Investment thesis
Intel (NASDAQ:INTC) has consistently delivered steadily improving financial performance throughout the year. Meanwhile, the company has not missed a single execution milestone, paving the way for much improved products that will pave the way for continued financial improvements as well. In other words, the downward spiral from the previous several years has been reversed in a flywheel propelling Intel back to industry dominance.
Since Intel is still in the midst of this transition between these two narratives, the stock (still) provides a very compelling entry point (as it has done for the past year). Get yourself an incredible early Christmas gift.
Background
As detailed in my previous coverage from September, the Intel turnaround has recently started transitioning from being just a plan, to entering its final stages of execution, evidenced by the signing and prepayment of an 18A foundry customer, the node with which Intel will reclaim industry leadership in 2025.
Q3 results
Revenue increased to $14.2B with reasonably healthy 45.8% gross margin, up through the year quite significantly from $11.7B in Q1, which itself was already a beat compared to guidance and expectations. This shows that the inventory digestion thesis has played out as the company and industry is coming back from the downturn. Although with the PC, data center and network/edge TAMs down, there will not be a quick complete recovery. Intel has now also exited 10 businesses, which further weighs down revenue by some undisclosed amount.
In addition, there also has been a recovery in terms of EPS and gross margin (back into the 40s). Intel is achieving its $3B cost saving goal, achieved in part by cutting its workforce by around 10k as well as those exits. Note that not just profitability improved quite significantly (evidence of the significant leverage from higher revenue, the reverse of when revenue started to decline, with Intel guiding for 60% gross margin leverage for Q4), but adjusted FCF of $0.9B was also sufficient to cover the reduced dividend. So Intel is again quite healthy, and the many overblown commentaries from when Intel was not for a few quarters are already outdated; showing that there was nothing structurally wrong with the business except for the downturn coinciding with the fab buildout.
Nevertheless, the YoY comps are still down, although this should reverse next quarter on easier comps. On the assumption that the PC segment has stabilized and normalized, and that the data center has also bottomed, with growth expected in 2024, then the current quarter might be viewed as a baseline for performance going forward. The main wildcard is the NEX segment, which is having its own inventory issue. Based on the general macro outlook, it seems that this segment likely will not achieve its long-term goal of mid-teens growth, which would be quite a headwind to reach some of the ambitious revenue targets from the 2021 investor meeting (especially since two of the three big bets, Mobileye and graphics, are not executing on point).
Guidance
Q4 revenue is expected to be another step forward. Combined with the cost saving efforts, the main significance is that it gives Intel enough flexibility to execute on its plan, as the downturn happened during quite an unfortunate/critical period of heavy investments in both R&D and capex.
Victory laps
Without doubt the most significant bit buried in the earnings call, besides yet another confirmation of the process roadmap remaining on track, was the signing up for no less than two additional 18A customers, with a potential fourth one targeted by the end of the year. As a reminder, the first 18A customer was announced in September, with the significant detail being that said customer had made a prepayment that caused Intel to accelerate its Arizona buildout plan.
At the beginning of the year, I promised you one [external committed foundry customer]. Here we are in the third quarter, we have three, and I hope to finish at least the fourth before the end of the year.
It is not sure who the 18A customers are (i.e. if they are top 10 foundry customers), and what kind of volume they will bring (likely not their whole portfolio at once), but after Intel started running test chips through the fab about a year or less ago, signing up the first customer is clearly a great next milestone and stepping stone towards building out a competitive foundry in terms of scale. Intel also announced that it has completed the 0.9 PDK, likely paving the way for continued customer interest in the node.
I have been studying SEM diagrams for almost 40-years. RibbonFET and PowerVIA are true works of art, the most exquisite transistors ever created.
For comparison, this article from a year ago stated that customer had access to the 0.9 PDK for TSMC’s (TSM) N3E, which was announced for manufacturing in H2’23. However, since Qualcomm (QCOM) has already launched its Gen 3 flagship SoC, it is not sure if there will be any SoCs in the market on N3E before the second half of next year. Hence, conservatively it could take perhaps two years from 0.9 PDK to product launch. This would imply that 18A foundry products could hit the market in the second half of 2025, which is in-line with the expectation of 18A starting volume production in early 2025, with the first Intel 18A products (Clearwater Forest and Panther Lake) likely launching from mid-2025 onwards (about a year after Intel 3). Overall, this suggests Intel really is not holding back with providing external foundry customers access to its leading edge node(s), fulfilling its promise of create a true standalone foundry business.
Yes. And first, I’d say we’re — come to a different conclusion than what you might have heard from them. We feel that our five nodes in four years, the leadership position that we expect with Intel 18A, this is a remarkable set of work. And as you heard me say in my formal comments, we think of 18A as a work of art. This is the finest transistor, right? And we’ve invented the last 30-years of transistors. This is the best one that’s ever been built, right? And that and PowerVia, we feel very confident that we are on track to the leadership position that we described.
In packaging, Intel announced two AI wins (which quite likely may include Nvidia (NVDA)), with an additional six customers in the pipeline.
But the other thing that we saw this quarter, which was a little bit unexpected was this huge surge in interest for AI customers and Intel’s advanced packaging technology. And this is the 1 that probably we didn’t even quite realize. We always knew that Intel’s packaging technology was the best in the industry, but the amount of interest that we’ve seen. So we completed two additional customers, so three 18A wafer customers, two additional packaging customers, and we have a pipeline that we’re in active negotiations with six additional packaging customers as well.
Intel plans an IDF industry event in Q1.
Further discussions
AI
With regards to Intel’s AI everywhere strategy, the company has been touting local AI, such as with the advent of the AI PC with Meteor Lake. However, the company hasn’t disclosed any potential financial benefits from such a possible trend. While every Meteor Lake SKU will include a Movidius NPU, it is unknown if this will lead to increase in ASP (average selling price).
Qualcomm
In addition, Qualcomm (QCOM) has made some high-key announcements recently on the PC side. While Qualcomm has dabbled before on the PC side, it hasn’t gained much if any momentum as it didn’t have a competitive architecture. This is now changing due to its Nuvia acquisition. While Qualcomm showed competitive or even slightly leading performance, this was against the current-gen chips from the competition, with both Intel and AMD (AMD) expected to launch their own next-gen by the time of its mid-2024 launch – Meteor Lake is in fact already shipping in preparation for its December 14 launch.
Nevertheless, Qualcomm showed some especially impressive power efficiency numbers, reminiscent of Apple (AAPL) silicon (which isn’t a surprise as Nuvia was created by former Apple employees). However, while this means Qualcomm should become a credible third competitor in the PC CPU market, the graphs Qualcomm showed failed to incorporate the turnaround Intel is completing in exactly the next year, as discussed above. Going from Intel 7 to Arrow Lake on 20A within the 14 months is effectively a full two node jump in area density and a three node jump in power-performance (7->4->3->20A). This means any comparison Qualcomm showed will be utterly irrelevant a year from now as this is effectively comparing pre and post the turnaround.
So it is nice to see some additional competition for Intel to compare itself and compete against, as this will all the more prove what Intel is accomplishing after having had to endure the strongest criticism for more than the last half a decade. Still, I was actually dissatisfied with Pat Gelsinger’s reply during the Q&A, saying he didn’t expect Qualcomm to pose a major threat. While this may have been to some extent a factual answer based on expectations for gaining market share, there was zero technical discussion regarding expected competitiveness. Investors may remember that “unquestioned leadership in 2024-2025” was one of the major points from the CEO appointment in 2021, but this hasn’t been further discussed since (only the process technology side).
GPU
Another point definitely worth mentioning is that Intel has partially backtracked on Pat Gelsinger’s intent of allowing investors to see the financial progress of its big bets. Following its reorganization, Intel is no longer disclosing the financials of its GPU division. In fact, in the wake of the departure of some notable folks including Raja Koduri and Rying Shrout, one might start to wonder to what extent Intel will even continue in this market.
While the data center side seems quite sure to continue, with the folding together of the Habana Gaudi and Ponte Vecchio product lines, the PC side seems more questionable. This would be unfortunate for gamers who were looking forward to a credible third alternative, as Nvidia continues to squeeze every last penny from its customers.
Note that Raja Koduri had guided towards $10B revenue in 2026, so underperformance here would be quite a letdown, especially given the surge in market growth.
Other
For those who missed it, Intel announced a 288-core SKU addition to its Sierra Forest roadmap, partially backtracking from the “roadmap simplification” touted by Sandra Rivera, who by the way is moving to the FPGA division which Intel is making into a standalone business yet again, similar to Mobileye (MBLY). While this is not quite the 384-core or 512-core SKUs rumored initially, it should allow Intel to compete more thoroughly on performance against the 128-core (256-thread) Bergamo and 192-core (384-thread) Turin. In 2025 with Clearwater Forest, Intel should achieve a more significant leadership if it launches a full-blown portfolio; if the initial Sierra Forest line-up was capable of 512-cores, then Clearwater Forest might stretch this towards 1024. In any case, the main point is that Intel’s move to chiplets is bearing results, as the addition of this 288-core SKU would “simply” have entailed adding a second compute chiplet between the two I/O chiplets.
Another notable mention is the divesting of the pluggable silicon photonics business, marking the tenth business Intel has exited. As someone who was a big proponent of BK’s diversification strategy towards becoming a data-centric company, this is the only part that I don’t quite agree with. Clearly all these divestitures are the result of the cost saving efforts due to the downturn (together saving $1.8B according to Intel). This isn’t just short-term thinking, but also very much bean counting reminiscent of the Bob Swan days, which is unfortunate.
Mobileye is clearly behind schedule on its robotaxi plans. Mobileye had previously already said it wouldn’t run a robotaxi business itself (a big missed opportunity), but it had originally disclosed launches in various parts of the world throughout 2022-23, which clearly hasn’t happened.
Gaudi pipeline has doubled from last quarter, when it ballooned to over $1B. While this is neat, it still pales compared to Nvidia, and even AMD is already speaking of $2B revenue in 2024, with Intel only able to confirm “hundreds of millions” of dollars so far. So despite having not one but two competitive products in TCO (Ponte Vecchio and Gaudi2), it not capitalizing that much on the GPU shortage.
Investor Takeaway
The conclusion is simply that the turnaround remains on track, going exactly as planned. This itself is incredible as progressing through two full-blown nodes in 12 months has never been done in the entire semiconductor history of Moore’s Law.
While in the near-term the focus remains on the technology side, as 2025 progresses I would expect the first meaningful financial improvements – beyond just those from the economic recovery – to become visible, becoming more pronounced in 2026 as the 18A portfolio starts meaningfully ramping (which takes about a year).
So while there is still some patience required by investors, as I have been pointing out for more than the last year, given the economic downturn in the semiconductor industry, Intel stock has been cheap (in absolute valuation compared to the Bob Swan and BK eras) during this time. This provides the chance for elevated alpha in the wake of the recovery followed by the results from the turnaround.
While explosive stock results such as AMD and Nvidia have seen is not guaranteed, the combination of very low risk and potential for elevated gains should make the stock worth considering for most investors.
Read the full article here
