As I’ve been covering more of these quarterly calls, I’m starting to recognize how and when companies are focusing on different parts of the numbers and metrics. Intel’s recent Q3 numbers are doing just that - due to non-cash based restructuring and impairment charges, headline numbers like Gross Margins don’t make a lot of sense until you peel away the onion, and Intel isn’t putting them in the infographics it sends round to the regular press. For a normal company doing normal things, we get a regular cadence of the same similar metrics and the scaling over time - here Intel has lots to talk about, and it requires a nuanced eye. There are more people who delve into the pennies who can do this better than I can, but here are my thoughts and takeaways, especially with a number of comments that CEO Pat Gelsinger and CFO Dave Zinsner made in the Q&A.
Headline Numbers for Hallowe’en:
💵 Revenue $13.3b, down 6% YoY (Guide 13b)
📈 Gross Margin 15% GAAP, down 27.5pp YoY
📈 Gross Margin 18% GAAP, down 27.8pp YoY (Guide 34.5)
💰 Net Income -$16.6b, down from $0.3b
🪙 EPS -$3.88, down from $0.07
The revenue beat the guidance and the market, and as we’ll see with the outlook and guidance, is why INTC 0.00%↑ is up 15% after market close. But why are the margins and income numbers so low?
$2.8b in restructuring charges in Q3 related to cost reduction plan
$3.1b charges recognised in cost of sales and depreciation of assets related to Intel 7 - i.e. they overspent on tools and space during the pandemic and are now writing those assets off (will likely be resold as the second hand market is quite potent)
$2.9b in impairment of goodwill, such as MBLY
$9.9b of non-cash charges related to valuation allowance against US deferred tax assets
In total, that’s $18.5 billion. If we take those numbers out, we’re looking at Gross Margins of around 41%, which is a step up sequentially and actually really damn good for where Intel is right now. Dave Zinsner in the Q&A said that the seasonally up Q3 on that metric is due to Q3 better sell-through with reserved inventory. He was asked this because that number decreases for Q4 guidance, where Intel doesn’t expect the same ‘bump’ to occur.
Outlook
💵 Revenue $13.8b, +- $500m
📈 *Non-GAAP Gross Margin 36.5%
*graph is GAAP except for the outlook number
The outlook shows a sequential increase in revenue, and a GM return to normality. Note that the 36.5% is non-GAAP, and the CFO didn’t rule out additional restructuring costs in Q4. On the $18.5 billion of costs listed above, only $2.3b were direct cash costs, and include Intel’s headcount reduction costs which Intel says is almost complete - but the official headcount numbers tell a different story.
Intel had 124100 employees as of September 28th, which is only 1200 down on 125300 the previous quarter. Intel previously stated a 15k headcount reduction (which on this call became a 15% headcount reduction, or 16.5k), of which we knew that voluntary redundancies were decided end of August to finish in September, and involuntary redundancies were being made in October for November/end-of-year. We suspect employee numbers next quarter to be more in the 110k range as a result, perhaps a little higher as certain regions (Israel) have stricter requirements for layoffs.
However when all is said and done, the street felt the outlay and the outlook were good for INTC 0.00%↑ . As I said, initially up 15% after hours.
Now let’s go into the different business units.
Intel Foundry
💵 Revenue $4.5b, down 8% YoY
💰 Operating Loss $5.8b, down from $1.4b loss YoY
📈 Operating Margin -134%, down from -29.7% YoY
Right now Foundry encompasses both the wafer manufacturing side and the advanced packaging side, and both of those are mostly Intel Products driven given the nature of the process nodes Foundry has at its disposal. The advanced packaging side of the business hit profitability this quarter, a substantial highlight.
The major knock on to the operating loss was asset impairment charges of $3.1 billion related to un-needed capital equipment and space for the Intel 7 process node. Intel 7 can’t be used by external foundry anyway, and rather than throwing good money after bad, Intel wrote these costs down, which also led to an operating margin loss of 71 points. That still means operating margin was -63%.
That being said, Intel has stated that 18A is still progressing well, with key milestones for the leading 18A products on client (Panther Lake) and datacenter (Clearwater Forest) are being met with relative ease. Ramp for 18A is expected in the second half of 2025, with revenue due moreso in 2026. For external design wins, Intel showcased a nearly signed deal with Amazon for an AI fabric chip built on 18A.
Earlier in the quarter, CEO Pat Gelsinger did state that 18A has a defect density rate (D0) of below 0.4 - he means defects per square centimeter. For people wondering what this means on yields, because yield is a function of die size and defect rate:
A smartphone SoC is ~100mm2, or 10mm by 10mm,
Using a defect rate of 0.40, with 0.2mm scribe lines and 5mm edge loss,
that translates to a yield of 67.9%.An large reticle AI chip is 810mm2, or 30mm by 27mm,
Using a defect rate of 0.40, with 0.2mm scribe lines and 5mm edge loss,
that translates to a yield of 8.25%
Aren’t numbers fun! Simply describing defect density as yield is erroneous because of this. It also doesn’t include parametric yield, such as frequency binning, or the fact that most chips have structures that can absorb defects through redundancy. All being told, a defect rate of sub 0.40 at this stage, a year before high volume production, feels like a good number. Pat said they have good line of sight to the work needed to be done to improve that.
Also a highlight for Foundry this quarter was a $3b win from the CHIPS Act Secure Enclave part. Investors in the Q&A though were more interested in customer wins and hard metrics they can judge the success of the ODM model on.
Pat did reiterate the desire to push Foundry to be a subsidiary of Intel, rather than a direct business unit, with its own independent advisory and fiduciary board with external players in the mix. Pat says it’s a needed element to make the foundry business work as a foundry, as (my words here) there are still a lot of players out there suspect as to how truly independent can Foundry be to products given Intel’s history.
Datacenter and AI
💵 Revenue $3.35b, up 9% YoY from $3.08 billion
💰 Operating Income $347m, down from $391m YoY
📈 Operating Margin 10.4%, down from 12.1% YoY
In the last short while, we’ve seen the launch of the Xeon 6 family start to roll out into a competitive market with some good metrics and good responses. The Xeon 6 family technically has four corners - two for the types of cores (P cores in Granite Rapids, E cores in Sierra Forest), and two for the types of platforms (6700 is eight channel memory, 6900 is twelve channel memory). Intel now has a mix of Xeon 4, 5, and 6 in the market, and so customers are oscillating as there’s no one clear platform taking up share.
Pat was keen to point out that Intel Xeon sits at the heart of >70% of all AI-focused GPU platforms, and that’s a lead they want to maintain. He also stated that having four corners for Xeon 6 was perhaps a bit excessive, and part of Intel’s strategy moving forward is to simplify its portfolio, reduce the SKU stack, and reduce complexity. With Diamond Rapids and Clearwater Forest following the same philosophy, Pat must be talking about a couple of generations ahead at least with Xeon 8 for that to be true.
Also in the quarter, Intel announced the formal launch of Gaudi 3, the AI chip, and an integration of Gaudi 3 into IBM cloud. Intel is stuck a bit between a rock and a hard place on Gaudi 3 right now - while promoting its competitiveness and TCO benefits against the competition, during the remarks Intel stated that they’ll miss their target of $500m from Gaudi revenue this year due to the transition from Gaudi 2 to Gaudi 3 but also on the software side as Gaudi isn’t compatible with the OneAPI software stack (that’s based on fundamental technical differences as Gaudi 3 was designed pre-acquisition). The thing is, we can put the $500m miss of Intel against AMD’s $5b+ and NVIDIA’s $60b+ for AI chips in 2024, and can see a big gap. Pat was asked in the Analyst Q&A about this, and his answer wasn’t the clearest. There was also no visibility into revenue for 2025 with Gaudi, which is what I perhaps might have asked. We’re also all waiting for Falcon Shores information at this point.
Client Computing Group (CCG)
💵 Revenue $7.33b, down 7% YoY from $7.87 billion
💰 Operating Income $2.7b, down from $2.8b YoY
📈 Operating Margin 37.1%, up from 35.3% YoY
Given AMD 0.00%↑’s financials earlier this week putting 50% of the company into datacenter territory, Intel is still very firmly a client-focused business, with revenues in client being double the enterprise business. In this quarter we’ve seen launches of Lunar Lake, the new TSMC-foundry based low-power with integrated memory chip to compete for AI PC, as well the initial revenue of Arrow Lake for desktop, which will show more in the Q4 numbers.
It’s an interesting one this, simply because of the statements made in the analyst Q&A. Intel has stated that the use of TSMC for Lunar Lake means they aren’t getting the margin benefits of using their own process node, but still states it was the right node for the right time. On top of that, using in-package memory for Lunar Lake also means a complex supply chain, and given the commodity nature of the memory, the margins for adding it in don’t reflect corporate average. THAT BEING SAID, and I say that in all caps because it’s important, Intel has decided to triple the volume of Lunar Lake chips orders because it wants to ramp this product due to AI PC.
Both Pat and Dave on the call wanted to showcase that AI PC is a big moment, and having a product like Lunar Lake for that market would be a showcase moment for the company, and so as a result it looks like they’re prioritising units over margins here and simply putting Lunar Lake into more of their portfolio. This is depressive of margins overall, even with the volume, revenue, and market share that comes with it. Triple is a lot. Not to mention, but Lunar Lake was also brought in to Q3, when we had all expected it to be Q1 2025, because it was deemed acceptable in production. That also increases costs, for what it’s worth.
However, it should be noted that these depressed margins, while here for a while, will start to evaporate with the introduction of Panther Lake in 2H25 and 2026. Using most of its silicon area at Intel, and using Intel’s 18A, Pat was keen to point out that the more wafers that come back to Intel, and the fact that using the integrated memory was a one-off for Lunar Lake (we won’t see it in Panther or Nova), this compounds better margins for Intel as the product stack replaces Lunar Lake towards the end of next year. It does depend on Panther Lake being ready - as stated 18A will be ramping in 2H 2025 for volume in 2026, so Lunar will be here to stay for a while.
Nothing much was said about Arrow Lake, mostly because that’s a Q4 product on the revenue side. Arrow Lake reviews are alive, to a mixed response. The press like the fact that it’s more efficient than previous generations criticized for massive heat output, but they’re finding its competitiveness on performance and power against the competition more of a fine tuning exercise as some of the quirks of the motherboards and firmware are sorted out. More to come on that one I think.
Networking and Edge (kinda)
💵 Revenue $1.51b, up 4% YoY from $1.45 billion
💰 Operating Income $268m, up from $100m YoY
📈 Operating Margin 17.7%, up from 6.9% YoY
One clear success story in Intel’s business is the NEX business unit - numbers are up all around, although that’s a bit hidden by the fact that this market has gone through some inventory digestion the last couple of quarters. Pat did make reference to a $300m write-down of inventory due to lower-than-expected sales, but nonetheless NEX is still up and significantly so.
No-one really covered NEX on the call - the highlights that Intel shares with everyone notes a design win with KDDI selecting a Samsung vRAN solution powered by Intel Xeon - however it was stated that the Edge side of NEX will be migrated under CCG, given the product focus of the Edge market is drifting towards Intel’s CCG product lines.
Altera (formerly PSG)
💵 Revenue $412m, down 44% YoY from $735 million
💰 Operating Income $9m, down from $263m YoY
📈 Operating Margin 2.2%, down from 35.8% YoY
After a couple of negative operating income quarters, Altera is back in the black, even if only slightly. The business was repositioned as it moves towards IPO with its own separate structures as an ongoing process this year, and in September it held its first Altera-specific developer event as this side of the business now starts to regrow under its dedicated CEO Sandra Rivera.
While nothing on the call was specifically tied to Altera’s YoY losses, it was highlighted that the rollout of Altera’s product families, such as Agilex 5 and Agilex 3 for the mid-range and low-range respectively are progressing. With new management so to speak, I can affirm to the fact that there’s a better intent to increase Altera’s product visibility in the market place. I personally now have a line to some of the Altera staff, and might be showcasing some of what they’re doing in the new year.
Intel did state that the pipeline for Agilex is sitting at $5b right now, with Agilex revenues set to grow 3x in 2025, though this is partly as the Stratix family winds down and Agilex gears up to replace it. In the prepared remarks, CFO Dave Zinsner stated that Altera is expected to have inventory correction with its customers through the first half of 2025 - this is something we saw with Xilinx in 1H 2024 and with Lattice in the last two quarters, so it seems odd that Altera’s exposure is continuing moreso than others.
There was also the mention on the progress of Altera’s partial IPO. The documents stated, and Pat confirmed, that they started initial talks with investors during this fiscal quarter, and they’re expecting to finish in Q1 next year with a lead to that IPO in the coming years. It sounds like they mean a private equity sale, perhaps up to a certain percentage, in the late 2025/2026 timeframe.
Mobileye
💵 Revenue $485m, down 8% YoY from $530 million
💰 Operating Income $78m, down from $170m YoY
📈 Operating Margin 16.1%, down from 32.1% YoY
Again, Mobileye wasn’t specifically spoken about much on the call, especially with everything else taking priority. Two main points that did come through are that Mobileye’s free cash flow for Q3 exceeded its operating income, and also that its exposure to China has been reduced. No detail was given to that point.
Final Thoughts
My consistent discussions with investors about Intel in the calls that I have revolve around one thing: is this restructuring going to work? This leads onto questions as to if Pat is doing a good job, or if rogue investors come in and try to carve it up. Intel is in a unique position, being both product and leading edge foundry in this space, and there are suggestions that if one succeeds and the other fails, which way does the company as whole end up?
Personally I’ve been saying that the best Intel has to execute on all fronts. From a technology perspective, 18A has to deliver on its merits. Pat has been talking a big game of regaining competitive leadership, and we need to see that in action, at volume, with fiscal efficiency. That’s what a lot of the restructuring charges are this quarter - over the last few quarters Intel has been criticised for ‘hiding’ behind new debt issuance, or by relying on risk mitigation. This time Intel has said ‘here are all our costs’, and rather than take the wind out with such bad numbers, the market has reacted positively, because the underlying business does work. In order for IDM 2.0 and Foundry to work long term however, customers need to come. This isn’t Wayne’s World - you have to build it, and sell it, and they will come.
From a revenue and products perspective, Intel is still heavily client focused right now. That leads to seasonality, which may not be a good thing long term. Pat is right in some regards that the complexity of Xeon 6, while leads to domain specific optimized plays, does introduce heavy amounts of design complexity, but in order to shift more on the enterprise side, they have to have cost effective products at scale that deliver performance. I’d argue that between Intel’s Xeon 6 and AMD’s Turin, the enterprise market has never had a better set of products to choose from. But in order for Intel to generate value here, it’s something that has to deliver at scale - even with the right process, you still need the right product.
The biggest miss in all of this is the AI side of the business. As I mentioned before, while NVIDIA is looking at $60b+ for the year, AMD at $5b+ for the year, Intel can’t manage $500m. Over the years Intel acquired Nervana ($400m, 2016), Movidius (2016) and Habana ($2b, 2019) to tackle the AI conundrum, with the biggest success being low powered acceleration in client CPUs and AI PC through the Movidius acquisition. Nervana was mothballed, and Habana’s designs rely on top level framework support rather than OneAPI support. Intel has to decide if datacenter AI is still a top target for the company given the situation its in and what it can dedicate resources - Pat’s answer in the Q&A I feel wasn’t convincing, especially with the comments about redoubling down on x86. When we look at Gaudi 3 deployment sizes, they end where NVIDIA’s start (eg MLPerf on GPT, biggest Gaudi system was 768 chips, smallest NVIDIA system was 1024). This isn’t the fault of the Gaudi engineers - corporate alignment (the miss with Altera being integrated too fast) has been difficult for Intel’s acquisitions over the years. I don’t have a good recommendation here without more knowledge into the Falcon Shores design.
Intel’s biggest product win this year seems to be Lunar Lake. It’s extremely competitive, and will be distributed widely. As I stated above, the downside is that it’s a lower margin part which doesn’t increase margin the more you ship, and is below corporate average. That lends itself to tremendous margin upside in 2H 2025 (probably Q4) as Intel moves to Panther Lake. Pat is spot on when he says moving wafers back to Intel helps in more ways than one - the faster Intel can accelerate 18A into high volume, and assuming it’s as good as they say, then investors should look to the business accelerating in that timeframe.
More Than Moore, as with other research and analyst firms, provides or has provided paid research, analysis, advising, or consulting to many high-tech companies in the industry, which may include advertising on the More Than Moore newsletter or TechTechPotato YouTube channel and related social media. The companies that fall under this banner include AMD, Applied Materials, Armari, ASM, Ayar Labs, Baidu, Dialectica, Facebook, Guidepoint, IBM, Impala, Infineon, Intel, Kuehne+Nagel, Lattice Semi, Linode, MediaTek, NextSilicon, NordPass, NVIDIA, ProteanTecs, Qualcomm, SiFive, SIG, SiTime, Supermicro, Synopsys, Tenstorrent, Third Bridge, TSMC, Untether AI, Ventana Micro.
Analyst Q&A
This Q&A was transcribed live, and mildly adjusted for clarity.
Q: Ross Seymour, Deutsche bank: 18A transition, talk about the metrics we can see externally to give confidence to ramp?
A: Three new customers on 18A this Q. AWS, Panther, Clearwater, two external. Next year, not much financial to 18A, so more qualitative metrics. Continue to give updates, lifetime deal value. As we get more customers, will be updating on progress. More impacts on 2026 financials, move to better margins on Panther and 18A. That’s the clarity. Super important for us and for foundry.
Q: On GMs, 41% without charges. Much better than guidance? What’s the upside, why down in Q4, what's the big picture puts and takes in 2H 25?
Dave: 3Q surprise was better sell through with reserved inventory. Into 4Q, that won’t repeat. Have more startup costs in Q4 with 18A compared to Q3, will put pressure on GMs. Our GM guide has less noise compared to previous, it’s a good guide. The puts and takes of 2025, we’re all in on AI PC, Lunar Lake - memory on package, weighs down GMs on products in 2025. Panther is the next one, better margins on product, but also more mix of wafers internally - helps foundry business. More volume in 2026 with Panther, that’ll help. On foundry, we’ll see GM improvement on foundry partly because of the reductions - reduce spending - $1b in cost of sales and efficiency. Additionally, mix is better to EUV wafers, better cost structure and pricing dynamic. Pleased with the new model managing the foundry with the P&L. Better decision making at foundry and products to optimize cost structure. Long term, as we continue our portfolio in product and foundry, commands better margins.
Pat: As we go through restructuring, next phase of transformation, financial discipline - first phase was getting back in the game.
Q: Tim Arcuri, UBS - Defect 18A was sub 0.4. Does that translate to yield? High volume - what does the customer want?
A: D0 is complex - big dies have lower yields at the same D0. 0.4 is a healthy yield number at this process development. It’s not a high volume yield, but we’re at where we want to be at this point of the process. For HVM in 2H, we have to be markedly lower than that - but we believe we see the signs, signals, to get there. Intel 3, we’re accomplishing D0 with the maturity we expect. Key milestones on quality, yield, then HVM but you keep working. Arizona ramp for 18A, that comes online 2H 2025, very important. First wafers coming out Q1 next year.
Q: Panther Lake, coming back in house. Most tiles still outsource, was that always the plan? Nova Lake, is still dual track, an option for outsourcing?
A: Panther, some will be external, but most of the mm2 are internal, 70%+ silicon area is in-house. Nova Lake has some SKUs continuing to leverage external, but most internal. Still some flexibility, but most are committed to Intel foundry. Executing on bringing wafers on home. TSMC has been a good partner, Lunar Lake demonstrates that. Partnership we’ll be using selectively in the future, but most coming home. Improves margin structure.
Q: CJ Mews, Cantor Fitzgerald - Optionality on leading edge capacity. All in 18A, if delayed, what negotiating capacity in 26/27. If 18A is more successful, how aggressive of adding capacity in Arizona.
A: We have optionality in the portfolio. More resilience in the supply chain. We value our relationship with TSMC, and the product portfolio is set up nicely. 18A capacity, shell ahead and investments, have lots of flexibility if market conditions to scale, or if foundry needs to scale as well. Margin stacking, every wafer we bring home brings the margin structure better to Intel. 18A, 18AP, 14A - aggressive roadmap, advanced packaging, finding more momentum in product and foundry. Benefits of long term strategy.
Q: Consensus guide for Q4 - seasonality into Q1?
Dave: No guidance for Q1. The average for Q1 is 8-10%, more color next quarter.
Pat: Hard to say, geopolitical factors for the world, no wisdom beyond that. Managing the business to a cost structure. We have lots of flexibility to scale up as necessary. Overall there’s a lot of uncertainty in the market, so strategy to deal with those. We’re committed to controlling what we can.
Q: Vivek Arya, BoA - 2025, one point trend of 3-5% top line growth? On GM, Dave you said muted expansion in 2H, do you mean 39.5% like Q4. You also mentioned headwinds in Q4 2025, getting confused?
Dave: On revenue growth, not guiding for 2025. We are managing the business, 2-3-5% growth rate over time and if things are better than that, it’s good fall through. Every Q will have unique aspects to it, only that 39.5% was a clean quarter and good proxy for 2025 calculations. More headwind in 2H vs 1H given Lunar Lake coming with more volume for 2025. That starts to improve with Panther as that becomes meaningful.
Q: Pat what does the future look like for DCAI if there’s no AI product? CPU centric, might be commoditized, what’s Intel AI strategy?
A: Three perspectives. CPU plays an increasing role in DCAI compute. Head nodes are an area of strength for Xeon. In enterprise, we play more prominent, DB, placing, refinement, more tuned to CPU. CPU plus accelerator playing there has a long life.
Second, Gaudi 3 is good. Seeing good early interest in customers, eg IBM win, good pipeline of activities. Xeon plus Gaudi.
Third, Also x86 advisory council. Seeing extraordinary interest in the industry. Front footed with x86, but also AI use cases, and industry is interested in us expanding world's greatest architecture of all time.
Q: Aaron Rakers, Wells Fargo - On foundry, as you ramp 18A, given the pipeline of the design wins. How large is your external at $4.4b this Q, was 700m last quarter. What’s the inflection in 25/26?
A: External foundry will be modest of foundry for next couple years. Foundry revenues will be dominated by internal. We’ll ramp through the decade. Objectives by EoD of 15b of external revenue. Periodic updates on lifetime deal value in foundry. Indicators of new design wins and new customers. Reported numbers on IF will be dominated by Intel products. Nice QoQ growth in our foundry external business. Advanced packaging was profitable in Q3.
Q: On DCAI, continued debate on recapturing share in x86. Granite rapids, clearwater, how do you recapture that business
A: Our goal is to stabilize our position and grow. This was a solid quarter on execution. Xeon 6 fully shipping, important milestones. AWS deal is a nice deal, expanding Xeon and customization. First hold share, then gain share. Strength for Xeon, AI head nodes, AI use cases, very good perf on GNR for AI. In future, being more competitive in base performance and perf/core/watt. Other Rapids shortly going to fab. Overall, that is a good pathway. For AI, all the energy has been in training, and cloud training. But training is making the model, not using it, we have to focus on how we use them - inference, retrain, DB embeddings, RAG. Those are CPU centric. The strength of CPU in AI is something we see the market coming towards us.
Q: Sri, Raymond James - Packaging. 18A, EMIB, Foveros. Given the tightness in packaging, I expected more interest in Intel packaging. Why are we not seeing that? Are there nuances?
A: We do see significant interest. Moving the supply chain is complex, and takes qualification. Designs often start in CoWoS, then look at Foveros. We see good momentum in that area. It’s not as revenue producing as wafers, but the pipeline is very strong. QoQ driven by packaging, most external revenue into 2025 in external foundry will be packaging. Healthy margins for that as well. Progress we’ve seen, design wins we’ve included into LDV, all give great optimization for the long term.
Q: On GMs, PC. Wafer outsourcing, packaging of memory, both depressive. How much of an impact that memory - helps ASPs, but hurts GM? Architecturally, why combine memory? Is this ongoing or one off?
A: It has an impact - Lunar Lake GMs. At one third volume, then we recognized AI PC and how good the part is, so we pushed volume, that’s pushing margins.
LNL was initially a niche product for performance and battery life. With AIPC, it moved to high-volume. We’re not talking about 50-100m units, but a meaningful increase. A bigger margin implication for the company overall. But pleased to have the option to scale LNL due to momentum around AIPC. Volume product in volume industry, you don’t want volume memory going through that channel. It’s a one-off with Lunar, not in Panther or Nova, build those with mem off packaging. But volume mem will be off packaging. It’s a great product, happy to have it in the portfolio with the AI PC category.
Q: Chris Casso, Wolf Research - CapEx, OpEx in 2025. Very clear what the plans are, and FCF. How much flex in those numbers in 2025 and 2026, investing in new nodes for changing conditions?
A: On $17.5b Opex, that’s relatively firm, we’ve right-sized the investments. CapEx - three components - investments for process, shell-ahead, capacity. Always going to make investments towards the process. Shell-ahead, we will, but we’ve caught up and now more measured. For capacity becomes the flex - tooling of shells, modulating that on demand. There is flex, as we progress in 2025 into 2026, our goal is to generate FCF for 2025 and onwards. Managing Capex accordingly. Another variable is offsets - aggressively pursuing offsets, eg SCIP. More meaningful impact from investment tax credit, that’s already baked in.
Pat: Now we have an EUV fleet, we have more flexibility. Intel 3, 18A, 14A - more building to overall capacity requirements rather than rapid 5N4Y. Now we’ve finishing 5N4Y, we’ll have a more regular cadence now. That’s more flexibility. Singular expectation is High-NA, the equipment bases otherwise are the same. The TD development has flexibility.
Q: Better together - explain the rationale of why that’s the case. Various opinions on that. If Intel will be willing, strategic what the current plan is?
A: Comfortable to the subsidiary model, gives three things. Greater operational and integrity as we do that separation, communicate that more clearly. Potential to fund and manage capital requirements of foundry. Mass volume will come from Intel products, customer zero and integrations, the benefits from the cash flow and managing balance sheet are beneficial. Distinct but better together.
Q: Joe Moore, Morgan Stanley - Narrowing the product focus and prioritizing x86. Are there areas you’re investing less in?
A: A lot of detail behind that. We created complexity in the server line, E, P, across sockets. That complexity was appropriate when business was growing, but we’ve taken steps to simplify the line, fewer SKUs, efficiencies. In client - simplifying the roadmap, fewer SKUs. How we are managing graphics, bigger integrated graphics, less need for discrete graphics. CCG and Edge coming together. Leveraging our core investments. Getting back to fighting shape.
Q: Prioritizing x86, does that shift the IDM 2.0 model. Arms length relationship, is there a focus?
A: Clearly external foundry requires a different view on how to manage that business. Wafers and cash flow comes from internal. Leading edge, western based, foundry model. Focused on making that successful. Years behind, 5N4Y to be global leaders. Still work to do, well on our way to completing most seminal restructuring ever. It’s been critical, bringing us to a point for the capacity to drive long term shareholder return.
Insightful summary. All I would add is that Pat also said they had secured two new “compute-centric” 18A customers. I assume one of these will be Nvidia for its upcoming ARM based laptop CPU. Not sure who the other one will be though. Any thoughts?
Really enjoyed your takes in "Final Thoughts" 👍🏼