight as the one not-very-small segment that seems to continue to grow to new all-time highs, confirmed by Intel’s comments.
• The expected AXG ramp is delayed due to the delays in the ramp of Arc and Ponte Vecchio.
• IFS, as expected, remains in the buildout phase.
• Mobileye had a great Q2 and will likely continue to grow, but will likely need its autonomous (L4) product and its robotaxi business to become a material business for Intel’s overall financials.

This is how these results would fit in the long-term trends:

	CCG	DCAI	NEX	AXG	MBLY	IFS
2019	37938	21696	6829	606	879	461	68409
2020	40535	23413	7132	651	967	715	73413
2021	41067	22693	7976	774	1386	786	74682
2022	32289	19981	9212	777	1774	649	64682

Click to enlarge

Overall, CCG will likely see a ~$10B or 25% hit, while DCAI would lose closer to 10%, for a large part attributable to market share losses. Revenue would drop firmly below 2019 levels.

For some further discussion, in ordinary times it just seems nearly impossible that Intel would report three straight quarters of multi-year low results in both DCAI and CCG, but of course these aren’t ordinary times. Since Pat Gelsinger said in early September that the results are trending towards the lower end of the guidance (but still within the guidance), there is no reason to model in any sort of recovery.

Nevertheless, one thing investors should be aware of is that Intel was quite firm in Q2 on noting that part of the results was a “once in a decade” inventory correction. So, at the time, Intel did actually guide to some recovery in Q4. Given how the macro environment hasn’t really improved since July, even if demand continues to soften, at least there should be some less inventory burn, both of which might to first approximation cancel out each other.

Hence, since (1) the “model” above indicates that Intel hardly needs any recovery to report full-year results within its Q2 guidance range, as well as (2) given Pat Gelsinger’s comment about being in the guidance range, and (3) given how bleak the Q2 results were to begin with, contrary to some investors, I do not necessarily expect a further downside guidance revision (although perhaps the top-end of the range may be reduced).

As case in point, the PC sale numbers that have been reported for Q3 certainly were not post-apocalyptic, and within pre-pandemic levels. Of course, there may have been some market share and average selling price changes since then.

On the other hand, though, a revenue shrink of “just” around 25% would actually be quite strong compared to what we’ve seen from AMD and Nvidia recently. In that sense the glass could be seen as either half full or half empty.

Lastly, for reference, my Q3 “prediction” (extrapolation) is in-line with the Street estimate, while for Q4 the estimates are modelling in a ~$1B sequential improvement.

Investment Thesis: Revisited

Given the discussion above, it can be concluded that revenue has indeed likely bottomed. If that is indeed the case, then Intel may have reached the financial bottom, and hence also the stock bottom since the stock has simply followed the earnings estimates. This is no guarantee for a quick recovery, but at least the Q4 guidance will either confirm or disprove this thesis by next week already.

Just given how weak the Q2 results were, the upside at this point seems a lot larger than any potential further downside, especially since none of this factors in any of the three “new” growth businesses/segments ((AXG, IFS, MBLY)), which currently still represent a rounding error in the overall results.

Business updates

There have been several moving pieces over the last quarter, which warrants a review (see below for a summary).

1. Raptor Lake and Zen 4 launch

At the time of writing, Raptor Lake has been announced but not yet launched, just a few weeks after Advanced Micro Devices (AMD) launched its own Zen 4 CPUs. This means there are only reviews yet of the AMD part, although as usual there have been plenty of leaks and Intel has disclosed its own benchmark measurements. Contrary to some investors’ (outdated) opinions, Intel has been very competitive again since the Alder Lake launch, and this remains the case.

Background

As a reminder, Intel launched Alder Lake less than a year ago with the game-changing feature of adopting big. Little (hybrid architecture), which is common in smartphones. However, Intel is using this feature to be able to deliver both single-threaded and multi-threaded leadership. The P-core (performance core) team is charged with developing the highest IPC (instructions per clock cycle) and highest clocking architecture, while the E-core team is focused on delivering the highest performance per mm2 of silicon. Note that contrary to smartphone little cores, the Intel E-cores are actually much closer to smartphone big cores; Intel has compared the Alder Lake E-core to its 14nm Skylake CPUs in terms of performance, but at just a fraction of the power. (Intel should perhaps have called them large X-core and small S-core.)

Raptor Lake

The top-end Alder Lake was equipped with an 8+8 config of P+E cores, but Raptor Lake doubles the E core count to 16. Intel has also tuned the process and circuits to improve the frequencies, which have also improved slightly. Overall, Intel is claiming low double digit single-threaded and around 40% multi-threaded improvement.

Zen 4, comparison and verdict

Although this is quite substantial, AMD’s Zen 4 claimed a similar (or even slightly higher) multi-threaded improvement, and a nearly 30% single-threaded improvement. This was achieved due to a combination of IPC and frequency.

Hence, the general expectation is that Zen 4 (Ryzen 7000) and Raptor Lake are closely matched in terms of performance, as the multi-threaded comparison isn’t changing much while Zen 4 is catching up a bit in terms of single-threaded performance. However, pricing is heavily in favor of Intel, especially considering that the Intel platform still supports DDR4 memory, unlike Zen 4. This means both cheaper motherboards and RAM.

Hence, despite being at a process node disadvantage, Intel will likely actually end up slightly edging out AMD in terms of performance per dollar.

So, overall, Intel’s competitive position in the PC market remains solid. Of course, though, Intel still has a substantial market share advantage due to its incumbency position; it could perhaps be argued that simply being roughly on par might not be enough for Intel in order not to be vulnerable to continued market share erosion over time.

Power consumption

A specific remark should go to power consumption. Last generation, Intel received quite a lot of criticism with Alder Lake for its supposed lack of efficiency. However, power consumption scales linearly to quadratically at best and cubically at worst with frequency. This means that the last few percentages of performance are responsible for the vast majority of the total power consumption. Hence, in the pursuit of the performance crown, Intel sacrificed power consumption, even though at lower power consumption the performance per watt improves substantially.

As case in point, Intel has claimed that Raptor Lake at 65W delivers the same performance as Alder Lake at 250W. This claim has already been verified at least at 80W.

This generation, with Zen 4 and the new AM5 platform (as the AM4 platform was capped at around 140W), AMD had opted to follow into Intel’s footsteps. As mentioned above, the majority of the Zen 4 performance improvement comes courtesy of higher frequency, which is inefficient.

Tests have indeed shown that this has offset pretty much all of the supposed benefit of the 5nm process technology. In the worst-case, the performance per watt (energy per instruction) is pretty much the same as Zen 3. Even worse, for gaming the fps per watt (energy per frame) has deteriorated substantially.

Surely AMD bulls will point out the eco modus in Zen 4, but the point remains that out of the box, both Intel and AMD are now sacrificing power (efficiency) in pursuit of the highest performance for their flagship CPUs. In other words, this shows that the efficiency of an architecture/CPU is not something inherent, but obviously depends on its specific operating condition; both Intel and AMD CPUs can be both efficient or inefficient.

2. Arc debut

As discussed in previous article, Intel is – to no one’s surprise given the specifications and time to market – playing the performance per dollar game in the mid-range segment in order to capture market share as it enter the GPU market.

Overall, this launch is a substantial milestone for what should be a new growth segment for the next decade.

For a more technical discussion (feel free to skip), the major issue with Arc is that the GPUs do not live up to their theoretical capabilities (as measured in TFLOPS), likely due to immature drivers and/or architectural bottlenecks. In principle, the A770 should or could have a performance between the, both much more expensive, GTX 3060 Ti and 3070, while in reality it falls short of the GTX 3060 Ti and actually rather competes against the 3060, where the . The reason this is in issue is because AMD still exists and actually already competes favorably against Nvidia (NVDA) based on performance per dollar, which further reduces any performance per dollar advantage that was left in the Nvidia comparison. Hence, reviewers tends to recommend AMD at best.

Nevertheless, in a recent discussion, although not really surprising, Intel said most of its engineers are working on Battlemage now, as well as that Intel isn’t planning on repeating the mistakes it made with Alchemist. Intel remains committed to its roadmap.

In particular, while it is currently leveraging Taiwan Semiconductor (TSM) (“TSMC”), given Intel’s highly outspoken target of regaining process leadership, a Celestial GPU on 18A in 2025 could in principle be highly compelling, given that especially GPUs can very effectively use additional transistors that are available with a process node advantage. As Pat Gelsinger said, even a mediocre architecture with a leading transistor will still a leadership product.

Although, as last technical remark, one criticism that I have had about Intel’s architectures over the years is that Intel often tends to use design libraries with relatively low density, far below the theoretical maximum transistor density of the node. This means that even if Intel is one node ahead, if it then uses a library that is 2x less dense than what AMD and Nvidia are using, then this offsets the theoretical advantage. Indeed, a comparison I saw passing on Twitter (but couldn’t find) compared the Arc GPUs against AMD and Nvidia in terms of transistor count and silicon area vs. performance. In simple terms, the 400mm2 Arc GPU has trouble competing against much smaller chips from the competition, which also refers back to the point above that Arc doesn’t even deliver the performance (TFLOPS) that it theoretically has in the first place.

3. Mobileye IPO

Although the IPO is receiving most attention, the real news is that Mobileye Global (MBLY) is nearing the initial launch of its robotaxi service, although no formal announcement has been made yet. A few years ago, the CEO had stated early 2022 launch, so there is a bit of a delay. Since Mobileye had said it is waiting for regulatory approval to launch its driverless service, it seems the delay is due to regulation and not technology.

A few years ago, I had put out a moonshot goal for Mobileye revenue to reach $4B to $5B by 2023. Although for this to happen to growth rate would have to accelerate to more than triple digits, the premise of the robotaxi service opening a large new growth market for Intel remains. In the bull case, robotaxi takes over all ride-hailing, becoming larger (as well as more profitable given the lack of driver opex) than the likes of Uber (UBER).

Regarding the IPO itself, it has been ridiculed by investors for the valuation collapse over the last year. However, I wonder what investors are complaining about, as this provides a great investment opportunity. Given how disruptive robotaxis will be (or at least can be in theory), there is a likelihood Mobileye revenue could surpass its IPO market cap over the coming decade.

4. IFS update

Capacity corridors and system foundry

A new tidbit of information came out about Intel Foundry Services. Intel is establishing dedicated capacity corridors for its foundry customers. No one, not even Intel, will be able to touch these corridors. Intel said the MediaTek corridor is a few thousand wafers per week. Depending on the wafer price assumed, a few thousands wafers per week at a few thousand dollars each starts to approach the $1B annual revenue mark, from just one customer. Note that MediaTek has so far only signed a deal for Intel 16, so the opportunity for land and expand remains strong.

Overall, a few billion revenue from about half a dozen large customers (Intel is engaged with 6 of the top 10, according to Pat Gelsinger in July) is exactly the recipe to build a large foundry business.

In addition, Intel has propagated a new marketing terms for all its foundry initiatives together (including things like packaging and its ecosystem investments): systems foundry. Together with the capacity corridors, this shows that Intel is taking the steps to become at least on equal to footing to pure play foundries such as TSMC. For example, one often heard criticism remains that some customers supposedly would avoid IFS since Intel would be competing against its own foundry customers.

Twitter.

However, this argument doesn’t make a lot of sense since for the competitiveness of some product, the only thing that matters is the transistor, not who made the transistor. As case in point, although it could be interpretated otherwise since Apple (AAPL) did eventually switch to TSMC, Apple had been a Samsung foundry customer for quite a few years. That was, effectively, as if AMD would use IFS to manufacture its CPUs.

In addition, since IFS was set up as a separate business, it actually does not compete against any of its customers, just like TSMC. And since Intel has also addressed some of the other criticisms (such as lack of ecosystem and EDA support), the decision for fabless customers will simply come down to things like process leadership, packaging and pricing – all of which Intel is well-positioned to compete on going forward.

TSMC already pressured

TSMC seems to be feeling the competitive pressure from IFS already.

Twitter.

TSMC indicated it is challenged by “overseas fab expansions”. While certainly Intel isn’t the only “overseas” foundry, nor the only one busy with fab expansions, it is the most logical candidate, especially since, indeed, Intel’s foundry business is still in the phase of building the fabs.

For comparison, it took something like 3-4 years after the launch of Ryzen/Epyc for Intel to acknowledge that the competitive environment was intensifying.

Gross margin de-risked

When Intel announced the foundry business, one concern was that only few companies can really attain high gross margins, as even market leader TSMC achieved “only” around ~50%, so it was questioned if Intel could make this a profitable business.

However, in the most recent quarter, TSMC gross margin reached a record ~60%, driven by price increases in the wake of the shortages. The operating margin was around 50%. Hence, even assuming that Intel is willing to sacrifice a bit of margins to lure customers, both TSMC metrics are much higher than what Intel has reported most recently.

Financial engineering: Internal foundry model and Brookfield

I recently discussed Intel’s new internal foundry model, which warrants some follow-up discussion as there seems some misunderstanding.

First, as it relates to the general topic of financial engineering, one comment said that I had misunderstood the previous piece of Intel financial engineering, namely the Brookfield deal, which I summarized as Intel building two fabs for the price of one, but also only getting profits for one fab. The commenter said that Brookfield’s maximum profit is purportedly capped. No source was provided, though, and Intel never mentioned anything like this, so my opinion stands.

Secondly, the misunderstanding about the internal foundry model seems to be that people think that it is only now that Intel is starting to apply IP protection/firewalls and a separate P&L for its foundry business.

Twitter.

This is simply factually wrong. Intel Foundry Services was conceived from day one as a separate entity with its own P&L. This is obviously mandatory to be a foundry in the first place. Investors may know that a common criticism is that Intel is (or would be) competing with its own foundry customers, so obviously IFS cannot just be allowed to give its customers’ IP to Intel.

Hence, this is the wrong way to look at things. Imagine you’re Apple or Nvidia, for example, and given the standards you as a company adhere to, as well as what your shareholders demand anyway, you must build the most competitive silicon out there. Especially in the case of Nvidia, if you can build a better GPU, you can charge higher prices and maintain a higher market share.

But since you’re a fabless company, you can freely go shopping around looking for foundries that make the best transistors. In the case of Apple or Nvidia, or even AMD, that means comparing the offerings of the three leading edge foundries (i.e., Intel, TMSC, and Samsung), and then picking based on features likes pricing and power/performance. Of course, if for some reason you have a prejudice against one company, fine, but then you risk not making the best silicon possible.

This has been true from day one for IFS. What the internal foundry model changes, though, is that now Intel itself also becomes an IFS customer. I suppose this entails that for example Intel’s client and data center groups will get the same wafer pricing as other foundry customers. In this way, Intel design groups will get apples-to-apples comparisons against pricing of other foundries, and hence Intel’s process development teams are challenged with providing and getting a lower cost than foundries.

While this sounds like a neat exercise and obviously will require some organizational changes to do this, my argument was that it doesn’t change any fundamental economics. It is just accounting. Even if IFS charges Intel’s client group with some arbitrary wafer price, this is just internal money that gets shuffled around inside Intel.

Twitter.

Dividend sustainability

Some recent articles argued that Intel’s dividend is unsustainable given the recent trends such as FCF. However, investors who analyze Intel purely based on its financials, as argued in a recent article, are probably the reason Intel is currently as cheap as it is. In other words, the argument is that investing in Intel purely based on the numbers is misleading.

In summary, Intel’s FCF has been negative in 2022 due to large fab investments in preparation for a return to growth in the coming years, while (coincidentally) revenue is also excessively underperforming due to unsustainable inventory burn and lowered demand. Nevertheless, in the case that there would be no growth, the fab investments would obviously decline significantly. In addition, there is further long-term (gross margin) upside as the investments to regain process leadership progress. Lastly, many of these articles seem to rely on estimates as if these are gospel. The reality is that given all the variables involved (for example, what if the robotaxi, GPU or foundry business becomes successful?), no one can reasonably predict Intel’s mid-term earnings and FCF (although obviously Intel as discussed at investor meeting does have some guardrails which they are managing).

As one more note, I would nevertheless not commit to any bet that Intel will certainly not cut its dividend, but such a thing would in my opinion depend more on the evolution of the overall PC and data center demand, rather than AMD. As case in point, AMD’s pre-announced Q3 PC financials were objectively far worse than Intel’s Q2. Intel reported a COVID-hangover combined with inventory burn, while AMD reported a total collapse.

Layoffs

Seeing the damage that the 2016-2017 reorganization did to the company, certainly I am no fan of the latest round of layoffs. However, one must also admit they are perhaps just the simple financial reality:

Intel cut its annual sales forecast from $79 billion to $67 billion in July amid a steep decline in PC demand and a softening data center market.

Intel hired on the order of 18k net new employees over the last 18 months, so it is unlikely there will be a complete reversal, but with around ~$15B in lower sales than expected (the low end of the guidance range is $65B), something has to give. So if as discussed above Intel is unwilling to cut the dividend, then apparently it are the employees. Nevertheless, the article that details the layoff plans mentions that Pat Gelsinger noted Intel’s low current gross margin profile, but Intel has only itself to blame for falling behind in process technology and the low yield of 10nm and 7nm. This is something that can only be fixed by Intel’s employees.

Summary

Some of the key points:

• Intel’s earnings may be bad in the near-term, but that can (almost certainly?) only get better.
• The dividend is not at risk.
• Intel has been competitive in the PC market since Alder Lake, which has been ramping in 2022, and despite the process disadvantage over the next year, the competitive environment isn’t changing materially over the next year due to Raptor Lake being a strong improvement.
• AMD has been pushing the power envelope of its highest-end CPUs (at the expense of efficiency), just as Intel has been doing for several generations node.
• Arc is not delivering the performance one might expect based on theoretical FLOPS (an issue that had been expected and readily admitted by Pat Gelsinger and the GPU team itself). In addition, since AMD was already competing favorably on performance per dollar vs. Nvidia, the value proposition of Arc isn’t quite as large as Intel portrayed.
• While the downspecced Mobileye IPO is at the expense of the amount of money Intel can raise, perhaps in order to fund the fabs, R&D and dividend, it provides a compelling entry point for investors to buy the stock. There is no reason to complain.
• The argument that Mobileye’s valuation has remained stagnant because of its current bear market valuation is nonsensical, and at best could be applied to many, many more companies that have seen their valuation being compressed like a souffle under a sledgehammer.
• Intel has been and continues to build a highly credible foundry business, offering a complete systems portfolio (including leading edge process technology as well as more niche offerings from the upcoming Tower acquisition), which even includes dedicated capacity corridors.
• IFS is open to all customers, including AMD, Nvidia and Apple, and given the rising wafer prices and accordingly gross margins that TSMC has been charging since the chip shortages (as well as its half a decade of standstill with N3 and N2), surely leading edge fabless companies have every reason to check out its offerings.

Investor Takeaway

Referring back to my relatively recent discussion about the Intel stock and financial trends, the thesis was that, in hindsight, due to underinvestment Intel might actually have been overvalued when it was trading for $50-60, as it was not reinvesting enough profits back into R&D. In addition, the gross margin erosion might cost another $10 or so in stock price.

However, the forward-looking thesis was that the gross margin weakness should improve over time, so with the stock price now firmly below $30 due to recessionary cautiousness, Intel is trading below any reasonable steady-state valuation. Moreover, if Intel does not revise its guidance downward, then this would be proof of the bottom.

In addition, there is much further upside that no analyst estimates will recognize (or would even be able to predict) until it actually materializes, namely (1) regaining process technology leadership, and, more directly, (2) the three new emerging businesses (robotaxi, GPU and foundry). Therefore, estimates should be taken with a grain of salt.

As discussed here, Intel is in-line on robotaxi (it is not sure if any delays are due to regulation or technology), a bit behind on GPU (but no fundamental roadblocks as it now comes down to improving the software stack further over time and executing on the roadmap), and a bit ahead on foundry.

Given Intel’s mid-quarter confirmation that it was performing within guidance, the thesis of Intel bottoming seems substantially de-risked, so investors don’t have to wait until the earnings to buy the stock. Although Intel does have a history of dropping after its earnings reports.