Pat Gelsinger is the CEO of Intel. I’ve been excited to have this conversation for a very long time — ever since Pat took over as CEO a little over a year and a half ago. Intel is a very important company with a huge series of challenges in front of it. It’s still the largest chip manufacturer by revenue and makes more chips than any other company in the United States.
The Intel Pat took over was struggling and losing ground in a variety of markets. But in the past year and a half, Pat restructured the company, turned over almost all of its leadership positions, opened a new line of business that competes with Taiwan Semiconductor Manufacturing Company (TSMC) and makes chips for other companies including Intel’s competitors, and generally tried to reset Intel’s famous engineering culture.
He’s also done a lot of handshaking with politicians. Gelsinger lobbied hard for the US to pass the Chips and Science Act, which will put $52 billion in subsidies toward chip manufacturing in the United States — a lot of which will go toward Intel. He’s expanding Intel’s chip fab in Arizona and building new fabs in Ohio and Germany.
In addition to that, he’s shipping products: we spoke about Intel’s new GPUs and the roadmap he’s been planning for the future.
Pat Gelsinger is the CEO of Intel. Welcome to Decoder.
It’s my pleasure to join you today, Nilay.
There is a lot to talk about. You just announced some new processors and GPUs at your developer conference. There is the CHIPS Act, which passed back in July. You are also building some new fabs in the United States, most notably in Ohio. Let’s start at the beginning. You have been back at Intel for a little over 18 months now. You had been at the company for a long time before you left, and I don’t think that was the happiest departure for you. You were then the CEO of VMware. Why did you come back to Intel?
Listen to Decoder, a show hosted by The Verge’s Nilay Patel about big ideas — and other problems. Subscribe here!
I joke that I went through puberty at Intel since I started there so young. From age 18, I grew up at the feet of Andy Grove, Robert Noyce, and Gordon Moore — who I call the Trinity of Intel — and I had deep interactions with all of them. When my wife and I had been approached about joining the board of directors of Intel, we said, “Okay.” We were interviewing and meeting the board, then, just before Christmas, they asked if I would consider being the CEO. I joke that it ruined my Christmas because, all of a sudden, there were all these emotions. “Should we come back? What would it take to turn the company around?” I had written 30 years earlier that I wanted to be the CEO of the company, and I thought that vision was dead many years ago.
Fundamentally, the decision boiled down to three things. First was restoring the iconic Intel, the company that I grew up in. Second was believing in Intel being foundational for the technology industry. A healthy industry is part and parcel to a healthy technology future. Third was rebuilding the Western supply chains, fabs, and all the things associated with chips, EU chips, and balance resilience supply chains. Oil reserves defined geopolitics for the last five decades, but technology is more important for the next five decades as every aspect of humanity becomes digital. Everything digital runs on semiconductors, so this is really important to the world. It was really for those three reasons that Linda and I took the dive and said, “We’re going to come back and turn it around.” I say “Linda and I” because anytime you take on an assignment like this, you better have your spouse or significant other aligned with you as well.
There is a lot to your answer. There is restoring the iconic Intel, which is really about the engineering culture of the company, and I definitely want to dig into that. Then there’s chips and supply chains, which is really about geopolitics. There is no “United States Strategic National Reserve of chips” — there’s kind of just Intel. That is the answer for a lot of what has been going on lately. As I’ve watched you doing this job, I think there is a lot of pressure and a lot of redefinition for what it means to be the CEO of Intel.
Let’s start with the engineering side. When you came back, you gave a big talk, which did go out to the public, but it was very obvious the audience was Intel’s employees and engineers. It was a storming talk about resetting the engineering culture, letting go of some baggage, and reinventing the company. You have now been in charge for a little over 18 months. Were your instincts about that culture correct? Was the biggest problem resetting the engineering culture?
It really boils down to two things. First is the culture, the methodologies, and the processes that we were using. Twenty years ago, we had established what was called the tick-tock methodology, a risk-managed cadence of execution, which had been undone when I was gone. It was like, “Why’d we stop? That was super successful. Let’s do it again. Put it back in place.” It’s how we have consistent life cycle management across the company and how we have quality A step tape-outs (the end of the design cycle), move bugs and software emulation. When an A zero goes to fab, there is no reason it can’t go to production. This kind of discipline is data-driven. We had stopped doing OKRs (objectives and key results). People wrote books about that as other companies adopted it. We invented it at Intel, but we stopped doing it.
I think most listeners of Decoder think that Google invented OKRs, by the way.
Well, guess what? It wasn’t. It was Intel. Go back and read Andy Grove’s book High Output Management about where it first began. A lot of it was, “What’s old is new again.” Engineering, the Valley, so much of the technology has obviously moved on. It’s not like we are just repeating what we had before. Many of the core concepts need to be brought forward, modernized, and reestablished up and down the organization. The other aspect is that we inherited a lot of products in flight. Much has been said about Sapphire Rapids (the codename for Intel’s 4th Generation Xeon server processors) and the difficulties we had in getting that out. That project was started five years ago, so it’s in-flight. I can’t just reset the methodology of a product that began five years ago.
It didn’t start well, either. It had way too much complexity in it, with three major new systems, or interfaces, in that design — a new CXL (Compute Express Link), a new PCI Gen 5 (Peripheral Component Interconnect Express), and a new DDR5 (RAM) — and there were no backups on any of them. It was like, “How did we take on so much risk and technical challenge in one program?” I can’t undo those products. We have to gut it out and get those finished, even as we are launching the new products on new methodologies with new approaches. So I said, “We have to flush the pipe of those projects. We have to finish them, get them done, and get the quality culture back.” We do this even as we are starting new projects with the disciplined culture that we expect and are going to be executing in the future.
“Not only do we have to rebuild the internal culture and execution, but we also have to rebuild our customer’s confidence.”
In the middle of that, not only do we have to rebuild the internal culture and execution, but we also have to rebuild our customer’s confidence. In some cases, we have earned our customers’ distrust. We have not executed the quality, the capacity, and the performance that they would expect from Intel’s products.
We are re-earning their confidence that they can build their business on us. We have made so much progress in a year and a half, but there is still a lot of work to do. If you ask our top 10 customers, they would say, “Oh, there is a defined shift. They are back to focusing on quality, on engineering, and on giving us transparency on schedules. We are rebuilding our confidence.” That is a lot of work to do for what was largely a lost decade in that respect. That is well underway. It’s not finished, but every day, we are gaining more momentum to be that engineering culture that people can build their businesses on, to have confidence and our competitors’ fear. When Intel says something, we are going to get it done.
So, was it the engineering culture that you needed to reset? Was that instinct correct? Was there anything else that you found in the last year and a half that was a bigger surprise to you?
I’d say it was a collection of things, like OKRs, discipline, performance orientation, and being data-driven. I summarize it as an engineering-centric culture, which is what we needed to reestablish inside of the company. We just had our Innovation Conference, and my shirt said, “Bringing the geek back,” but it was spelled in ASCII binary. Everything about us is that engineering technology centricity of what made Intel great, what made it famous, and what made it this iconic technology driver — the company that was going to rebuild Moore’s law and be committed to it. I like to say, “Moore’s law ain’t dead until the periodic table is exhausted.” We are going to keep driving physics, inventing molecules, and finding new ways to advance the state of the art of technology.
This is a key point. You mentioned the tick-tock cycle, and you said the words “risk-managed,” which I want to push on a little bit. The tick-tock cycle, very famously, was where Intel would invent a new processor design and move it to a smaller process node, increasing its efficiency. Then you would have a new process design, and you would move it to an even smaller node. This was a very easy way to consistently increase performance of processors over time. Implicit in that is Moore’s law — the idea that transistor density is going to double every 18 months — and that Intel is going to invent some new process technology to make that happen on that cadence and de-risk it because that processor design is proven. That has not happened for Intel in the past decade, though — this “lost decade.” The packaging improvements, the process node improvements have not happened for Intel in that cadence. They have happened elsewhere, at TSMC and at Samsung. Part of it was because Intel didn’t take enough of a risk on some of the process technologies.
“We took too much risk, and we failed.”
It’s actually the opposite. We took too much risk, and we failed. It is a managed risk methodology in that sense. When I came back, we had said we were going to do five nodes in four years, which is an audacious plan to lay out. Even inside of the process of technology development now, it is a very risk-managed, cadence-focused delivery, where we have essentially parallelized the 10-7 team (10nm - Intel 7 processes) from the 4-3 team (Intel 4 - Intel 3 processes) from the 20A and 18A team. These teams may build and leverage each other, but they are parallel, each becoming its own tick-tock execution environment. Many years ago, I was running a project for Inel, and Andy Grove joked at the finish of it, “I gave Pat an unlimited budget, and he exceeded it.” Similarly I went to our head of technology development when I came back, Dr. Ann Kelleher, and I said, “You have an unlimited budget, and you are going to deliver five nodes in four years. We are going to get back to unquestioned process leadership.” Some look at that and say, “Hey, that’s just nuts! Nobody has ever achieved that cadence in the history of the technology industry.”
There are three things that we have done toward that goal. First is that we have put in the teams, the capital, etc., because a lot of innovating at transistors is that you run more wafers, run more experiments, and have more ideas going on. We gave them the capital and the R&D to go forward. Secondly, we created this model of parallel execution inside of the TD teams so that the 20A and 18A team is now operating independently of the 4-3 team. Third, we had a whole bevy of component research ideas — inventions that were in our period where we stumbled. We had a whole bunch of ideas that we could quickly put into production. This was all complemented by Intel becoming somewhat arrogant. We were like, “Hey, just drop the equipment off at the front lobby, and we will take care of running it.”
The industry had moved rapidly, and we needed to learn from those capabilities and bring those experiences. We are now leaning into those relationships at scale to garner their best learnings into our development. For instance, I regularly meet with the CEO and the CTO of ASML (also known as ASML Holding), the EUV (extreme ultraviolet lithography) equipment provider. We have committed in a big way to EUV. That was a fundamental mistake that Intel made.
Right. This is what I was saying. I think it wasn’t a risky enough bet on EUV.
We were betting against it. We had taken a lot of risk in Intel 10 when we were like, “Hey, we don’t need EUV. We will go to advanced quad patterning of the lithography.” We were doing other things to avoid needing EUV, and those things just weren’t panning out. It might have been a good decision when we did it, but as those things slipped, we were on the wrong side of EUV. TSMC grabbed EUV because of that. By the way, Intel drove the creation of it. How did we not monetize and leverage something that we created? At a minimum, we should have had a parallel program on EUV that said, “If we get this wrong… If we get quad patterning or the other techniques we’re doing in this self-aligning wrong…” We should have had a program for that, but we didn’t. We were betting against it. How stupid could we be?
So now, we have leaned into the EUV relationship very aggressively, and that is going quite well for us. Intel 4-3, our first volume deployments of EUV, are going extremely well currently. It’s seminal to our 20A, 18A programs, which are going very well. I joked with Peter Winick, “Send all your best engineers to my Oregon site.” He came back and said, “Well, Taiwan is closed. I can’t send any engineers there because of COVID, so that’s easy.” We had a couple of good jokes. We are learning from what they have done with TSMC and Samsung. We are benefiting from that.
We are getting back to what we do well. For instance, we have the 20A and 18A, and we have the RibbonFET — a new transistor technology we have developed. Intel has done every major new transistor architecture at scale in the last 30 years. We are on it with our PowerVia and Backside Power Delivery. We are embracing EUV, and we are leveraging a sustained leadership that we have had for decades in advanced packaging. These together are what give us confidence that Moore’s law is alive and well for the future. We have all the work going on beyond 18A that will allow us to keep saying that through the end of the decade and so on.
My joke about Decoder is that it is fundamentally a podcast about org charts because I think CEOs mostly mess with the org charts. I’m looking at some quotes here. You said at the Evercore Conference in September that 70 percent of the leadership team at Intel is either new to the company or new to their role. That is turnover that you instituted. Why did you decide to turn over that many folks?
When I came back to the company, it needed to be refreshed. We had people that were good leaders in jobs that they, and I quote, “weren’t qualified for.” They had the experience level, and they were good leaders, but they weren’t in the roles that fit their skill sets. We wanted to bring people in that gave us skills and business capabilities in the areas that we needed. Onstage today is Greg Lavender, our CTO, who I worked with closely for a number of years. He has been resoundingly and enthusiastically accepted as a key leader at the company to lead our software and CTO efforts. We also have Nick McKeown, who has taught everybody in networking for the last 25 years and is now running networking for us. These are super quality leaders. Our general counsel has been GC at four other firms before coming to Intel. Dave Zinsner, my CFO, was at Micron before coming here. He lives and breathes the semiconductor industry. Our government affairs leader, who was seminal in getting the CHIPS Act across the line, was under the Secretary of Commerce in the Obama administration. He knows how to run government affairs. These are people who I can have confidence in that they can lead this incredible turnaround.
With every one of these different aspects of the company, I want people that I can put in the chair that know what to do to go a certain direction. We have so much work to get done, both for Intel and the industry. We don’t have time to say, “Let’s do some on-the-job training.” We need to go, and we need to go quickly. We reorganized the company in a very specific way, around six business units. We are going to have empowered business units, strong functional units, and a single voice to the customer with our sales. Christoph Schell, who was at HP before coming here, sat on the other side of the table. He knows what it’s like to be sold to as a customer by Intel — the good, the bad, and the ugly — so he’s a super qualified leader for that. One by one, we are gathering people who are unquestionably qualified for their jobs. In many cases, I don’t think there is a better person on the planet for the jobs that are part of the team. There is a clear, simple organizational structure and a lot of accountability and expectation for what they are going to achieve. My job is to essentially knit them together as a team.
You said six business units, but then you have one face to the customer. Are you organized in divisions or as product lines?
There are six businesses. There are the core four: CCG, our client business; data center and AI, our data center business; accelerated computing and graphics; and networking and edge. Then the other two are Mobileye, which we hope to IPO, and the foundry business, which is really quite distinct and unique. Those are my six business units. Then there are the key functional units — manufacturing, technology development, software and technology, and design engineering — which support all of those six business units.
It is one face to the customer. When we show up at Dell, there is one face saying, “Hey, they are buying our accelerated computing, our clients, our servers, our networking.” When I show up at Amazon, there is one face saying, “Hey, they are becoming a foundry customer; they are buying our data center products and accelerated computing.” One face to the customer. Those business units, which are accountable for those P&Ls, are essentially reportable segments. The financial community gets to see how we are doing in clients, data center, etc. I’m holding them accountable, not for spending targets but for operating income targets. You run it like a business you are responsible for. If they say, “How much can I spend, Pat?” I say, “I don’t know. Tell me what your business plan is for next year. I’m only going to give you operating income and operating margin targets for your business that you are running.”
Every structure has inherent tradeoffs in it. It sounds like you kind of cleaned the slate and drew a new structure for Intel. In your mind, what are the tradeoffs of this structure?
When you look at a structure like ours, I have had people say, “Hey Pat, why are you launching these new business areas like accelerated computing and graphics? Why don’t you just finish fixing the core? Just go and get client, data center, and the process technology done. Why are you worrying about that new area? Foundry sounds great, but why don’t you wait to launch that until you get the process technology leadership back in place?” Those would be some of the criticisms that have come up, and I’d say they are fair. It’s not my strategy, though. I want to bring accountability. We have had a large networking business, but it was never treated like a business; it was always submerged. We were doing integrated graphics for two decades, but we were never getting a monetization. Then there is independent software vendor (ISV) presence, and GPUs have become a big computing platform in their own right. We were never treating it that way, even though we were spending most of the R&D to satisfy that segment.
One of the seminal decisions that I made was to become a foundry. I joke that there are three types of semiconductor manufacturing companies: big, niche, or dead. We are way too big to not be big. Because of that, we have to be at scale. There are only three companies that can do advanced processing on the planet, and those are TSMC, Samsung, and Intel. Before I showed up, none of the foundries were Western foundries. We said we were going to build and balance this global supply chain, and we were going to open the doors to our fabs to allow us to be unquestionably at scale, to be that foundry partner for our customers. So much semiconductor innovation and design innovation is occurring outside of the Intel fabs. “Why aren’t we enabling that ecosystem to work on us?” Our customers want us to take advanced packaging technologies and make them available for some of their chiplets as we’re going forward.
Thus, we have announced this idea that our foundry is going to become a systems foundry. What does that mean? We said a systems foundry has four components. First is that you have to be a good wafer foundry. I am so committed that we are going to be a leadership process technology company again. Secondly, we are going to enable our advanced packaging so people can put the pieces together. On top of that, we are going to have a rich ecosystem of standards-based chiplets. Third, we are driving the UCIe chiplet architecture so that you, Nilay, could walk in and say, “Hey, I have a couple of components of TSMC. I want to combine that with a couple of your chiplets, Intel. I need this power component from TI and this IO device from GlobalFoundries. Since you have the best packaging, I would like you to assemble them to be my supply chain.” The answer is “Yes” because the rack has become a system, and the system is becoming an advanced package, and we are going to be enabling that ecosystem. The fourth component is the foundational software pieces that go against it. Obviously, with our one API initiative, we have more software in that than anybody else in the industry, and we are enabling that against these core platform components. We said, “We are going to be a great wafer foundry, but we are going to usher in the systems foundry era as well.”
So that is the path that we have laid out. Generally, it feels pretty good. We are getting more and more resonance with that approach. At the Innovation Conference yesterday, I had video testimonials of TSMC and Samsung, along with Intel, on the shared initiative around a standardized chip-led architecture.
I want to push on that. That is a big decision that also directly plays into the geopolitical nature of your role, which we talked about at the beginning. Let’s just abstract into the classic Decoder question for one second. How do you make decisions? What is your framework?
We have laid out a strategy. To me, that is not a one-year, one-quarter strategy; this is a decade-long strategy that we have laid out. We have called it IDM 2.0, where we are going to rebuild manufacturing, engineering, culture process, technology, etc.
I understand that it’s Intel’s strategy. I’m asking you, Pat Gelsinger, how do you make decisions?
My decision-making is, first and foremost, to deliver on that strategy. I am not making decisions about the next quarter — I am making decisions against that decade-long strategy. That is the high-order bit that I have laid for the company. That is my north star as I’m measuring these tradeoffs and asking, “Do I do this or that?” I am going to be super consistent on that. Secondly, in setting that north star, was building the agreement with the board of directors. This is an expensive journey. Please understand we are going to spend tens of billions of capital to go execute that journey. So I need to make sure I have their alignment, and I have to keep checking in with them. Part of my decision-making is to make sure the board is on board if I could be against that direction.
Third, I am a data-driven guy. If you walk into a meeting with me and you don’t have data, just leave. Don’t even bother showing up. This is a very Grove-esque principle from hearkening to Andy Grove. I want data, and I want your opinion. It is not a consensus process, but I largely want to build agreement in that direction across my leadership team. The next principle is that we are going to have open debates and open discussions. “It’s all in the meeting, so let’s talk about it.” When we leave the meeting, we are going to have had a real conversation that enables people to do their jobs, having participated in the decision-making process. That is sort of how I think about and make decisions, in the context of how I treat the board and how I work with my leadership team. The last piece might be that I’m a fairly detailed guy.
I can tell. Just from this conversation, I can tell.
“You can’t make decisions on these things just by looking through a spreadsheet.”
Hey, bring the engineers in the room. I want to know the real tradeoffs down in the bowels of the company. The more that I understand how these things play out, the more competent it enables me to be in the leadership team as we make decisions. I believe technology companies should be led by technologists. You can’t make decisions on these things just by looking through a spreadsheet. They are much more intuitive at the tradeoffs of financial technology, products, and markets.
I really want to talk about Foundry Services and about your new chip, but let me ask you a philosophical question first. You said you are really data-driven, but chips are a long-term business, and you just said you have to make some intuitive decisions. My feeling is that data can really only tell you about the past. If you actually want to invent the future, you have to go invent it based on your gut. How does that play out inside of Intel for you as you think about chips, which are a really long-term business?
The past and the data needs to inform the future. Let’s just make an example. We are working on 18A, and I have regular data updates. When you are developing a new process technology, you are fighting defect density and transistor performance. You are always pushing the envelope upward as you try to converge to the process technology target. History has shown us that we can get so much percentage, per quarter improvements, on those two vectors.
The actual transistor design that is going into 18A, we have never done before. This is much more intuitive. There are new structures, new sciences, new chemicals, and new physics going on. In our history, though, we have only ever been able to do 5 percent per quarter improvements. “Team, if you came in at 4 percent this quarter, I want to know about it. Oh, 6 percent this quarter? We’re looking pretty good, baby!” That is where the data of history guides the intuitive invention of tomorrow. This applies across many domains. For instance, when we are doing a major new microprocessor design, we know that “At this point in the design cycle, we need to be within 15 percent of our power performance target because we have never been able to get more than that in the last six months of the design. So if you’re not there, then I have a problem.” Sometimes it’s just as new — you don’t know.
Some of these things, like in our graphics development, we thought we could scale the integrated software stack into discrete performance levels. We were wrong. They weren’t plumbed with the bandwidth, the performance, and the latency that we needed. So we stumbled out of the date, largely because we had assumed the first generation of our software could be better leveraged than it was. I’ll tell you what: with the second generation, we ain’t making that mistake! We may make a new mistake, but we are not going to make that one again.
Let’s talk about Intel Foundry Services and your role as a geopolitical leader now since there’s kind of no other way to describe it. You came back to Intel, and you have done a lot to reform the structure of the company, the culture of the company, and the engineering leadership of the company. I’ve watched you over the past year and a half, and boy, have you talked to a lot of politicians, been at a lot of events with the president, and lobbied hard for the CHIPS and Science Act.
The act passed in July, and it’s $52 billion for chip manufacturing in the United States. There are not a lot of companies that make chips, so Intel is going to receive a lot of that. You have to do Foundry Services just to live up to the policy nature of the bill, which is more chips you made in the United States, not just Intel chips. Are you seeing that role for Intel? Are you part of the geopolitical drama that is playing out over at CHIPS and particularly TSMC, which is in Taiwan?
It’s part of what we signed up for and what I signed up for when I came back to it. Though, I view it from a somewhat different lens than you have suggested here. We have used the phrasing “geographically balanced and resilient supply chains.” We say, “We shouldn’t have unique, extreme, acute dependencies on individual suppliers for something that important.” It’s not like TSMC isn’t going to be part of the semiconductor supply chain a decade or two from now. Far from it. They are critical and built a great ecosystem, but we shouldn’t have acute dependencies on individual geos, locations, etc., for something that is so important to the world. It’s like having one oil reserve in one country. No! We need to have a multitude of oil locations and refineries and of different energy sources over time because energy is foundational to every aspect of human progress.
Similarly, if you go back to my earlier comment, technology and the digitization of everything are affecting every aspect of human existence. How we connect right now, how you live in community, your social networks, your healthcare for the future — every aspect of your existence is dependent on semiconductors. So let’s make sure we have a globally balanced, resilient supply chain for semiconductors and the economic benefits of that. Why are we moving these manufacturing jobs uniformly offshore? These are good jobs. We want them here for the economy.
Look at national security. Of course, we want our supply chains for national security, and that is a statement if you are in Asia, the US, or Europe. These things are foundational to the nature of the role that nations have across the globe. We have stepped into that in a pretty aggressive manner since I’ve been back. It’s the right thing. Yes, this is good for Intel — and I am not hesitant to say that — but it is the right thing. I have interacted with political leaders in the US and in Europe. One of the conversations with a European political leader was that we should only get involved where there is market failure. “Hey, we used to be 80 percent, now we are 20 percent. Europe used to be 44 percent, now at 9 percent and predicted to go to 5 percent. If that is not market failure, tell me what is!”
For something so foundational to every aspect of the economy, human existence, and national defense, of course this is important. We have leaned into something and said, “This is really important to the world.” As the leading technology supplier in the US, of course it is going to help us. We should be the national champion. We are half of the semiconductor manufacturing in the industry. If we are not healthy, then there is no hope that we can turn this around.
You said “national champion,” so this is my follow-up question to that. I talk to other CEOs and executives from huge, US-based tech companies, and they think of themselves as global citizens. They are huge global companies that have massive footprints around the world — some of them have massive footprints in China in particular. They don’t like the idea that they are national champions of the United States. You even said it yourself before I asked you about it. Do you think of Intel as the national company of the United States?
We manufacture globally, and we are distributed across the world. We have a presence in Europe and a presence in Asia. We are a global citizen in that sense, and we are quite proud of that. At the same time, when you look at the US, yes. I want the US economy to look at us and say, “Wow, this is really important to us because we only have one of these in the United States.” We are also deeply invested in Europe. There is no other leading-edge manufacturer in Europe other than us.
It is really important to the future because leading-edge manufacturing is going to be essential for AI, for high-performance computing, and for advanced communications. I believe that we need to be both. We need to be proper global citizens. I personally chair the US Chamber of Commerce to China as well. Why? Because that relationship is so important. We are also deeply invested in our partnership in Europe. We need a globally resilient supply chain. I believe that if we create a manufacturing network that has supply chains balanced across the globe, it enables world peace.
Since we have acute dependency in one area of the world, minor provocations become major responses. They have to at that level. So it all makes sense when you tie those pieces together. I need to make sure this is a great business. I am very committed to having the majority of our manufacturing in the US and in Europe, and we are absolutely committed to being a proper partner in the global supply chains of the world.
You are expanding in Arizona, and you just had a groundbreaking in Ohio that is a major initiative. Neither of those is going to be done for years. How long until the majority of your manufacturing is done in the United States and Europe?
The majority of manufacturing is already in the US and Europe. We have a large facility in Arizona, a large facility in Oregon, and a smaller one in New Mexico. Then we have a major fab in Israel and a major fab in Ireland. We have some in Malaysia, Vietnam, and China. However, the majority of our manufacturing footprint is already in the US. What we don’t do for the world is foundry.
At the Ohio site, I said, “That fab module right there, I want to put your logo on it, and I want to say it was made here.”
We had a number of CEOs from fabless companies at the Ohio site, and I said, “That fab module right there, I want to put your logo on it, and I want to say it was made here.”
You’re going to put an AMD logo on the side of an Intel fab?
Hey, if they choose to manufacture with us, I will be thrilled to do that. It’s the right thing at that level. These are large investments, and they are important for technology — they spawn technical communities. We opened the new module at our Oregon facility, and we had Senator Wyden, Senator Merkley, and Governor Brown there. They gave me a plaque showing the location before Intel and the location after Intel 40 years later. We named it the Moore Center, and it was a super emotional moment for me to be able to honor Gordon Moore that way. It was a field before; now, it is a tech metropolis. It has Google, Microsoft, and Intel in this big center location. It has ASML, applied materials, air products, and so on.
It is just this exploding metropolis of technology. It is the right thing for these communities. There is such a pent-up enthusiasm for Ohio as the silicon heartland for exactly that reason. It’s not just about what I’m going to build there. It’s going to be big. We are going to manufacture for lots of companies, and I am excited about it. It draws all of these other technology companies because we are building the water, the electricity, the roads, and the tech pipeline in those areas.
I have heard a story like this before. It was in my home state of Wisconsin with Foxconn. I was just there, actually, and there is nothing. I’m sure your people prepped you for this question. Are you actually building this factory?
It was fun when we were doing the groundbreaking. It was like we were the Secret Service saying, “Okay, send the earth-movers in now.” I mean, we are moving dirt.
Foxconn also moved a lot of dirt, man. Are you actually building a factory?
Oh yeah, absolutely.
When is it going to come online?
It takes about four years for one of these factories to come online. Our objective right now is to be online and start producing by 2026. My bigger objective isn’t just that I build the first two modules — which was what I did groundbreaking for a couple of weeks ago — but that as this entire movement and our Foundry Services build that I then go build modules three, four, five, and six over time. That is the vision that we have laid out. Yes, building is underway, and we are pretty excited to get it in place.
What process node are you building?
It will be the most advanced technology nodes, and we are building these as EUV-capable fabs. There is also 20A, 18A, and the ones that come beyond that. We may choose to have some capacity there for some of the older nodes as well, like Intel 3, and some of the automotive products that we will manufacture. I expect that this becomes a major manufacturing location for a variety of nodes, and it will at least be capable of manufacturing our most advanced technologies. Obviously, when you build a new node at a new factory like this, you want to equip it with all of your latest and greatest insights for your next-generation process technology. In this case, building a site that is EUV-capable is a big deal.
I have to ask you a hard question here. The CHIPS Act was passed in July for $52 billion, and it was really tied to the Ohio plant — right after it was announced, we had the groundbreaking. Biden was there, and it was a whole thing. At the end of July, Intel announced a 5 percent dividend increase to shareholders and a $4 billion reduction in capital expenditures. This is where there is criticism that the CHIPS Act was a huge subsidy for Intel and that the American taxpayers are now funding a dividend to the Intel shareholder. That is what it feels like. Shouldn’t the CapEx have gone up and the dividend have gone down?
There are two things there. First is that we actually only slightly changed our gross CapEx for the year. You might recall that we were also able to announce our semiconductor co-investment program. We talk about gross CapEx versus net CapEx. We were able to bring our net CapEx down, which is essentially what we carry on our balance sheet. We are having success for some of our capital offsets. When we initially set our net CapEx targets for the year, we didn’t build into it the success that we have had in establishing this Brookfield.
We have actually only changed the gross CapEx, and most of that was driven by equipment push-outs, where the equipment supply chain has drifted out. From my chair, I have hardly changed my capital investment policies at all this year. Now some of it is that you are always working on these things, supply and balance, and so on. With gross CapEx, we have communicated that. A couple of press people continue to say, “Hey, you took your net CapEx down.” It’s like what they really care about from this conversation is what the gross CapEx is, and that had very minor change — and the only reason it changed at all was because of equipment delays.
On the dividend side, we have continued to have a dividend policy that hasn’t changed. We have been running this policy of having a quarterly dividend and regular increases in that dividend level. To some degree, the news would have been about me not doing a dividend since we have been doing it for so many years. I understand some of the critiques that have come back to us as a result, but neither of them have any foundation.
I’m not allowed to invest in tech stocks, so I can only cosplay as a tech investor here. If I’m an institutional holder of Intel, getting a tiny bit of cash back every quarter as part of my dividend versus seeing the company invest in the long-term roadmap, I would rather see the investment. Why couldn’t you make the sale that “we are actually going to reduce the dividend and push out into the long-term plan,” which you have said is your north star?
When I came back into the job, we told investors that we are going on an aggressive capital build-out campaign and that we are going to help with capital offsets to make those investments. The reason for capital offsets is that if I was building these facilities in Asia, they would be substantially cheaper than if I was building them in the US or Europe. We have been very clear in saying, “If I am going to build at this level, they have to be competitive. I need capital and tax offsets to make sure they are competitive.” We are also going to find creative ways to fund that level of capital. I do $5 billion of dividends per year against the $30 billion gross CapEx spend.
These are on different scales at that level. That is what our semiconductor co-investment program is about. That is where Brookfield basically came in, and they are helping to capitalize $15 billion of the Arizona site. It is just at a different scale of investment. We have told investors that this is a multi-year journey.
“People have commented that the Intel stock hasn’t performed very well. We know that investors believe in our vision, but it is going to take a while to get there.”
As you know, people have commented that the Intel stock hasn’t performed very well. We know that investors believe in our vision, but it is going to take a while to get there. The dividend essentially provides some level of continuity of total shareholder return, even over this period that I have dubbed “the saddle.” We just have a couple of years until process leadership, product leadership, and capital is in place.
That is why we have held the dividend policy, but we immediately curtailed any buybacks. We have to invest in the capital going forward. That is how we position it to analysts and investors. I have had numerous conversations with them on this journey. Most of them think, “Okay, that’s the right balance.” They haven’t seen that total shareholder return that they would expect, partially because we are investing super heavily to establish the position that we need for leadership technology and capacity as we go to the second half of the decade.
That is a long-term vision. But you also have products coming out now. You just announced new GPUs and CPUs that are going to carry you through the current moment – GPUs being the highlight here. It took a long time, but the Arc GPUs have finally been announced. As you said, that is a product line you inherited. Performance-wise, they match Nvidia’s RTX 3060s, which came out in 2020. Nvidia also just announced the 4090s.
The 770s are much better than the 3060s. They are going to be very competitive, and we put them at a very aggressive price point at $329. We will have a range of price points associated with them. Those are going to reviewers now. I have one I’m playing with.
What is your game of choice to test out a GPU?
Well, I’m not much of a gamer myself. I actually get my son, who is a big gamer, to test them. But I still have to plug it in, boot it up, load Windows, etc.
Do you build your own PCs?
I have help from IT here because of all the security I need since I happen to be the CEO. Trust me — I’m tinkering regularly. I think I have five laptops at home with different builds, and I have my favorite laptop of the month. I’m actually getting the Unison build that we just announced installed soon. I am definitely a geek’s geek, but I do have a lot of things that pull my time away from being as geeky as I would like.
I feel you. When do you think you’re going to match the performance of the Nvidia 4090s?
We set October 12th for the launch date with the A770s. We will let the reviewers speak as they look at that. We are pretty satisfied with the performance that we are seeing in our labs, and obviously, we put them at an aggressive price point. I think the performance price characteristics we are expecting the reviewers to come back with will be very favorable.
You have also announced the 13th Generation processors.
Well, before we leave this topic, we have also released the GPUs. The data center GPU version, which used to be called Arctic Sound, is now available. So you have the discrete graphics and the data center version of those products, a hardware-based AV1 multi thing. We can do inferencing video processing on simultaneous streams. There are some great demonstrations of that. Overall, that’s the pair. Then there is the high-performance computing with Ponte Vecchio. There is a continuity from integrated to discrete, to data center GPU, to HPC.
We now have the full suite of those available in the fourth quarter of the year. I have been working to get those in place. You have the client leverage between integrated and discrete, you have the data center leverage from discrete to GPU, and then you have the leverage into the high-performance computing. With our Argonne National Labs, we are combining Sapphire with high-bandwidth memory and the GPU. That whole family will be available shortly.
That has been a long time coming, right? That’s a long road, and you’re finally here.
I think it’s a fair assessment that you have to prove that out, that it’s going to compete with Nvidia.
Absolutely. I have a fun little story there for you. When I left Intel 12 years ago, I had taken the data center job at the time, and I had made a list of 10 things I was going to get done. In five years, I finished nine out of 10. The one that I didn’t finish was discrete graphics. So I am now finishing that one that I started 12 years ago. We are getting it done. From here, Gen 2 will get better, Gen 3 will get better, and so on. Software will be more robust, and there will be more ISV enabling and more refinements of the hardware. I am very happy that we now have all elements of the family shipping.
The ISV piece is actually the part I was going to ask about. We have seen over and over again, in GPUs in particular, that the developers are going to support what the consumers have, and the consumers are going to buy what the developers support. You have to kick off that flywheel, where you get people to buy into your drivers, your APIs, etc. How is that going for you?
That was part of why the launches earlier in the year and some of the lower-end graphics weren’t as successful. The software wasn’t as robust, the drivers weren’t cleaned up yet, they weren’t performing, and we didn’t have game compatibility in a number of places. It has been a rough year to get those pieces put in place. Now that we are largely done with that work, I would say the vast majority of games are validated, and drivers are there across both Windows as well as Linux embodiment. So I feel like we’re pretty good.
Part of this week with our Innovation Conference was a very developer-centric set of activities. When you think about this, people have been optimizing for Nvidia and, to a lesser degree, AMD for the last decade. I don’t expect that we are going to show up and have 100 percent compatibility here on day one. It’s just a lot of work in those software stocks. I do think we are now to a point that we have critical mass, we check most of the boxes, and we are getting good responses from developers, but we have to earn our way back into the segment. Again, we have not been here for over a decade, and a lot has happened over that decade. We have to earn our credibility.
The flip side of this is not just in Nvidia but AMD as well. It’s in the two major consoles and the Steam Deck — basically, it is in everything except the Switch. Ryzen chips per dollar seem to be offering more performance, and there is a lot of back and forth between Intel and AMD. Is that just a crown you can take? Or are you saying, “At the end of the day, we are going to do Foundry Services, get that Intel logo on the side of the fab in Ohio, and it’ll be fine”?
Hey, we are here to win in this space. Make no mistake about it. We are going to build the fastest high-performance computers in the world, the fastest GPUs in the world, and the fastest discrete GPUs in the world. We will have continuity to our integrated graphics as well. With increasingly integrated graphics in a heterogeneous architecture where you have multiple chiplets, just wait until you see Meteor Lake next year. These become quite interesting products, somewhat as Apple has shown. I see that continuity, and we are going to be very competitive.
“I want to win Qualcomm’s business, and I want to win Apple’s business.”
At the same time, I’m going to be the foundry for Nvidia. By the way, they need a more resilient supply chain. They need these technologies that we are working on. I don’t know how much of it will win, but I want to win their business. I want to win Qualcomm’s business, and I want to win Apple’s business. We want to be that provider of a choice. Again, we haven’t been in the foundry business. TSMC has perfected that business model for 30 years. They are good at it, and they have an ecosystem around them. I’m not going to displace that, but to earn that business, we have to start coming in with unique perspectives, the best transistors with capacity corridors, and the PDKs (process design kits) and EDA (electric design automation) tools. The natural tailwinds will work for us. We are getting good momentum, and we are pretty excited about that. One by one, we are getting the media tech design win. The largest fabless company in Taiwan is committed to the US foundry company. That is pretty nice to see. Rick Tsai used to run TSMC, right? He is a very helpful customer. His commitment has helped teach us how to be a good foundry.
That’s great. You’re very good at this, as you have already said the keyword of my next question. I want to end by talking about Apple and the M Series. Before I do that, though, let’s pull back a bit. You are trying to run maybe the most complicated business model of all. I see this dynamic at play in every piece of the tech industry, where you run the platform, and then you want to supply the content on the platform. You run the foundry, but then you want to make the best chips that the foundry makes. You have to trade off against your customers there, or you have to piss off your own company because customers are actually paying you money.
Inside the company, you are not doing the accounting that way. You are going to go with the external dollars instead of prioritizing yourself. This plays out. Facebook has this problem. YouTube has this problem. Netflix has this problem. Google has this problem. Over and over again. Everywhere you look in the tech industry, this plays out. There is this dynamic of, “We want to be a service provider for everyone else, but we also want to use our own services to beat those competitors.” How are you planning on handling this issue?
There are a couple of aspects to this. First, I need to create a bit of separation between how we manufacture for others and for our own product groups. I have said, “I want there to be auditable boundaries between those,” so that they can look at the contracts they might do with the TSMC, Samsung, or GlobalFoundries and respond, “Huh, I get the same contract from Intel.” Same terms, conditions, protections, IP flows, service agreements, capacity corridors, all of those things. I have to earn my way into that business. Was everybody happy with TSMC the last couple of years when they couldn’t satisfy supply? Of course they weren’t. I know that. People are anxious to have a broader supply chain here. There is a tailwind for us to step into, but we have to earn our business there since they have done such a good job as a services company over the last number of years.
We are going to go do that. My product teams use a lot of foundries as well. Part of IDM 2.0 is that we use external foundries as well as internal capacity. To some degree, I have to make this work in both directions, and I have to create some cleaner lines between those. We want the external foundry customers to look at us and be able to say, “Intel is a good foundry for me.” In doing that, it also makes my own product teams better because those external customers are demanding us to become a cleaner, more productive, and more efficient foundry as well.
Thus, I have used the language, “IDM (integrated device manufacturing) makes IFS (Intel Foundry Service) better, IFS makes IDM better.” I see it driving us to be better on both sides of the business. I don’t view this in any way as a negative because if I become a better foundry for those external customers, then they are going to be demanding better PDKs, better IP libraries, better support for synopsis and cadence, etc. That is going to make my internal design teams more productive and efficient as well. As my internal design teams are more productive and efficient, they are going to be creating IP that I am going to monetize for my foundry business. If this cycle starts to really work as I hope it does, I think this becomes a very positive cycle for the company. Again, getting there means we have tons of work to do. In that sense, I agree with the dichotomy that you are pointing to.
I think we see this play out. Microsoft has Surface products, but they also make Windows and sell Windows to competitors at the Surface team. Once you see it, it’s almost everywhere, and everyone manages it differently. I don’t think I have ever seen it applied to the chip industry itself. We will have to have you back in a year to see how it’s going.
So I started by mentioning this talk you gave a year ago when you first took over the culture. You took some shots at Apple and the M series there. You said they can’t just be the winners all the time. You have now mentioned them again. They are ahead. They are on ARM. Our last guest was the CEO of ARM, Rene Haas. Do you think you can, with X86, deliver more compelling products in a laptop than the M Series?
Of course. That doesn’t mean that they’re not good, but at some point, I hope I’m doing better chips than them so they will tell me, “Hey, I want your chips.” In the meantime, I hope they become a foundry customer and still use a lot of my I/O components. We are trying to earn our way back into their ecosystem, but I have to enable the PC ecosystem to have better products than what is done by the Mac. Period. My client group has that mission. By the way, Raptor Lake (the codename for the 13th Gen processor) is a great product with ECORs, PCORs, performance, etc. We win on numerous benchmarks across the point.
As evaluators get their hands on them and look at it versus the M1, I think they are going to come out and say, “Oh, Intel is doing a pretty good job here.” I think Apple needs broader, more resilient supply chains, and I want to win their business in that area. We see this as a very natural competition model and one that I am deeply committed to be successful in — with them and against them.
Do you think Raptor Lake is better performance per watt than the M series today?
Yes. There are a few points where you can say, “They still do a good job,” but as you look across the spectrum of different power envelopes, Raptor Lake is the best product.
Well, do me one favor. Just make sure you don’t publish graphs without any labels in the axes when you put out the comparison. That drives me crazy about Apple. Pat, it was great having you on Decoder. Thank you so much for coming on the show.
Very good. Thank you.
Decoder with Nilay Patel /
A podcast from The Verge about big ideas and other problems.