Sharp Tech with Ben Thompson - Gelsinger Out at Intel, 20 Years of Structural Challenges and Strategic Blindspots, The Board and What’s Next

Episode Date: December 5, 2024

A closer look at Intel's fall from grace in the wake of CEO Pat Gelsinger's sudden retirement and with the company facing a fresh round of questions about its future. Topics include: Ben's overview of a 20-year run of paradigm shifts and strategic missteps, Gelsinger's strengths and weaknesses, CHIPS Act funds and a looming inflection point, and the murky path forward for American made chips.

Hello and welcome back to another episode of Sharp Tech. I'm Andrew Sharp and on the other line, Ben Thompson. Ben, how you doing? A little worried, concerned. This is our second take to start Sharp Tech. We somehow talked about the weather for two minutes and it wasn't even funny. It was very boring because like the worst cocktail part you ever did to. Yeah, it is what it is.

But look, we're going to spare the audience any weather chatter here. We have a lot to get to. a lot to wrap our arms around with the state of Intel. Once more onto the Intel breach here on this episode. For anyone who missed it this week, about 20 minutes after Monday's podcast published, news broke that Pat Gelsinger, Intel CEO, would be retiring after nearly four years as the CEO of Intel. You have to correct you. It not would be retiring.

It's had retired a day ago. Retroactive. Backdated. So that's how you know the board meant it. There were subsequent reports that he was ousted by the board. He was at Intel for about 34 years overall. He joined the company when he was 18 years old in 1979. A pretty incredible run overall.

Obviously, the story was a little bit more complicated over the last couple of years. Intel, one of the most successful companies in the world for most of Gelsinger's time there. But he left in 2009 after being passed over. over for CEO. And then after returning as CEO in 2021, Intel's sales have fallen by nearly a third and the stock has lost 61% of its value. So you wrote a fantastic overview of Intel's decline and some of the missed opportunities in the modern era. I mean, you have a wink here. I'm not sure which article that goes to because I feel like I've written that article five or six times through the years. Yeah, 10 years ago you were writing that article. And you've just been sort of

of continuing to beat the drum, you do now get being right points. The decline, it was happening very slowly and almost imperceptibly if you were looking at the stock price. And then this past year, it's happened all at once. I've been completely right about Intel, but I think the, it's been right in a sort of academic sense or a podcast sense in a takesman sense. the problem, and this is really the core of the problem for Intel, it's the core of the problem for what happened this week. It's just, it's a problem with the Intel board where a lot of this goes back to. The fundamental problem is that chips take a long time. It takes the design of a chip is three years, maybe if you're really fast, the building of a foundry, four to five years.

And the overall sort of roadmap of what you're doing, doing, we're talking. sort of decade-long time scales and the investments that go into sort of new technologies. You take something like EUV, the big thing and a factor in, you know, in Intel sort of falling behind in manufacturing. Intel invented the concept of EUV in, I think, the early 2000s or something. It might even have been the late 90s. So we're talking these these huge lead times. And it is such a challenging portion of the industry because you're making decisions a decade ahead of time that don't hit the financial results for years and years later. And so the issue with me and analyzing Intel to make this sort of solipsistic and, you know, self-serving here is I was in

2013, when my first article is on Chichikri, I wrote that Intel needed to change to being a foundry. And the reason for this is if you zoom out in the long run of time, you need volume as a foundry. Like this, the, to make these chips, the upfront R&D, the upfront, so there's two factors is upfront R&D, that's similar to, say, a software company where that's the actual, like, talent to create the new design and the concept. But also, what's different than software is you have upfront capital costs. You have to actually buy the equipment. And the equipment at this point is on the order of tens of billions of dollars to, to sort of get these

off the ground. And so the issue is if you don't have volume, you're not going to make that up. These are costs you're spending regardless. So your profitability over time is a function of how much can you spread those costs over discrete units, all of which on an individual basis don't cost anything. It costs a few hundred dollars for a wafer that goes into things. And off of a wafer comes tens or hundreds of chips. And so like the cost of goods sold for, actually the biggest factor is frankly electricity to make this. But the actual cost of goods sold for a chip is minuscule.

All the cost of a chip and its profitability is depreciation and also sort of your amortized R&D. And so you have to have volume. The reason why Intel was already in trouble in 2013 when I wrote that article. And the funny thing is when I wrote that article, my sense then was that, well, I just started to checkery, so I need to say this, but I might be too late, which was they missed mobile. And by, now, the reasons they miss mobile are there's a lot of, the commonly accepted story was one that Intel told, which was, and I think Paul O'Doini, at the time, he's, he's since passed away, but basically say something in the lines of, oh, yeah, Steve Jobs came to us.

We thought it wasn't going to be, you know, make enough money and no, turns out it was this huge market. Oops, what a shame. I've passed on that story before. I had Tony Fidel, who was, you know, basically the creator of the iPod and the hardware portion of the iPhone. This is actually how I had it on my podcast. He came out. He's like, no, you don't understand this. We're going to talk about this. And basically his take was no, Intel is never competitive. The issue for Intel was they were always focused on performance. Even when they had armed chips, they owned an entity. called X scale that made arm chips. They were all, Intel's always sort of assumed it's plugged in.

And so they've always been focused on performance. And the problem with the phone is obviously the most important thing was efficiency. And he's like, they just didn't have the view and mindset or product to do that. And so he said that actually Steve Jobs desperately wanted to go with Intel. And, you know, it was clear they quit for the iPhone. Yeah. And according to him, and this is all. in a struck in a struckery interview that will link to when the iPad came out jobs was actually

still on okay let's go let's go with intel for this one and they had to be like if you put an intel chip in this we're walking out the door um it's just not it's not going to be sort of what's going to work and that's how apple and that actually what is what drove apple then to acquire PA semi which is like oh okay we're if we want the performance that we want from an intel but we want the efficiency of arm we're going to have to do it ourselves and and the crazy thing is is these discussions were happening, like, what chip goes in the iPad. The iPad launched in 2010. The discussions weren't happening then.

They were discussing in 2006, 2007, before the iPhone even launched. Because this is the timelines you're talking about. And so this timeline bit is sort of really important. So, you know, I'm doing a massive weave here, but I promise to come back to the beginning. Okay. So the issue is Intel misses mobile. Arm becomes the standard. And this was devastating Intel on a few regards.

One is which when you restart a platform, when you start from scratch, all the software locking that Intel had sort of didn't matter. And this is actually where Gelsinger enters the story. So Intel starts out with this, what's called a CISC architecture. This doesn't really matter, just the way that instructions are sort of ordered. And there was generally, there's CISC versus Risk. And the general thought is that risk is just fundamental. more efficient, but for various reasons, the initial Intel 8088 processors and that 8086,

which is where the X86 normalclature comes from, were CISC processors. And so the Intel made the, I believe it was, they had the 8086, they had the, I think it was then the 286 or 386, somewhere along those lines, there was a real movement within Intel to say, we need to restart this. We're not on the most efficient path. And Pat Gelsinger was the leader of the faction that said no. What we need to do is actually we now have five, six, seven years of entities building low-level software, assuming X86 chips. We own the PC market.

Yes, in theory, the risk approach is better than ours. but it will take at least two years to rebuild the sort of software layer on top of that so it will be functional. And in two years, guess what? We will have learned how to make the chips even faster through our manufacturing process. And so Pat Gelsinger was, and he ultimately won the day. He actually was the lead designer of the 486. I mean, he's very technically capable. But the big thing was the strategic bit was he was, he was.

won the day by saying it doesn't matter that we're sort of structurally less efficient because of the software moat that we have and we can make up for the lack of efficiency through our superior manufacturing processes over time and he was 100% correct it was absolutely the correct call intel came to dominate was one the most profitable companies in the world massive you know for for years and years and years because they own x86 now the whole cisc versus risk thing is sort of less important over time. Like internally, the way these processors operate has changed a lot and there's sort of a translation layer.

But X86, so it's sort of a less important detail today than it was then. Other than to say Galsinger was critical in this major strategic inflection point within Intel, which where the outcome of the inflection point was that there was no inflection point. We're going to keep on the path that we're doing because we have this sort of lock-in. So you fast forward to mobile. And so number one, you can see. this Intel sort of intrinsic dismissal of efficiency is long rooted, right? It kind of goes back

to that discussion. Galsinger won by pushing the, we don't have to be the most efficient because we'll make up for it manufacturing. We need to have our software mode sort of view. And again, it was 100% the right thing to do. He was right. If they try to reset to risk, you're resetting the whole ecosystem, you're giving up your advantage in the pursuit of an more elegant solution that opens the door for other companies to come in and compete with you. And so it was absolutely the right thing to do. But you fast forward to mobile. Just Intel structurally and culturally is a company that's optimized on efficient, is optimized on performance, sorry, and they gain that performance through their manufacturing process. And so that's number one. So you get to mobile. And the problem

with mobile, as I just discussed, is efficiency matters most. So that's number one. Number two, the key thing that Apple did with Microsoft, they were just trying to do a small version of Windows on the phone. And they sort of assume they'd have sort of a compatibility and, you know, it'd be easier to port their programs and all those sorts of things. Like Microsoft Mobile had a start button on it. Like just to, like it was, it was definitely just a small version of Windows.

and Apple correctly and appropriately realized, no, we need a complete reset. Yes, we will use the kernel of macOS, but everything in user space, everything that sort of matters is going to be completely rebuilt from the ground up. Guess what? If you're completely rebuilding from the ground up, suddenly all those little advantages, and if your number one priority is efficiency, the gains of staying on X86 kind of disappear because you're rewriting everything anyway. And this was such a critical moment for Intel.

It wasn't just that they didn't get those sales. It's that mobile inspired a ground-up rewrite of all software that actually undid sort of Intel's core differentiation and moat in their sort of advantage. And so you fast forward. And so Intel has this. And then, of course, you have the classic disruption story of arms gets better and faster and there's all this volume behind it. But so Intel had this big problem in 2013, which was number one, we don't have the volume of mobile. Having volume is critical, especially as every time a new technology is introduced, like you go to FinFet or you fast forward, you go to EUV.

It gets so much more expensive. You need volume to do that. And when you say you need volume, you need volume from third parties. and that's what they could have done had they gone the found your direction 10 years ago. So you just need volume, period, wherever that volume comes from. In a perfect intel world, they have the best performing mobile chips. And so everyone is just buying mobile intel chips like they did for PCs. But that was my point in 2013.

They had already missed the boat there. You missed the boat. So you need to get in back on that wave. The only way to get back on that wave is to become a foundry so that, and what you can do is you can go to an Apple and say, look, well played, you got it, arms the winner. We have the best manufacturing. Let us help you make the fastest possible chips. And they obviously, number one, did not do that.

And then number two, everyone points to this as the Intel failure, but as we just recounted, actually, they were in serious trouble before this happened, which was they got stuck at 14 nanometer and then 10 nanometer. And so TSMs are past them. The reasons for this are long and complicated. You know, getting EUV to work was a real challenge. It was massively expensive. Intel felt confident they could use their old DUV sort of approach, do this sort of

quad patterning. And they could get it to work. This is sort of relevant to the Smick and Huawei 7 nanometer thing. They could make a 10 nanometer chip. They just couldn't do it in an economic fashion, which means their yields were too low to spend the amount of money to scale that up, they wouldn't have been able to make it up, make it up in volume, as it were.

And so, so, so they get surpassed. And suddenly that advantage that I was highlighting in 2013, look, you can still partake in the future. You just have to be a manufacturer. You're going to need volume in the long run. This is actually a much rarer capability than chip design. Because the other thing that happened was TSM came along. And to have a foundry, you had this explosion.

of fabulous chip makers where they could create their own chips without having to create their own foundry. And so suddenly, Intel, like, the design bit was less important than it used to be. Yeah. Right. And so, so this is the reality is that Intel, Intel by 2013 was already screwed. Now, to take this back where we started, Intel's results and stock price returns in the 2010s were phenomenal. And this was, this is the challenge that I felt as an analyst where I was, I was, 100% right.

And yet if you believed me and chose stocks based on what I wrote, you were getting screwed. If you shorted Intel, you're really screwed. But at the minimum, you're missing sort of all this sort of upside. And the problem is, in general, I'm sort of skeptical of the people like to blame the stock market for sort of making short-term decisions. But this is an example that is very much in sort of the critics favor. the stock market theoretically is valuing the long run terminal value of a company and all these sorts of things.

That was not happening with Intel because here's the reality. By 2013, when I wrote that article, Intel was screwed. The extent to which they were screwed was not fully visible until 2016, 2017 when TSM passed them, but they were already screwed. And that was not reflected in the stock price until 2022 or so when Pat Galsinger is suddenly back as CEO. Right. Well, and you talk about timing. I mean, Pat Gelsinger takes over and now tries to employ the plan that you proposed basically a decade ago, but they are too late. Oh, wait. Well, here's another, here's another bit of history that is back Celsinger related.

So one of that, one of Galsinger's big projects before he left was called Larrabee. So I was pretty critical of Larrabee, but just to sort of explain what it was, Larrabee was basically a, an Nvidia competitor where it was a, what's the word, GP GPU or GC, it's a general computing on GPU, GCPU, where basically, Nvidia started by making chips that were just for 3D graphics, but then they realized, wow, there's actually a completely different kind of computing you can do on GPUs, this massively paralyzable computer. And Intel started out Larrabee as sort of a graphics competitor, but they sort of pivoted to, like, no, this is just going to be a parallel compute architecture.

and Larrabee, as you would expect from someone that was that was led by Pat Galsinger was X-86 based. So all these little cores on there were X-86 cores. And so on one hand, you could, like, that's what Intel did. It's what Pat Gelsinger did in particular. No one was more married to X-86 than he was. On the other hand, you, like, you could see that could potentially have real gains in terms of programmability. And like there's like just understanding how it works. Maybe it was the wrong approach.

is trying to shoehorn X-86 into literally everything. They wanted mobility X-86. They wanted, you know, all these, all these bits and pieces. So there's a lot, like, I generally thought Werribe was a over, you know, a little to, the X-86 focus didn't necessarily make sense. But in broad strokes, in sort of the grand scheme of things, it was in line with the future, which is this paralyzable sort of capability. And that Invidia was a long-term threat in terms of taking over processes.

workloads. So Pat Galsinger leaves in 2009, goes to EMC, eventually becomes the VMware CEO, and a year later, or six months later, Werribee's killed. And, you know, again, political infighting, whatever it is. Again, some people don't just think it never would have worked. We're not here to sort of legislate Larrabee, other than to note that Intel also is suddenly sort of, is nowhere in terms of the future of computing, which is AI. And this, again, speaks to the timescales. Like, there's a, I just read the Nvidia Way, my interviews with Take Kim, the author of it, on Shrekri, which, you know, would be coming out today as well.

And he recounts this meeting in 2013 with, with Jensen Huang, where he's all in a panic because this AI sort of approach is we're a little late to this, all hands. on deck were resetting things, we're moving forward. And, you know, InVDVD was actually already very well placed. They had just in general, the idea of Shaders, this paralyzable platform. They'd already invented Kuda back in 2006, which by the way, NVIDIA was murdered in the stock market for doing. Like, they were spending all this money. They were spending money on Kuda in two ways. Number one, the R&D costs were astronomical. So they were, to build this out. They're basically building out an entirely new.

platform and operating system and all these sorts of things, not an operating system, but you know, that sort of platform. And then number two, they were putting Kuda dedicated cores onto their chips. So onto the chips, it's a, it's, it's, it's, it's, it's, it's actually a cost because either number one, you're making your chips less performance because you're taking up space to put on these other cores that aren't useful for games. Or you're making your die bigger, which means your yield's going to be worse. because the likelihood that there's a defect on the chip goes up, the larger the chip is.

And so it was hurting their margins. It was hurting sort of their cash flow. And Wall Street was killing them for that. They had an activist investor on board who was like a very upset about this and then you have to do this. And yet you look back this period, 2006, 2013, is the foundation of everything that's happened to Nvidia over the last four to five years. Like this is how it works in semiconductors. The lead times are even longer than you think.

And at that time, Nvidia was making the right investments and Intel was killing them. And killing the appropriate investments on their side. And the whole point of all this is this is this is the period, this history that I'm talking about that seems ancient. This is why Intel's stock is in the dumper. This is why the results are down. They lost the X86 advantage because of mobile. Mobile starts this rebuild of low-level software. You fast forward, and not only does Arm progress extremely rapidly, but also when an AWS comes along, says we should put Arm in our servers, there's already a ton of low-level software written for it because it was already developed for mobile.

And so just Arm becomes dramatically more capable. One of the big things that Renee Haas, the current CEO of Arm has done, in part with Tony Fidel's help, who joined the board a few years ago, but is sort of, Arm had this poverty mindset, which is, oh, we are just like the chip that's used for like your power windows in your car. And we can't sort of raise the price too much because, you know, someone else is going to come on and disrupt us. What he realized is actually the whole mobile revolution has built up a software moat around Arm that is just as good as the X86 one used to have. So guess what we can do? We can jack up prices. We can force, you know, these new contracts.

And that's why, like, the arm sort of like, I got the Arm IPO wrong in. part because I was a little too anchored on their, they're just crappy business generally and sort of historical returns. And what Renee Hos got right is actually, no, we have a lot of capacity to jack up price because everyone is stuck with us, just like Pat Galsinger realized everyone was stuck with X86 back in the 90s. And all that's downstream from mobile.

And so now you have arm in the data center that is infringing on, on Intel and AMD both. You have AMD with frankly better. architecture and also on TSM, a better process that is, that is pressuring Intel. All this pre is before, happened in the period. Well, some of it happened when Galsier was there. Again, the miss for mobile. He was there.

But all this intervening period, it all happened then. And so there's a lot of mistakes that Galsinger made, which we can get to in a moment. But the key thing is that all the problems Intel has go back years and years and years. Right. Gelsinger's failure was a failure to stop a spiral that he didn't necessarily start, although it is interesting that some of his decisions were part of the failure to... Okay, well, so let's get to what his plan was. So you go back to right before or right around when he took over.

I wrote an article saying that Intel needs to split up. And the reason is that, look, the design business over time, like Intel owns markets like on premise and government and they do very well in PCs. They do have a lot of surrounding technology. And also like an Amazon or a meta or Microsoft is going to do the work to get AMD at par. If you're sort of like you've been building on Intel forever like a government agency, you're not going to buy anything else. It's just like it.

So they have they have like they have good safe markets like it's not growing, but you can bring a lot of money out of it sort of over. time. And that's sort of like, you become just a fabulous business. You become like a, you become like a, and then you have this foundry bit, which kind of, uh, is behind and doesn't have a reason to exist outside of sort of national security concerns, to be totally frank. Like in, there's just no economic reason. TSM is ahead. They have scale. They're structurally better because they don't compete with their customers. And you, and you really, and so. So before we get to Gelsinger's plan, that was one of my questions. It's like, and we're now 25 minutes in. And,

great overview of Intel history and where it all went wrong. But why is Intel important today? Is Intel important today? They don't make the best chips anymore. They don't have the best manufacturing. And, you know, what used to be a central role in the story of American technology now looks more peripheral or ceremonial. So, like, is that an unfair assessment of where Intel stands in 2024? Economically speaking, no. There's no reason for any company to, uh, work with Intel when they can work with TSM. And TSM is kind of going for the kill. Like they're building a ton of two nanometer capacity.

They built a ton of three nanometer capacity. They're not giving space to Intel, which is, I mean, kudos to them from a business perspective, but that is one of the many problems sort of Intel faces. And then, yeah, the X86. I mean, yes, there are niceties and advantages that come with Intel chips as opposed to AMD,

like, and there's things like Firewire and some of the various standards that they're involved in. not Fireware, Thunderbolt. But so there's things that, but by and large, no, there's not really a reason for them to exist. There is a long run. I think one of the critiques of TSM is they did bring EUV to market. In the history of semiconductors, every few years, there's some sort of transformative new technology that has to come to keep Moore's Law going on.

And the thing about Moore's Law is it's not inevitability. It's a choice. You choose to figure out what's, we hit a wall here, what's something else we can innovate on to sort of figure out what's going forward. People are locked onto lithography because most people never knew what a chip was

until three years ago and at that time, lithography mattered. Withography was just the current one. Before that, before lithography, it was FinFET. So FinFET was changing,

basically making transistors 3D instead of 2D. And of extremely difficult, costly change, that was pioneered by Intel. EUV was pioneered by Intel. Backside power, which is a new shift coming forward, was pioneered by Intel. The reality is that Intel for all their flaws has been the driving force between Moore's Law.

They are the company that just has relentlessly not just push forward their chips, but has been investing in thinking about long-term technologies to the extent where TSM has just been copying what Intel did for almost its entire existence. So does that culture still exist? You wrote on Tuesday that they spent a decade throwing spaghetti at the wall. Culturally, can they... That was a product perspective. From the actual, like, what's happening, like, so gate all around is a new transistor of type and backside power. That is downstream from Intel.

Like, I think there's legitimate questions and concerns. Is TSMC, yes, again, the EUV was great. But again, that was ultimately an Intel invention. And ASML played a big role in that. And for various reasons, Intel didn't. but is TSM culturally capable and willing to push through Moore's law to keep it going? Can they be the standard bearers for it? I certainly hope so, but I would say that remains to be seen.

Is that a challenge that all of these companies face? I mean, it's similar to Intel, where they had really successful business lines 12 and 13 years ago and didn't want to disrupt that. Here's the core intel problem. This goes back to my my general view on life and people in general, like when I give career advice or whatever it might be, which is you have to understand two things. Number one, your strengths are your weaknesses. There are two sides of the same coin. Number two, you should focus your career on accentuating your strengths and doing the bare minimum necessary to amelior weaknesses. Like that is how you get a friend.

And so many people, they get so frustrated with their weaknesses or they see the, weaknesses and others without appreciating that is part and parcel of the strengths and they want a world of no tradeoffs where you just get all the best thing of everything. But then number two, they spend way too much time sort of a million their weaknesses. And the reality is Intel is, if they were a person, it's the cockiest son of a bitch you've ever seen. They think they're the best in the world. They think they can defy physics. They can defy the limitations of of materials, of light, of quantum mechanics. Right.

And you know what? For 50 years, that's exactly what they did. And the problem is that the sort of entity that does that and is capable of that is really horrific to work with. Everyone hates Intel in tech. Everyone hates them. They've treated everyone like crap. They screw you over.

They like, well, if you try to work with them, they'll still your tech. And their whole way they're processing work was their manufacturing was such king of the world. They would like tell the design what to do. You need to adjust to figure out to work with us, XYZ. All of this is the exact opposite of how TSM works. TSM is there to serve you. They're going to help you out. They're going to figure it out.

They're going to have great customer service. They're going to have all the pieces you need. They're going to like adjust, like, try to figure out, oh, can we adjust our process to get this to work for you? All these sorts of things. And I think the, and so the problem with Intel becoming a foundry is they ended up in a situation where their success depended on getting really good at the exact stuff that they were bad at. And that's just, that's just a tough place for a person to be. And it's a tough place for a company to be. And on the flip side, the worry I have about TSMC in the long run is they are too good at customer service.

and do they have the arrogance and the annoyingness to work with to force the industry forward by sheer force of will? To finally push it all forward. Yeah. That's right. And so like, and so the problem, the point I came to a couple months ago. And I was so angsty writing this article. I felt angsty afterward. And I still feel angsty today. Like, is that this cultural problem can't be overcome. I just don't see how Intel ever acquires a meaningful customer base outside. a national security incident. And that's the answer to why is Intel important today? It's a national security concern going forward is the way the American government has framed it.

Yeah, the U.S. is like, we talked before, like Trump has accused Taiwan of like, you know, being like the mafia and holding him up over chips. And the reason why, number one, that's correct, because number two, it's true. We are completely dependent on Taiwan. And Taiwan does not properly invest in my estimation in multiple facets in its own sort of self-defense. because it just assumes the U.S. will save them because the U.S. is no choice. And so. Everyone's right. Yeah.

The people, again, it's an all-time example of people getting really mad at Trump for saying something that's like just 100% true because of the way he said it. Well, and Taiwan and TSM is they are also correct that we have no better option but to defend Taiwan. That's right. And so the thing I kept coming back to is like Intel's never going to fix this cultural problem. I think that's a reason they don't have any customers. No one trusts them. No one thinks they'll do it.

Now no one's going to really trust them, right? Especially after they just kneecap the guy that was building out this whole thing. And so just to go back to Gelsinger's plan. His plan, he comes back to Intel. And he's like, yeah, we're going to become a foundry. We're not going to split. We're all in. And there's a lot of logic to that.

If you're going to build a foundry and you want to get customers, but you have, you need scale. And Intel has its own scale. They sell a lot of chips. And so they were sort of their first best customer. The concern all along is if you don't truly. split Intel can you actually have a customer centric foundry? And I think that was

the sort of the concern of lots of folks. At the same time, by all accounts, Gelsior did a really good job fixing their technical processes. Like they're coming out with 18A next year. Now, it's still TBD how well this is going to scale. They have their own internal 20A

or do they skip 28, like Intel 3, their costs, their margins were horrible last quarter because their costs about auto control and they tried to scale it up. This is the big problem with the whole five nodes in four years. There's a lot of learnings and tacit knowledge that comes from not just inventing a process, but scaling a process.

And it's TBD if Intel can scale 18A in sort of a meaningful way. But 18A is a big technical achievement. It's new transistors, gate all around transistors, and it's a new, it's backside power where it used to be, you would have sort of the transistor level,

and then you'd have all these communication wires, you have power on top. And the problem is these wires are getting thinner and thinner. you'd have this real problem of like interference of like trying to send power down these big thick cables and all these communication cables around it. Now you're putting power on the bottom,

transistors in the middle and the communication layers on top, which is much more elegant. You don't have the interference problem. It's also much more challenging because you're, in your process, you're laying down all those power layers before you do the transistor layer, which is the most fragile,

which means if you screw up on the transistor layer, you're throwing away a lot more work than you were sort of previously. That's why transitions run on the bottom first. You do the hardest stuff first because if you screw up, then you lost sort of less process time. So we don't know if it's going to scale. But it is a technical achievement. It is the sort of achievement that is helping push the industry forward, which is what Intel does. And my view is, look, at the end of the day, there's not an economic reason for the Intel Foundry to exist.

to the extent we want it to exist, the U.S. government needs to pay for it. If the U.S. government's going to pay for it, it needs to be viable in the long run. It needs to be separate from Intel. A standalone entity, yeah. Because the third party customers, we've talked about this in the past,

but the third parties that would potentially be working with Intel are taking a risk on a process, a manufacturing process that may not be as reliable as TSMC. It may be more affordable in the short term, but like also, If it works, it becomes, it's possible that Intel does retake the process lead, by the way. Well, but the other concern for third-party customers is that if Intel is also serving itself with that foundry, then Intel is going to get the priority going forward. And so it makes more sense to sort of sever the connection if you're trying to market as a service business to third parties.

Is that right? Am I remembering our previous conversations correctly? I think that's correct. And you also can't oversaw. state the extent to which everyone hates Intel. Like, like, it's, it's a very real and true thing that I think is underrated by, and this is one where all the analysts just, yeah, there's a lot of semi-analysts that got mad about my article and really disagreed with it. And I think they, they had all intellectual reasons that were correct. I think this is one area where I was influenced by being in Taiwan. And in Taiwan, I've, I know lots of people that work in lots of suppliers and lots of areas that deal with semi-chips. And you just can't escape that sentiment.

Like there is a sense where you have people on an intellectual level want Intel to succeed because they understand why it was important. And they're also just delighting in their downfall because they fucking, excuse me, because they hate those guys, right? Yeah. And so to your point, the other bit about is if you go with Intel for a major chip order, TSM's not going to let you back in. Like you're back of the line. And guess what? You can't afford to be back of the line today with TSM. There's some very competitive line with TSM these days.

That's right. That's right. And TSM was, you know, I think TSM could be hardball in their own right. They actually be very hardball. And this was a, or play hardball. And so, yeah, you were just sort of stuck in this limbo. And I do think that, you know, so, yeah, we can talk about sort of paths forward, but that was the situation. And so I do, I do think there's a bit where if only. Again, Gelsier's overall view of the world may have been incorrect. It was a view of the world that led to Intel missing mobile. And so in Werribe, I'm not sure was the right. It was in broad strokes in the right area. I'm not sure if it was the right approach. But if Intel would have embarked on this manufacturing shift in 2012, 2013,

when I sort of wrote about it, not only would they be partaking in the volume of mobile, they'd also be partaking the volume of AI. Like that's how you, like, and this was my point. The way, like if you're a founder, you're going to take part in everything. And which is not just a good thing. It's also a necessary thing because you sort of need the volume. And so I think Galsinger's plan was the right one a decade ago. If Intel had done that and embarked on this, they would be in great shape right now.

The problem is it is by the time you fast forward a decade, is it doable and viable or is it sort of gone too far. And so this is why I have a real, on one hand. There's a real symmetry. Both you guys were dead on, but off by 10 years. So it's great. Yeah. And this is why like I gave him a lot of rope because I think this are necessary. Now big mistakes he made. If you're embarking on this, you have to realize every dollar of cash is critical. He needed to kill the dividend day one. They've spent billions on dividends since he was CEO. They needed to lay off. a lot of people on day one. You needed to realize you're going to have a massive cash crunch to pull this off.

And I think there's a bit where, and again, is that his fault? Is that the board's fault? The reality is Intel's board is an absolute disaster. It's one of the worst boards in the history of technology. They've been a problem for years and years. They've hired the wrong CEOs again and again and again. They've been focused on short-term shareholder outcomes again and again and again. This dividend has been their number one priority again and again and again.

And the reality is they were a decade behind doing what they needed to do. They were 15 years behind having missed mobile and thus fundamentally changing their long-term trajectory. And this piece of crap board has not recognized or realize a single thing. And if you are going to bring Pat Gelsinger back in 2021 and he says this is the plan and you say you support him, then it's like, wow, you actually are putting your rear end on the line because this might not work. It's super aggressive. It's going to be really hard. But hey, if you're all in, you're all in.

And then to turn around three years later and fire the guy because of results that are downstream from your lack of governance is disgusting. Every single person on that board is disgusting and should be ashamed of themselves. And yet they seem to have no, no awareness. This is the issue. You can go back and say, actually Gelsinger was the wrong guy for the job because Intel was already dead. And they should have embraced that and they should have split the company and had this pathetic X-Nad6 design team that can muddle along and make their little chips for PCs and government entities and get someone to take on the foundry and try to build something better. That would have been a legitimate outcome. That's exactly what I prescribed when Gelsinger took over.

But if you're going to commit to this guy, if you're going to go down this direction and then give up. after three years, you have no right to be anywhere near the boardroom of a chip manufacturer, a business that requires decision-making in decades. These guys can't look back a single quarter. It's disgusting. Wow. A lot of heat there. I love it.

So we should also note another, a mistake that Gelsinger made is he did not lay off enough people because this was going to be really expensive. But like, in the question, like, I also feel sympathizing. for him. The one thing that really screwed Intel was COVID. And the reason it screwed Intel is because they had phenomenal results that were undeserved. And when you're getting phenomenal results, your stock prices back up, it's hard to lay off a bunch of people and it's hard to kill your dividend. And so I'm actually, even though those critiques, I'm fairly sympathetic to him, where I'm less sympathetic is Intel, to take this job and to try this, you have to be a really optimistic guy. And again and again, Intel's, forecasts were too optimistic. I think there was a real CFO failing as well. Gelsinger needed someone in there who was just the biggest wet blanket of all time. That was like like that that would have seen the COVID overhang ending. I mean, I came away from the Gelsinger interview on Sertechery a couple years ago rooting for him and also feeling like maybe this can work because there was just a lot of positive energy. And Intel needed that internally. That's what's required.

Yeah, exactly. Intel needed that internally, but they needed a CFO that was. was going to make sure they didn't over forecast and then misprojections and all these sorts of things. And, and again, could he have realistically pushed through a cut to the dividend and a cut in workforce during COVID when their results are great? Probably not. That doesn't change the fact. It was still a mistake.

Like, like that should have happened sooner. But Gelsinger was clear. And I asked, the first question, I've interviewed him twice. The first time I interviewed him, this is the first question I asked was about this split or whatever. And his answer was, if you wanted to split, I was the wrong guy for the job. The board agreed with me. They're behind me.

This is what we're going to do. Well, guess they're not agree with you and behind you anymore. This idea that you're going to execute this plan and see results in three years. It's infantile. Again, you can go back and say, okay, Pat, you're not the right guy for the job. We're not going to hire you because we're not sure we can do this. Okay, that's mature board management. To hire the guy and say you're behind him and to fire him after three years, it's disgusting.

There's very few episodes in technology that have made me more disgusted than this one, even as the board might be correct. It might be the wrong plan. But it doesn't matter. You don't sign up for it and go for it and give up after three years when you're the idiots that didn't see this coming a decade ago. There you go. It sounds like the arrogance at Intel has actually yielded real benefits over the course of the last several decades. The board is just arrogant.

No, there's all these finance people and like what, like they, the, now again, I don't want to, I'm not necessarily going to the mat for Pat Galsinger. I think the plan might have been a bad one. There's, you know, he's also a, by all accounts, a pretty arrogant guy that thinks he's always right about everything. I think he made a lot of really critical mistakes. Intel regularly, like, was carrying too much inventory. They didn't understand consumer trends.

They had some real misses in terms of products. Their margins exploded out of control last quarter because they had scaling issues and things like that, which they didn't forecast anyone on the street. So it was a total surprise. That's what sort of triggered all this sort of right now. He's made some real fundamental management. It's interesting. I'm criticizing these folks for all being finance folks. Gelsiger screwed up a lot of the financial stuff. They're going to use a couple of finance folks. Yeah, exactly. But, but again, just it's the broader, it's the broader trend of Intel having its eyes closed as they wandered into this ditch. Well, and it's also, it's just completely incoherent.

to bring Gelsinger back in 2021 and commit to this path. And now a couple years later say, oh, my God, look at the stock price. Someone needs to take responsibility for this. And we need to go a different direction. I hear what you're saying. I love the energy out of you. You said you've got a good night's sleep for the first time in a week since coming back to Taipei. I can feel it here on the podcast.

We've gone too long, though, without talking about what's next. So there's an email we got from Sean. and I actually had a similar question reading your daily update earlier this week. So Sean says, hey guys, in Tuesday's daily update, Ben noted that the Chips Act grants Intel $7.9 billion for investment into American fabs, but requires Intel to own at least 50.1% of Intel foundry. Ben then concluded in his update, the Intel board has cold feet about Intel's foundry efforts, and this provision forced their hand. The best way to, quote, put our product group at the center of all we do, which is a quote

from Intel, is to preserve the ability to separate it from the foundry business and potentially go fabless, just as AMD did 15 years ago. That option, however, is about to be foreclosed. So, and Sean's question is, is Ben implying that Intel's board is now on a path to, one, selling Intel foundry off, and two, returning the about $8 billion back to the Department of Commerce. This seems ludicrous to me for a company that lost over $16 billion last quarter driven by the Foundry investment. Or, he asks, is the case that Gelsinger wanted to take the money to go all in and bet the

company on Foundry? But the board said, maybe we just take the chips money, half-asked this Foundry thing, go Fablis and contract everything to TSM, then ask the government for, more money when the fab stuff fails. Do you have thoughts on that question? Because I wasn't sure what exactly you meant in terms of the board getting cold feet and the path that they've chosen to go down at this point. So at the end of the day, it's all about cash flow.

And Intel is the reason why all the panic is about Intel is they have a potential like having cash flow problem. They have like $25 billion on hand or something like that, but they lost like $6 billion last quarter. And again, this isn't about profits. because the profits for a semi-conductor company are very divorced from the actual reality. Because, again, there is cash out the door to buy this equipment that only shows up in the earnings as depreciation over a number of years. And when you're building like a shell, that shell might be on a 30-year depreciation schedule.

The equipment might be on a five-year depreciation schedule, I think, is what it usually is. But the point is it's not actually hitting the numbers until you're actually producing this sort of stuff. but the money is going out the door. And so there's a bit where Intel could take the world's largest loss of all time by simply stop buying stuff and writing down all the stuff they've bought. But it doesn't change. That's not like they would report like tens or hundreds of millions of dollar loss. But that doesn't, it actually improves their cash flow position because they're not sending money out the door anymore.

Right. Like they're successful chip in the market right now. Lunar Lake is made by TSMC. like TSM is spending the money there. And so there's a bit where like, sorry, Sean, it's not ludicrous. Like it's a cash flow issue. They can stop buying stuff and fix their cash flow sort of issues. Right.

Now it would be, again, a massive accounting loss, but from a cash flow loss, you know, they can certainly be a better position. So the issue with this Chipsack Grant is, again, it's not very much money. It's only $7 billion. $8 billion is really not much money. It's not much money. This is just an aid, a help. It's not even remote.

These fabs cost like $20 billion. It's not going to really help or it's not going to pay for it. And it comes with the tie that you're locked in to the foundry business because guess what you can't do? You can't sell it off for pieces, right? Now, Intel could sell. They have all this equipment, right? You know, this could be a boon for tea. They bought high.

NA EUV machines from ASML, which is like to get to like one nanometer and beyond. TSMC, I think, is bought like one from ASML. It's driving ASMC crazy. They keep playing on their earnings calls because TSM's like, oh, we don't need it yet again. Number one, TSMC being conservative. Number two, they're, they can hold out. There's no rush. You know, this is why when people say, oh, why does it ASML take all the profits in the industry?

It's because they have one buyer. That's why. Like, they're gated by sort of the overall thing. And so maybe it'll be like, hey, we'll sell off our. high and A machines, we'll just sell them to TSM, right? Because we're closing down our phone. Like, they can get some of this money back. They can sell the pieces.

They can like, Intel, by the way, has been selling a ton of stuff to the Chinese over the last, they've been selling off everything, all their old stuff. Anything that moves Intel's been selling. We need cash flow for a while. No, they have been. They have been. You talk to people in the supply chain. It's been a big thing of all this intel equipment, old intel equipment that is just

sort of being sold off for pieces. Our sad, beleaguered national champion. No, it is an irony where they're the national security champion and they're just selling equipment left and right to China. Whatever you guys need. We just need cash. Yep. It's been a lot of it happening. Would they actually give the $8 billion back in that scenario? Well, so I was reading through the Intel disclosure and basically all the penalties are like you have to give the money back. There was one, which is and other things that are available to the Department of Commerce.

It's not like basically this is my view on what happened. And this is influenced by the suddenness of this decision and why over Thanksgiving weekend and he's retroactively retired to a day before. Taking this money and spending this money is an aspect of we're truly committed to the IDM 2.0, the Pat Kelsinger vision. Yeah. And I think the board has had cold feet about it for a while. And this was a forcing function where we're truly locked in. and I think that they got very, this is pure speculation to be clear.

My sense is, like I said, they got cold feet and like, well, we need to preserve our options, XYZ. And I think Packhouseer was like, you said you supported me on this.

We're going down this road. And they're like, eh, and he's like, if you don't go down this road, I'm out. And they're like, so we put the retirement for yesterday.

That's my, that's my, I think just the, the circumstance around it. That's my view on what happened. By all accounts, the board was pretty blindsided by this. They have no one in place. There's no success in plan. Gelsior's like, yeah, I'm not staying on as a consultant or it's not chairman emeritus or whatever. He's just, he's like, fine, I'm out. I'm done. And by the way, I think it's completely justified. He came back on the point that they were

going to support him in doing this. And I think the way, the abruptness of this exit, the ugliness of it, I would bet more details come out, speaks to the fact that he felt stabbed in the back. them. Now, again, big picture, this might be the worst person in the world was right scenario where the board might be right. This is the right way to go. But, but the, the, that's my view on probably what happened is this was a forcing function. Now, again, I think here's life advice for you. I think people do. There's a bit with fallacies where people get so obsessed with avoiding fallacies. They actually make bad decisions. Like, I always talk about, oh, confirmation bias, I want to be very careful of that because the way that I write.

And then I wrote an article because I was like looking for disconfirming evidence that showed that I was wrong. And I thought I found some. So I wrote the article. And I was totally wrong because I actually had was right all along. I got so obsessed with ameliorating my weakness in that regard that I like just wrote the wrong thing. And I think there's a bit with sunk cost fallacy where it's a bit where everyone's wary of sunk off jealousy.

Don't be overly biased in your decisions by what you've already spent money on or what you've already done. I think there's an inverse of that where you. you've already made decisions. Like, it is what it is. You're at the position you are in life. You're at the position where you are as a company. There's so much to go back with Intel and change.

You don't get to make decisions as if it stays zero. And I think this is maybe a case. Like, I wrote the Intel split thing in 2020 or 2021. That was the time to realize we're not going to go down this route. But once you commit, commit. And they didn't commit. And that's why I have such scorn for them, for the board. And so, yeah, that's my view.

This is speculation. We just napalms the entire boardroom. It's wonderful. Well, so what did the Intel board is disgusting. It's been disgusting for 20 years. Like, that's the thing. If this isn't a one-time thing, this is an entity that has prioritized financial engineering and financial results for 20 years.

And Intel's disaster is 20 years in the making. I mean, that's the thing. Like, looking at it from afar, it's like half a block. Blockbuster story where this is some, to some degree, there's just like natural disruption that's happened as the paradigms have shifted and chipmaking has become more modular and the trajectory makes sense to me. And then there's also this sort of Boeing angle to everything where it's a Boeing story. I'm pretty sympathetic to Blockbuster. They're a they're a bricks and mortar retailer. Why do we expect them to figure out streaming? Like, well, totally. But that's I'm also sympathetic to Intel. in certain respects. No, I don't have any sympathy to Intel and Boeing. They were pursuing financial results.

They drove out long-time engineers. They moved manufacturing out of the Northwest because they wanted to deal with the union. They move their headquarters. This is, don't get me start. Boeing and Intel really are two peas in a pod. There are two entities. And airliners are another industry

that requires decades out sort of investment. This idea, we're just going to do another 737, even though it was created in the 1960s, is too low to the ground because, because it wasn't even conceive of modern jetways or anything on those lines, instead of doing a clean sheet design because it costs too much, or we're going to outsource the entire production of the 787 without figuring out of making it first because we're trying to be cheap.

There are two peas in a pod. They're long-term companies that depend on long-term thinkers at the top that prioritized and got obsessed with short-term financial results. They are very much two versions of the same story. Yes. Okay. I don't want to argue too much on Intel's behalf, But where I'm coming from with the sympathy is you look back on missing mobile.

Can't what to call you an embecile. No, you look back on why they miss mobile. Well, they were really good at making chips for the things that they did well at that point. And you talk about some of Gelsinger's decision making. That was rational until it wasn't in terms of some of the sort of warrants that he signed for them. and like I understand how that could happen to a company that was already really successful, unless I'm misunderstanding the history that we went through at the beginning of the episode. Some of those decisions were rational.

You're good, Andrew. I completely agree. No, I think you're totally right. I actually, I'm more sympathetic to Intel missing mobile than a lot of people for the exact reasons that you said. They miss mobile to the same reason Microsoft did, which like in the grand scheme of things, like on like an objective basis, it's bad. on the other hand, this is what happens. You're,

you're focused and optimize. That's right. That's right. I think, but that's why what Microsoft did, and it took him a while to get there, but what they ultimately did with missing mobile

and why I was so critical of Steve Balmer was he was he couldn't let go of the paradigm. He couldn't let go of Windows. And what Microsoft did was they let go. And they realized, okay, given the fact that we lost mobile, how can we participate in the future? growth of this industry, and that was through Azure and the cloud. And that prepared them for the AI wave to come along.

They were already ready for it. Intel didn't do that. The answer, and by the way, the cloud, a much lower margin business than Windows, they could step down to that because it offered opportunities for growth and participating in the future of technology. Microsoft pulled off what Intel did not. You're totally right, Andrew. It was understandable that Intel missed mobile.

But what they didn't do is they didn't pivot to leverage their strengths to participate in the future, even if it was lower margin. They couldn't do it. The board could not do it. And it was fine for a while because the cloud wasn't important. Intel sold all those processors in the cloud. The cloud went hand in hand with mobile. But in the long run, they're not a part of AI because they weren't investing like in video was. 20 years ago. The way to be a part of AI would have been to be a foundry making Nvidia chips.

The way to be a foundry making Nvidia chips is to have pivoted to be a foundry making mobile chips in 2010. And that is the great bifurcation of these two companies that came up together. Okay. So I'm going to make you chairman of the board for Intel right now. As they search for a CEO, how would you define success for Intel in five years? Like if you're chairman of the board, What's the North Star that they should be aiming for? Okay. It's over. We're just punting.

For all Gelsinger's flaws, you're not going to get anyone of his level and capability. Who's going to take this job, number one? No one's going to trust them ever. Like, why would you imagine anyone that had signed up, if Apple had signed up with Intel, it's only they, I guess, you know, I guess theoretically he wouldn't have been pushed out if that would happen. But regardless, the, the path is to figure out a way to spin out. So what should happen? They just become like AMD.

They become a fabless business. They make their chips on TSM, whatever. The real question is more a U.S. government question. What do you do about if you want a U.S. fab? That was my follow-up, yeah. I think the U.S. government was going to need to fund this.

And I think there's a bit where, of course, I'm generally anti, like the socialization of an industry or whatever. You just got to take it over. It's got to be built. And I think the way you get volume is there was this sort of, there was like a congressional report a few weeks ago about a Manhattan project for AI. Okay, let's do that. Let's do all up.

We're going to build our own AI systems. We're going to design our own chips. We're going to make our own chips and figure out what that looks like, what it is like. And I think there's a bit where we talked about the trade stuff. You want to bring manufacturing back. How do you do that competitively? Well, you integrate because U.S. has differentiation in software. And so you leverage that to overcome the higher cost of doing stuff domestically.

There's an extent to which Tesla sort of as has pulled this off to a certain bit. And is there a route of the U.S. building out its massive AI capability and actually integrating all the way down to the chip level and making sure that all the key pieces are domestic? I think you go more all in as the U.S. You take up the fabs. You say the volume is going to come from us. We're building our AI systems. We're going to build these hundreds of thousands,

million dollar chip clusters. So we're going to be the biggest customer. And over time, it pushes through and becomes successful and can bring on third parties because it's proved itself and maybe can be spun out to a private company again.

Again, this all sort of makes my skin crawl. But I think if, like the reality is, and this was, Pat Kelsey was a gift to the U.S. government because I do think there was a bit. And this came across in interviews. Like just being fairly, like, it was strategic to be patriotic because they needed government help.

But I also think he was patriotic and doing IDM 2.0 in defiance of Intel's reality because the U.S. needs its own foundry. That was a great gift to the U.S. federal government because they wanted that. And the reality is there's a bit where the incentives were just always a little bit off. It never made economic sense. If the U.S. wants this, they need to pay for it and they kind of need to own it. And I think that's the answer to the fab business. Interesting. And the red is you can kind of get it for free.

Like if Intel didn't have the fab business and just started making all their chips at TSM, they would immediately be a very profitable company, right? They're going to be fine. So the losses are all derived from the foundry business that is currently floundering. It'd be really hard to spin this out and figure out a way to do it effectively. It's not, like, AMD struggled for years and years and years after doing this. I'm not trying to say it's easy. I am saying the core Intel design business has a functional market.

It has products and they can make chips at TSM and make money. Okay. Good. Because a couple minutes ago, I thought we were just hanging up the Intel shingle for good after an illustrious 50-year run. But there's a version of Intel that doesn't need to shudder entirely. An Intel that doesn't make its own chips isn't a sadder Intel. Yeah.

Well, and in terms of funding the sort of project that you're spitballing here, There would need to be just way more money than the Chips Act has currently allocated to these companies. Right. I think you justify that by saying like this is a national security concern. We're allocating on the order of a Manhattan project in 2024. Okay. That's right. Yeah. Well, a lot of ground covered here.

And I'm sure there will be more to cover next week. We got an email about Bob Noyes that I'm rolling over to the mailbag. I'm excited to hit that. and who knows what else we'll get into. Maybe we'll talk more about the Manhattan Project that may or may not ever get off the ground here. But Ben, great energy across the board here. Any final thoughts before we sign off for the weekend?

Oh, it's sad. It's kind of a bummer. But I do wonder, I probably, like I said, I wasn't, it's the thing. I wasn't fully on board with the Elsinor plan, But I wanted to succeed. Yeah. And I wanted to succeed.

And like I said, from a board perspective, if you're going to say you support it, you got to support it. And it's a weird thing where they may be doing the right thing. And yet in a way that only just increases my disdain for them and their performance over the last 15 years. Absolutely. Boeing for the last 20 years. Boeing without the regulatory capture seems to be the Cliff Notes version of the Intel story. Yeah, Boeing with network effects. Yeah, well, unfortunately, they're all we got.

So maybe we can fold the Intel Foundry into the future here. For now, Ben, this was delightful. And I will talk to you on Monday for another Monday mailbag. People can continue to send questions. Email at sharptech.fm. And until then, I hope you have a great couple of days. I'll talk to you soon. Talk to you later.