Good afternoon. I'm Timothy Arcuri, I'm the semiconductor analyst here at UBS, and very pleased to have Intel. We have John Pitzer with Intel. John and I competed for many years, and this used to be John's conference. So anyway, John, you're back.
You've done a fantastic job, by the way. This has been a great day, so I appreciate that.
Perfect. So anyway, OK, John, before we start, I have to read a statement on your behalf. Before we begin, please note that today's discussion may contain forward-looking statements that are subject to various risks and uncertainties and may reference non-GAAP financial measures. Please refer to Intel's most recent release and annual report on Form 10-K and other filings with the SEC for more information on the risk factors that could cause actual results to differ materially, and additional information on our non-GAAP financial measures, including reconciliations where appropriate, to the corresponding GAAP financial measures. So thank you, John.
Thanks for reading that. Appreciate it, and thanks for the opportunity to talk to everyone today.
Perfect. So John, let's talk about just the shortages that you're experiencing right now. Demand's definitely better, but part of it also is that you have tightness on Intel 10 and 7, and you are diverting products, or you have a mixed issue such that you're short on supply. So can you actually talk about both PC and server demand and also supply?
Yeah, it's a good question, and thanks everyone for joining this afternoon. I mean, listen, I think I'll break this down into the PC market and then the server market. I think we've been in a relatively robust PC environment all year, Tim. I think in the first half of the year, we were concerned that the strength was really tariff pulling, people trying to get ahead of potential higher costs in the second half of the year by buying PCs in the first half of the year. I think clearly as we went into Q3, that concern kind of dissipated, and we think there's some sustainability to what's going on in the PC market. I think more importantly, the bigger shift was really what happened in the server market in Q3.
I mean, it was a relatively OK market in the first half of the year, but it wasn't sort of a very strong market in the first half of the year. As we went into Q3, I think clearly customer conversations changed, their demand signals changed, and it wasn't just near-term volume. As we talked about on the earnings call, we now have multiple CSPs that are looking for longer-term supply agreements that stretch out beyond next year. And that has put us into an uncomfortable position, as you pointed out, that we now are undershipping demand both on the client side and on the data center side. We said on the earnings call that the shortages are most pronounced in 7-10, but really we're short just about everywhere.
The reason why they're most pronounced in Intel 7 is that's still the mainstream of our volume today, both in client and in data center, which is why we called that out. We also talked about the peak of the supply shortages occurring in Q1. That's going to be the most difficult quarter for us to get wafers out. Things will get better beyond that, but we're still going to be in short supply even after Q1.
And so we have a situation where there's multiple hyperscalers now asking for long-term agreements. Can you also talk about that?
Yeah, I think it's simply that. I think one of the things that I get asked often, and we're trying to get our arms around, is what's really going on with traditional server demand. And I think there's probably three dynamics at play. One, you have relatively easy comparisons. For the last several years, the hyperscale community has not been refreshing their installed base and been rightfully going after AI infrastructure build. And so they've put that a little bit on the back burner. I think the second dynamic, which is somewhat related to the first, is they're also very power constrained right now. And if you were to take, and we talked about this on the earnings call, a server in the installed base that's five years old, and you were to upgrade it to a new server, that new server is going to be 80% more power efficient.
And so there is a significant ROI to actually start more aggressively sort of refreshing the installed base. The third dynamic, which I think is probably the most interesting one, is just I think as you see AI transform from LLM training into agentic and inference, it's just absolutely putting demand on traditional infrastructure that I think surprised us. Quite frankly, it surprised our customers. And I think that's one of the reasons why we're seeing sort of a desire by the CSPs to get some longer-term supply assurances in place.
Do you think that that drives some upgrades? Do you think that it drives an upgrade cycle?
I think clearly it's driving capacity needs, quite frankly. I think as you think about the LLM and the training model, all the spend has been done for that. But the LLM is basically the brain of AI. It still needs your personal data for you to get real insights out of it. And that's really a RAG model that is putting, I think, tremendous demand for traditional compute that has some sustainability to it.
You know, John, one thing I hear, and you actually got these questions coming out of earnings, was that, well, the customers aren't buying the new product. So there's some perception that customers are still buying the old product, and that's part of all of this. Can you?
Yeah, I understand the question. I would remind you that the vast majority of our capacity today is still on Intel 7, and that's why we're tightest there. Quite frankly, if we had more Granite wafers, we'd be selling more Granite. If we had more Lunar Lake wafers, we'd be selling more Lunar Lake. If we had more Arrow Lake wafers, we'd be selling more Arrow Lake, and so I think we feel pretty good about where we are in the AI PC transition, and we feel very good about where we are in the initial phase of the Granite ramp, which is our latest generation server part.
Great. So let's talk about 18A. And Panther Lake is out. Yields, I think, were initially unpredictable and low, but I think that's back on track. So can you just talk about 18A and progress? It seemed like Lip-Bu was a little more bullish on the progress last call.
Yeah, it's a good question. I think that we feel good about the progress we're making on Intel 18A. We had committed to everyone to get our first SKU of Panther Lake out by year-end. We have done that. I encourage everyone to get to Las Vegas in early January because you'll hear us, and more importantly, our OEM partners talk a lot more about the excitement they're seeing around Panther Lake. We'll ramp multiple additional SKUs in the coming months in the first half of next year. You know what I would tell you is we're still not satisfied with where yields are, but relative to when Lip-Bu came back to the company in March, he was clearly unhappy with where yields were absolutely, and to the point that you made, he was probably more displeased by the fact that progress on the yields were very unpredictable, very haphazard.
I think that he kind of tasked the team himself and Naga to really double down and focus on getting yields right for Panther Lake. And I think we're starting to see the benefits of that because at the very least, yields are still not at the levels we want them to be, and they will continue to get better over time as Dave talked about on the earnings call. But we are now in a position where we're seeing predictable improvement month on month that's in line with what you would expect as an industry average.
Great. And then can you kind of dovetail that into 18AP? And you've made progress on the PDK milestones and the plan to leverage the 18A family. And you've said that you're going to keep on leveraging that family for the next three generations of both client and server. So when would you expect to get more news on 18AP and for us to have more milestones?
Yeah, it's a good question. I think there was a little bit of confusion coming off the Q2 earnings call. 18A is both a node and it's a family. And I think when you think about 18A as a family, there's 18A, there's 18AP, and there's 18APT. 18A is going to be mostly an internal node. As we transition to 18AP and we've seen good maturity on the PDK there, it will be used both for the internal customer and will re-engage external customers with that node as well. And we've got good engagement with customers going on now with Intel 18A. And like I said, the PDK maturity has come along nicely. In addition, 18APT will be our base die for advanced packaging, and that will be used for both internal and external customers.
And it's one of the reasons why on the Q2 earnings call, we kind of went out of our way to just tell people, we think this is going to be a very healthy, long-lived node that generates a very good ROI. I think Dave talked about peak wafer starts for 18A really not happening until the 2030 time period. And so this is a capacity point that we think is going to be a good ROI for our investors.
Great. Can you remind us what the outsourcing ratio is going to be? I think you said it's 30% this year. Can you remind us sort of some milestones, maybe the direction of it in 2026 and 2027?
Yeah, so it's a good question. So what's been driving the outsourcing ratio this year is really the fact that both Arrow Lake and Lunar Lake are parts that are fully outsourced to external foundries. As we start to think about Panther Lake, that starts the process of actually bringing wafers back in-house. And what we've said in the past, and it's still true today, is that about 70% of the logic compute, the logic tiles for Panther Lake, will be done internally. That's versus 0% for Arrow Lake and Lunar Lake. That gets even better with Nova Lake because remember, Panther Lake is just a notebook part. Nova Lake is also desktop. So as we move into the end of next year, into 2027, we'll get the added benefit of bringing desktop wafers back as well.
Now, what I will tell you is we try to navigate through this current period of tight supply. We are going to lean more heavily on our external foundry partners around Arrow Lake and Lunar Lake. And you and I have talked about this. I think one of the bigger changes from Q2- Q3 is at the end of Q2, I would have told you that Lunar Lake probably peaks from a volume perspective in Q3, is kind of flattish in Q4, and then starts to fall off after that. I think now, as we think about trying to augment the shortages in PCs as we move client wafers to data center, we're leveraging more Arrow Lake and Lunar Lake. Lunar Lake is going to grow sequentially in Q4, and it will be up next year, year- over- year.
That has some challenges at the gross margin level because of the embedded memory that is really a pass-through cost for us.
Great, John. Can you also then extend that to 14A? How is the node development progressing? What are the next milestones we can expect on 14A?
Yeah, we're pleased with the early engagement. As I've talked about in the past, simplistically, I think about every node having three phases, sort of definitional development and then high-volume manufacturing. 14A is a very different node for us because on the 18A node, in definition, we were only engaging with Intel products. And all the decisions we made at the transistor level was really to optimize for what they needed, which was a CPU tile. It wasn't really until the development phase that we actually started engaging external customers. And to be fair, for many external customers, the fact that we optimized for Intel products didn't matter. For others, it did. And we underexecuted on 18A, and had we executed better, we probably would have had better results to show. I think importantly, on 14A, we are engaged with external customers in the definitional phase.
Quite frankly, we're making optimization choices more to favor what they need than necessarily the internal customer. And what that means is we're getting earlier, more, and better feedback about how we're doing on the node. The other point that I'll make is, as you know, with 18A, we tried to bend physics in two new ways, right? Moving from FinFET to gate all around and then implementing backside power. There's a lot of complexities on 14A, but at the very least, it's a second-generation gate all around. It's a second-generation backside power. And we've been very clear that if you look at yield and performance today on 14A and compare it at a similar point in time on 18A, we are meaningfully ahead on the 14A node, which is also encouraging.
And then the third thing that helps is I kind of tongue-in-cheek like to joke that we didn't know how to spell PDK until we were about halfway through Intel 18A development. And quite frankly, we struggled a bit with getting our PDKs to be very industry-standard-like. We're not having that problem either with AP or with Intel 14A. And so I think that customers are pleasantly surprised that our PDKs are just much easier to use.
What's the next most important milestone that we should be listening for on 14A?
It's really customer announcements, quite frankly. There will be incremental progress that we'll learn, but I think the big one is really when can we get an external customer on either Intel 14A or Intel 18AP? And what we've said on that front is the timeline to think about for when fabless customers have to start making hard decisions about 14A designs, that window opens up in the second half of next year into the first half of 2027.
And even at the same time, though, you said that even if you do sign a customer, you might not even tell us about it. So is there anything else tangible besides just to?
No, I mean, listen, I think we have a guiding principle that our job as a foundry is to try to make our customers successful. It's not to try to market their names. And clearly, I think if the customer would like to put out a press release, we're not going to be adverse to doing that. Quite frankly, I think we've had some good success with some advanced packaging customers that we haven't talked about. And I think that that's just par for the course of a good foundry. You don't see other external foundries talking about specific customers, and we're going to follow that example.
Great. I actually wanted to ask you about advanced packaging. So the external advanced packaging revenue is still pretty small, but we do hear about a lot more engagements there. And you've been very clear that the back-end engagements will lead the front-end engagements. So would we think we're about to see an inflection point here with respect to EMIB and EMIB-T, given some of the tightness we're seeing for CoWoS?
Yeah, I mean, we're pretty excited about the technology. I mean, if you go back and think about the journey we've been on in advanced packaging, we were pretty excited about the business about 12-18 months ago, in large part because we saw a lot of customers coming to us to be spillover capacity because CoWoS was so tight. And for full transparency, we probably underhit the potential for that business. I think TSMC did a very good job increasing CoWoS capacity. We probably underperformed a little bit in getting Foveros where it needed to be. But the advantage of having that happen is it brought customers in the door and allowed us to start moving from tactical conversations to strategic conversations. And when we did that and we started to show them our roadmap against EMIB and EMIB-T, they were pretty excited about what they saw.
We do think that EMIB brings capabilities into the market that you just can't get with CoWoS. So it's not a spillover. It's really a technology buy that people are focused on. We've got some good momentum there. I think the proof is in the pudding as to when we actually start to see revenue move through the P&L. I think that's probably something that starts to ramp in the back half of 2026.
Got it. Let's talk about how you prioritize capacity between server and PC. Obviously, these hyperscalers are coming to you wanting to have LTSAs, and they're talking about that. Obviously, and you've also said that you're prioritizing wafers to server. And so at the same time, PC is getting better as well, demand. So are you happy losing some PC share? You're okay with that through 2026? Because I ask that because AMD put up a pretty aggressive share gain target in PC.
Yeah, I wouldn't say happy. Never happy and losing share. But I do think as we think about how we're going to navigate through this tightness in supply, we've been fairly clear that we're going to de-emphasize the low end of the PC market. Quite frankly, that's an area where our closest competitor actually doesn't play. So it's going to be interesting to see how market share trends play on that front. But we're really going to focus on trying to optimize for revenue and profit share in the PC market given our tight capacity. Now, we're doing things in that business today that are both gross margin accretive and also create gross margin headwinds, as we've talked about. I wish our gross margin story was a little simpler. But for example, de-emphasizing low end, positive for gross margins.
On some of our lower end SKUs, we are going to raise pricing, positive for gross margins. Offsetting that, though, we are going to work closely with our OEM customers because we know we're undershipping demand, and we will do demand shaping and change pricing on Arrow Lake and Lunar Lake that allows our OEM partners to bring those CPUs into lower price points within the PC stack, and that obviously is going to create some gross margin headwinds that we need to manage through next year.
Just on that front, Lip-Bu has been kind of rethinking the entire data center roadmap. I think Pat was maybe a little more bullish on P-core on E-core. The idea was originally that Diamond Rapids was going to stem some of those share losses. It sounds like now maybe that optimism is pushed out to Coral Rapids. Can you just talk about how the data center server roadmap is changing?
Yeah, it's a good question. I mean, I'll remind people we're still very early in the Granite Rapids ramp. And we actually do think Granite Rapids does a good job of helping to close gaps in many parts of the market. And so we should benefit from that over the next several years. As you think about Diamond Rapids, Lip-Bu has made some proactive decisions to take some of the SKUs off of the roadmap, mainly at the low end, where he came to the determination along with Gevorg that we're just not competitive on cost and we're just not competitive on performance. And while it's not an ideal position to begin, it's actually very healthy for him to do that because it sends a very strong signal internally that we're just not going to accept subpar products.
And it also frees up resources to go off and work on products that we think will be competitive. As you know, Tim, it takes multiple years from cradle to grave to get a design out the door and into a product. I think when you think about the opportunity that Gevorg and Lip-Bu have to really embed their DNA into a server part, that first opportunity is probably going to be Coral Rapids.
Coral Rapids, okay. So John, let's just talk about PC. Do you feel like you're losing some ground in PC? I mean, and how do you think about you're fighting off AMD, you're fighting off ARM? How do you think about your PC roadmap?
I mean, we feel pretty good about the PC roadmap. Quite frankly, I think that this year was a challenging year for us at the high end of the desktop market. We've been pretty, I think, transparent about that. I think a lot of the damage that has occurred there has already happened. I think that as you think about Panther Lake, I think it shores up our position in the notebook market. And as you bring Nova Lake out late next year into 2027, we'll have a competitive part across the PC stack, across both notebook and desktop. And so we have very strong share there today. With Panther Lake next year, I think, share stabilization, notwithstanding the fact that we might de-emphasize the low end of the market, just be mindful of that. That could impact kind of a unit share number.
But share stabilization from a roadmap perspective is what I would think about. And as we can bring Nova Lake to market, I think we are very happy with our competitive position.
Just on PC, John, how have you been following these markets a long time? How do you think about memory price increases impacting the market? I mean, it's historically been, at least people think of it as being a fairly elastic market. Can you just talk about that?
Yeah, it's a good question. I mean, this surfaced in mid-November. The market got very concerned about this. And so I spent time with the team kind of going through some of the historical data and just looking at DRAM as a % of the BOM, how it's trended over time, and what happens to units and mix when it gets to sort of a pain threshold. Because we're clearly going through that today. I thought what was interesting is the data is not conclusive that it has an impact on either units or mix, quite frankly. I know it creates a lot of chatter in the marketplace, but the supply chain is pretty resilient about figuring this stuff out. Now, it's clearly a dynamic that we're hyper-focused on.
I will tell you today, if you go out and talk to our OEM customers in the PC market, what they're most concerned about is our inability to get them units, and so that's not a great place to be in, but at least it's a high-class problem and not a low-class problem, but we will monitor this situation very closely, and quite frankly, if on the margin, the market is a little bit less robust next year in a market where we're having trouble meeting demand, I'm not sure it's going to bother us all that much.
Great. John, let's talk about AI. Can you just update us on where the roadmap is? I know there was some optimism some time back about Falcon Shores, and just talk about where the accelerator roadmap is.
Yeah, so you're asking a specific question around AI accelerators. I'd like to remind people that AI kind of permeates almost everything we do. There's the AI PC strategy. There's clearly the uptick in traditional servers that's being driven by agentic AI and inference. I would say our advanced packaging is being driven by AI. Quite frankly, our wafer business is being helped by the AI trend as well. But specific to the accelerator market and the GPU side, we clearly have work that we need to do. I think Lip-Bu has defined a strategy that suggests the part of the market where we think we can make a difference and solve a customer problem is really on GPUs that are optimized for power, for inference. We're not going to go and try to duke it out in sort of the training market and the AI accelerator business.
The other thing that we talked about on the last call, which is relatively new, is this new ASIC business that Lip-Bu's put under Shrini. ASICs is actually an area where we have a pretty healthy business today. It's mostly in networking, but we have multiple customers there. And I think people would be surprised by the revenue run rate we're seeing in ASICs. And quite frankly, as you know, AI is driving a lot of demand for networking. And if you see the forecast for revenue for that business, it's following sort of the industry trend of what we're seeing in other networking assets driven by AI. I think what Shrini is tasked to do is to build that into a real business. We've sort of treated it as a hobby.
I think he's going to be, I think, much more focused on trying to bring, whether it's an ASIC CPU or an ASIC accelerator, to the market. And I think we have a lot of the IP blocks that are needed. And I think Lip-Bu and his customer conversations have seen a real need here. So we're pretty excited about that opportunity.
Can you talk, John? You've been there now for several years. Can you talk about Lip-Bu coming on board? How has the company changed under Lip-Bu?
Yeah, it's a good question. I mean, at the end of the day, whether or not we're successful in our strategy or not, it's going to come down to culture and incentive structures. And I think Lip-Bu has come in and been very aggressive in trying to change culture and incentive structure. And I think some of the visible signs is I think when he joined the company back in March, we had 11 or 12 layers of management. We've now taken that down to five or six. And so he's flattened the organization tremendously. In addition, we sort of had an engineering group that didn't report directly into the CEO. It reported up through the business units. He now has engineering sitting on his executive staff.
In addition, those engineers who never want to go talk to customer; it's now mandated that whenever we have a customer conversation, there's an engineer there to talk with the customer, and so it's really getting back to kind of an engineering-focused, customer-centric sort of culture with a lot of transparency. I mean, again, I think the decision that Lip-Bu made on some of the SKUs and DMR is very healthy for changing the culture of the organization. Because historically, if we were to miss milestones on new products, we usually just gave them exemptions. Saying no is a pretty powerful tool, and I think that that's kind of being viewed in spades, and I think it's healthy. In addition, the last point I'll make is return to office happened on September 2nd. It's great to get people back in the office.
We're back in about four days a week now, and it really is helping to drive incremental collaboration.
John, you used to be an analyst. So you sort of think about the company being on the inside. I'm sure you also think about it still in your old hat, too. So oftentimes people argue, well, Intel should be broken up into the foundry and into a fabless chip business. How do you think about that? I mean, it's easier said than done. You don't have any external customers now for the foundry. So it's kind of a moot point right now. But how do you think about that? What boxes have to be checked for you to get to that point if you were going to do that?
Yeah, listen, we're clearly taking steps to create that optionality, right? And a lot of that is being driven by engagements with external customers who would like to see more separation between the foundry business and the rest of the company. And so we've got an advisory board today for Intel Foundry. We're moving that into its own legal entity. And so we continue to move down that path to give ourselves that optionality. To your point, financially, pragmatically, it's probably not something we can do today. You've all seen sort of the split out of the two P&Ls. And I think first and foremost, Intel Foundry has to get healthier financially. I think we're on a path to do that as we bring 18A wafers in.
But I think from a longer-term perspective, if it's the right decision to create value, to create more separation, I'm pretty confident that Lip-Bu, Dave, and the board will move in that direction.
And John, one of the big keys to improving profitability is to get off of Intel 10 and 7. And the wafer costs don't change very much, but the wafer prices go up a lot. So that helps margins in foundry. However, the transfer pricing to the product business goes up alongside of that. So is the product roadmap in a good enough place that it's able to absorb those price increases when you think about this transfer pricing model?
Not as good as we would like. We have work to do there. I would say client probably further ahead than the data center roadmap. But I want to be very clear. As we move to Intel 18A, it's very accretive for Intel Foundry margins. It's also accretive to Intel corporate margins. To your point around the higher price that Intel products need to pay for 18A wafers, it does create some mixed dynamics within Intel products that they need to go figure out. But net-net, it is absolutely a positive to bring wafers home.
But doesn't it? So it improves profitability in foundry, but it brings profitability down in the product business.
Yes.
But you think it's still.
All else being equal.
Still a net.
It's still a net positive.
Got it. And then let's talk about CapEx, John. So I think the original plan was this year it sounded like it was going to be down, and now it sounds like it's biased down. But can you talk about CapEx in light of you being short of supply?
Yeah, I mean, listen, we're trying to balance two dynamics. One, we still fundamentally believe that we can run our fabs more efficiently and just drive more output out of the current footprint, which would argue for CapEx being down next year. On the other hand, we've got a situation where we're undersupplying the market, and that's a very uncomfortable position for us to be in and for our customers to be in, and we want to respond to that. I will say that I do think that Lip-Bu and Dave fundamentally are willing to run fabs probably structurally tighter than the historical model would, and we're trying to balance that a lot. I think you characterized it well. We guided gross CapEx to about $18 billion for this year. We were emphatically saying down at the end of Q2.
We were less emphatically saying that at the end of Q3, but I think that that's still directionally right. But it's a plus or minus for next year at this point. Now, I'll also point out we are not pre-building capacity for external foundry. So if we were to win a significant external foundry customer, that would put upward pressure on our CapEx.
Got it. And then can you just talk about, I'm not asking you to guide gross margin for next year, but can you talk about just the puts and takes? And is the 40%-60% drop-through still the right way to think about gross margin as we kind of go into next year?
Yeah, it's still the right rule of thumb. I'm looking forward to a point in time where the number of sort of levers in gross margins goes down because there's a lot of moving parts right now, and it is hard to forecast, especially because it is so mixed dependent. I would tell you that the biggest sort of incremental change from Q2 to Q3 is something I mentioned earlier, which is just the outlook for Lunar Lake, which we thought would be peaking right about now. Looks like it's going to continue to grow as we augment sort of shortness of client supply in what we build internally with more external wafers, and that is going to put pressure. I think we've done a good job, I think, pre-buying a lot of the memory for Lunar Lake, especially given what DRAM pricing has done.
But we still have some work to do there.
And John, last question. How has the involvement of the U.S. government changed the behavior of the company? Has it changed anything at all?
No, I don't believe so. I mean, at the end of the day, even before the U.S. government was an equity holder in the company, they were a critically important stakeholder in the room, and you see that across the technology supply chain. The number of CEOs that are making visits to Washington, D.C., I think they're visiting D.C. now more than they're visiting New York City. I think having them as an equity holder was the right decision for us and for our shareholders. It fully aligns incentive structures, and we're glad that they're a good partner of ours.
Great. Well, John, we're out of time. Thank you again.
Thank you. Appreciate it.