(Preview) The Roots of a Global Memory Shortage, Thick, Thin and Apple, Shopify is Fine, Actually

Ben and Andrew discuss the global memory shortage and answer a listener’s question: how did this happen? Topics include: What memory chips have in common with logic chips, why Intel left the memory market in the 1980s, how the international shipping market explains today’s shortage, how major players will address the problem going forward, possibly with some help from the Chinese. From there: A look at the “thin” future of AI computing as memory and performance takes precedent, AI computing that will still be done on device, and Apple’s AI strategy. At the end: Responses to Shopify concerns in the AI era, a question about the future of brands as AI finishes what digital advertising started, Shopify preferencing its own AI tools, and a looming test of Apple’s commitment to the Vision Pro.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Andrew Sharp?

  Hello and welcome to a free preview of Sharp Tech.

• 0:08Andrew Sharp?

  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?

• 0:20Ben Thompson?

  Dismayed, Andrew. Dismayed at our wildly, wildly diverging trajectories.

• 0:26Andrew Sharp?

  I'm feeling good. It's good to be back. I'm going on vacation in a couple of days. My life is good. How about you?

• 0:34Ben Thompson?

  Well, that's exactly it. You're going on vacation. We will not be recording next week because you'll be sitting on a beach.

• 0:39Andrew Sharp?

  Yep.

• 0:41Ben Thompson?

  I've spent the week visiting vet clinics because my dog unfortunately has gone blind, which is very sad.

• 0:48Andrew Sharp?

  It is.

• 0:48Ben Thompson?

  Eight and a half years old. There's something called SARS, which I hadn't heard about. It's like sudden, sudden acute retinal detachment syndrome. I might've got the SNA wrong, but apparently it happens sometimes. It's a small percentage of dogs, but they're just, their eyes become basically disconnected. There's like wildly varying theories about why it happens, but it happens very rapidly and they are just suddenly blind for life.

• 1:12Andrew Sharp?

  Yeah. And it involved a couple of hours and multiple visits to the vet this week. I hope Coco can adjust to this new reality. Apparently dogs can adapt without that much trouble.

• 1:24Ben Thompson?

  Well, I mean, there's a bit where like in retrospect, I can kind of mark when it was deteriorating, but it wasn't gone. And then she's like suddenly running into walls.

• 1:33Andrew Sharp?

  Oh, boy.

• 1:34Ben Thompson?

  So she still finds her way around the house pretty well, but the reason to bring it up is not only for everyone to feel bad for me, but if anyone has had a blind dog or dealt with this, open to tips.

• 1:49Andrew Sharp?

  Write in. Email at sharptech.fm. We'd love to hear from you. But yes, no podcast next week.

• 1:57Ben Thompson?

  So yeah, you're going to be sitting on a beach and I'm going to be, you know, feeling sad about my dog.

• 2:01Andrew Sharp?

  Grinding through. Yeah. Well, look, at least it's warmer where you are than it was about a month ago because I was there. 20 degrees below zero. You had like

• 2:11Ben Thompson?

  It was actually 27 degrees below zero. It was 63 degrees yesterday, which was a 90 degree swing in a matter of like three weeks.

• 2:18Andrew Sharp?

  Unbelievable, you know, so things are looking up out there in Wisconsin. For now, as for the show, we are going to begin with memory and a global memory shortage that is underway. Bloomberg reported on Sunday that Sony is now considering pushing back the debut of its next PlayStation console to 2028 or 2029. Nintendo is considering raising the price of the Switch 2 in 2026. Samsung is now reviewing memory contracts on a quarterly basis. Various Chinese phone companies are slashing production because of memory issues. So I want to talk about all that and how it relates to AI.

• 2:57Andrew Sharp?

  But I want to start with a note we got from Jonathan. Jonathan said, goats, help me understand. I love.

• 3:05Ben Thompson?

  Oh, have I been promoted to being a goat?

• 3:08Andrew Sharp?

  Yeah, you are definitely the third chair on goat. All things considered, Jonathan says goats.

• 3:13Ben Thompson?

  Hey, look, you know, I've always talked about on my tombstone will probably be like author or stratechery. But I'm equally proud of sort of the entire ecosystem of paid newsletters, the Substack world, all those sorts of things. But what I'm truly proud of is when Sports Illustrated slashed their whole staff, Open Floor Podcast was ended. I was on the horn with you and Ben, barely knew you guys at that point, saying, this must continue. You are going to do a paid podcast. I'm going to help you do it. And so, yes, on my No Tech Ben Twitter profile, it does still say the patron saint of the goat. Patron, patron, whatever it is.

• 3:53Andrew Sharp?

  You, look, you've been there from the very beginning, that first Zoom call. I'll always remember that. This is, of course, a reference to Greatest of All Talk, my NBA podcast. That's for the 80% of the sharp tech audience that isn't a basketball fan. Jonathan is though, and he says, goats, help me understand the dynamics between memory and chips. My normie understanding is chip fabs require tens of billions of dollars of investment and three to five years to manufacture. High fixed costs and long lead times create tenuous economics. If the fab is not full, the company could lose billions. So how does this compare to memory? Why are we experiencing a memory shortage?

• 4:32Andrew Sharp?

  Again, my normie understanding is the memory manufacturers are prioritizing high bandwidth memory for AI data centers. I'm assuming high bandwidth memory is more expensive with better margins compared to quote unquote, regular memory that goes into phones, laptops and Playstations. I'm also guessing the big tech companies will pay through the nose to acquire as much high bandwidth memory as possible. So what is stopping someone from building capacity to serve the low end market of non-high bandwidth use cases? Is it just as expensive and risky to build new memory capacity? Have the regular memory fabs been converted to high bandwidth memory fabs? When will I be able to afford a new laptop?

• 5:17Andrew Sharp?

  When will the next generation of video game consoles come out? Who will win the NBA championship? I didn't read that last part of the email until just now. I gotta go with the T Wolves, Jonathan. But, um, oh, I would love that. Before we get to what you wrote, I just, I appreciated what Jonathan wrote because I had the exact same questions. I've been hearing about the memory shortage for like 18 months now because it's actually been a bigger problem in China. How did this happen and how will it possibly be addressed over the next four or five years?

• 5:48Ben Thompson?

  Yeah, when I was reading this email, I'm like, I probably should have, like, I've referenced just the memory issue in passing for also for like months and months and months.

• 5:56Andrew Sharp?

  Yeah.

• 5:57Ben Thompson?

  And so it's clearly a thing that's happening. But I realized that I probably should have done a more all up sort of piece around it than sort of just mentioning it in passing. And I think this email is illustrative because there's certain things to understand how it works that will actually answer a lot of these questions. And I've been derelict in doing that.

• 6:18Andrew Sharp?

  Well, here we go. That's what the podcast is for.

• 6:20Ben Thompson?

  I'm gonna do my best here to sort of get at what's going on. Okay, so memory and logic, they both require being on the leading edge. So advanced memory fabs today also use EUV as an example. In fact, I think, and I should have checked this. I think that some memory manufacturers might have been using EUV. No, I think maybe TSMC was first, but like, the point being is like, they are just as advanced as a logic sort of fab is. And they're just as expensive. We're talking in the, like, the 20-plus billion-dollar range to build a new fab.

• 6:58Andrew Sharp?

  Okay.

• 6:58Ben Thompson?

  So a lot of those issues apply here. What makes memory fundamentally different than logic is a logic, it's kind of funny. It's like the difference between a, I guess this is not funny, it makes total sense. It's just a CPU and a memory chip. They're all chips, okay? The difference is a CPU is a very complicated chip with a bunch of different things. Because it's a, it's a central processing unit. It's a, it's almost like the CPU should have been GPU as a general purpose processing unit.

• 7:26Andrew Sharp?

  The CPU is for doing everything you can possibly do on a computer.

• 7:30Ben Thompson?

  That's right. And even a GPU has many more functions than a memory chip might do. And so what that means is to do logic well, you need the capability of designing all these different things and then making a chip that does all these sorts of different things. And so there's lots of fact, so you go back, like, the memory story people are probably in tech the most familiar with is sort of the story of Intel leaving memory and to go into logic. So back in the day, memory was where everything was. That was the most important thing. That's where people made the most money. Intel was in memory sort of to start. Along the way, they started making logic chips.

• 8:11Ben Thompson?

  And what happened was they were losing their shirt on memory to primarily Japanese suppliers. And meanwhile, they're making a lot of money in logic. So there's this very famous story of Andy Grove and Gordon Moore. Intel's like practically going bankrupt.

• 8:26Andrew Sharp?

  Yeah.

• 8:26Ben Thompson?

  And they're like, the story, you know, Andy Grove says to Gordon Moore, whatever, if we were outsiders, if we hadn't grown up with Intel, if we didn't identify as a memory Manufacturer is that being our core business.

• 8:37Andrew Sharp?

  Why would we ever go into this business?

• 8:40Ben Thompson?

  And people came in, what would they do?

• 8:41Andrew Sharp?

  Yeah.

• 8:42Ben Thompson?

  They're like, we would get out of memory. And that's, and so Intel exited memory in the early eighties. And focused just on logic, where if you think about it, it's not just about manufacturing. It's also about the design component. And Intel was an integrated device manufacturer, which means they did both the design and they manufactured it. And it's that design bit that increased the value of the chips that they made. Because they could do these sorts of things. And then they focused on that and obviously were fabulously profitable and made all this money and locked in their position. As being the dominant provider first for, you know, first for Windows and then for servers sort of broadly. With AMD sort of dipping at their heels all the time.

• 9:24Andrew Sharp?

  They did that because the Japanese manufacturers, for instance, couldn't match the design capabilities that Intel had. And so that.

• 9:32Ben Thompson?

  Well, it's not just design capabilities.

• 9:33Andrew Sharp?

  Pricing power.

• 9:34Ben Thompson?

  There's a software angle, which is like, does this software work with this specific chip? So they had the x86 instruction set, which they, going back to IBM, forced them to share it with AMD to be a second supplier. That's why AMD has always been there along with Intel. That was a condition of IBM using the Intel chip in the first PC. Is that they had to second source. So they had to share the x86 spec with AMD, give them a license to it. There is a whole string of litigation about like, to what extent do they have to share? Do they try to cut them off, adding custom instruction sets? But it's differentiating hardware with software at a very low level. But that was the moat.

• 10:14Ben Thompson?

  The reason why it was so profitable to be in logic for, for Intel is only they could make x86 chips. Again, with AMD, there's lots of complexities here. But the point being is there was a higher order of differentiation. What makes memory different? And the reason why they're getting their clock cleaned by the Japanese in the late seventies and early eighties is memory is standardized. All it is, it's a much more basic chip. Now, it's basic in the fact that speed still matters, bandwidth still matters, all these things matter. So it benefits greatly from being on the leading edge, but it is a commodity market by and large. Now, we'll get into the details why it's not fully a commodity market. But it is basically relative to logic, it's a commodity market.

• 10:57Andrew Sharp?

  Mm hmm.

• 10:58Ben Thompson?

  In a commodity market, the price is set by the marginal unit of memory. So everyone pays the same price based on like basically the last memory chip that's created and how much they can sell it for. Because you could just go, you could just switch easily and go use someone else. In a commodity market, your profits are dictated by your cost structure.

• 11:20Andrew Sharp?

  Right.

• 11:21Ben Thompson?

  So if you, so like we talked about with Amazon, with AWS, one of the reasons they did so well in the CPU-denominated cloud era is because they did all this work to basically have a lower cost of serving than Microsoft or than Google. So if you have a unit of compute in the cloud, Amazon provides that unit of compute more cheaply than Microsoft or Google does. So if you have someone that's just picking purely based on price, then Amazon makes more money because their cost structure is lower.

• 11:53Andrew Sharp?

  Right.

• 11:54Ben Thompson?

  So below the marginal cost. So the, you know, that's not, again, that's not a perfect example either because there's tons of software that goes in the cloud, trying to lock people in and all those sorts of pieces. But basically, there's two ways to make money in business. Way number one is you have a differentiated product so you can charge more than it costs you to make.

• 12:13Andrew Sharp?

  Mm-hmm.

• 12:14Ben Thompson?

  And then you make a lot of money.

• 12:15Andrew Sharp?

  Better product or better cost structure is the two paths.

• 12:18Ben Thompson?

  The other way is, in a commodity market, your cost structure is lower and the cost is dictated by the, so if you have three providers, one provider, they're forced down to basically their marginal cost.

• 12:30Andrew Sharp?

  Mm-hmm.

• 12:30Ben Thompson?

  And for a marginal cost, and this is why fabs are so hard, because the marginal cost is zero, basically. Okay? So you have to buy out this EUV machines, you have to buy all this other equipment, all of which is depreciation that's going to hit your balance sheet, but you are forced, the price has dictated you by the market. Now, you're not going to, you have to buy silicon wafers. That's your main input cost. So you're not going to price below the cost to buy silicon wafers. At that point, you might as well shut down the line, but you will produce, as long as you're making more than it costs you, your inputs. But that means you're also going bankrupt at the same time because you spent all this money that you're not covering your fixed costs for to buy all the equipment.

• 13:11Andrew Sharp?

  It takes that much longer to profit enough to cover the upfront investment.

• 13:16Ben Thompson?

  Yeah, a really interesting industry that I by chance happen to know a lot about, thanks to people that I knew in Taiwan, is shipping. And shipping is actually a good analogy for this. So if you think about what is the price of freight? If you have a ship, it's basically the cost of fuel and crew. But and then there's dock fees and all those sorts of things.

• 13:37Andrew Sharp?

  Yeah.

• 13:37Ben Thompson?

  So if you have a bunch of ships in the market and if there's more supply for shipping than there is sort of demand for stuff to put on there, the price is going to fall to that level, to the cost of fuel, the cost of crew and your fees, you sort of at various ports or whatever it might be. That doesn't include the cost of the ship itself, right? So the issue is, is that that cost of the ship itself has already been paid. And so as long as the price you get that you can charge is higher than the marginal cost, the marginal cost of the operational cost to actually deliver a container, you're still going to be running that ship, even though on your books, you're bleeding money

• 14:22Ben Thompson?

  because this depreciation for that ship is hitting the balance sheet. And so that's why companies will operate while losing money because they're still making money on a marginal basis. They're just not covering their upfront costs. And this is all shipping company. That's why it's tricky to buy ships because you're buying them for 30 years. And so that's what happened during COVID, for example.

• 14:45Andrew Sharp?

  All right, and that is the end of the free preview. If you'd like to hear more from Ben and I, there are links to subscribe in the show notes, or you can also go to sharp tech.fm. Either option will get you access to a personalized feed that has all the shows we do every week, plus lots more great content from Sir Techery and Mr. Techery Plus bundle. Check it out, and if you've got feedback, please email us at email at sharp tech.fm.