Inside the Memory Squeeze: A conversation with Claus Aasholm, Founder of Semiconductor Business Intelligence

As memory pricing tightens and demand surges, Fusion Worldwide’s Market Research Manager Andrew Czuczwa sat down with Claus Aasholm, Founder of Semiconductor Business Intelligence and a seasoned analyst of the global semiconductor supply chain, to unpack the forces driving today’s market and what they signal for the next semiconductor cycle. Their discussion offers a candid look at the factors shaping supply, pricing, and sentiment across the memory landscape.

The following transcript has been lightly edited for clarity and flow.

Andrew Czuczwa:

Thanks for taking the time to meet with us today, Claus. We really appreciate it. We’re excited to get your perspective on what’s happening in the market right now. Right now, everybody’s talking about the memory market. Since we’re an independent distributor, we’re often on the leading edge of these issues. Customers start reaching out to us first whenever they can’t get support. Over the past few months, it’s been busy — DDR4 DRAM chips, DDR4 modules, and now, with increased data center demand, DDR5 RDIMMs. What areas of the memory market do you see most at risk for supply issues through the rest of this year and into 2026? And what do you think is driving that?

Claus Aasholm:

Yeah. So, you know, the answer is always AI. I don’t know what the question is, but it’s always AI. That’s what drives everything at the moment. But there are a couple of underlying issues.

We have what we call the semiconductor cycle, and in reality, it’s really a memory cycle. You know that very well. If you’re a distributor, you know we’ve got a couple of years where things are fun and a couple of years where they’re not so fun because prices are dropping. It’s really the memory products that are driving this cycle. During the upturn, at a certain point, DRAM gets so expensive that people stop buying PCs, phones, and servers. That’s been the traditional way of doing things.

Even corporates stop buying DRAM for upgrades. In the old days, cloud data centers would say, “I’ll wait a little bit before upgrading.” But now the demand is not driven by corporate OpEx or consumer spend anymore, it’s driven by infrastructure spend.

So, what happens then is, first of all, the big memory companies they are really running at full capacity at all times now. They’re not shutting down capacity during the down cycle; they are producing at full speed, all production lines, 24/7, 365. Even when they’re running at a loss, they still run at full capacity.

This last down cycle was the deepest the memory companies have ever seen. Normally they run at a loss during the down cycle, but this time they had negative gross margins.

It was brutal, and they really got scared. So, you saw that when memory pricing started to firm again and they got back into the black, they didn’t invest in CapEx. They were too scared. They were worried — “What’s happening here?” So, we’ve seen a delay in CapEx investment during this upside.

They’re getting there, but they’re still behind the curve.

What happens inside these companies is that they’re not here to serve the market; they’re here to serve their shareholders. That’s what drives decisions. And you’re right to point out that we’ve seen, over the last two quarters — mostly in the last quarter — that DDR4 has been squeezed. And now the time has come for DDR5. So, I think this is the immediate thing happening right now. This is going to be brutal, in my opinion.

Andrew:

Yeah, that lines up. A lot of our customers started switching from Samsung, Hynix, Micron to Winbond and Nanya once DDR4 started tightening.

Claus Aasholm:

Exactly. And when Micron reported — I think it was about ten days ago — their margin on non-AI products increased by 18% with shipments flat. So, all the automotive, industrial, and consumer customers are paying 18% more for exactly the same shipments. That tells you there’s no price elasticity anymore. They’re just screaming to get product before the next guy does.

Andrew:

That’s a strong signal.

Claus Aasholm:

Right. If you look at the broader picture, the big three memory makers — Samsung, SK Hynix, and Micron — are running flat out. During the down cycle, DRAM and NAND get close in revenue because NAND is more resilient — it serves industrial and automotive customers. But in the upturn, DRAM outruns NAND. And right now, the memory companies only care about DRAM — and inside DRAM, they only care about one thing: HBM. The profits are huge. HBM3, then HBM3E 8-high, now 12-high, and soon HBM4. Each generation needs more capacity per bit because stacking reduces efficiency. You need room for vias. So, they need more wafer capacity to make the same amount of bits.

And there’s only one place to take it from: DDR5 and DDR4. All of them have said they’re phasing DDR4 out. They haven’t said much about DDR5 yet, because it’s used in data centers, but that will come.

This is going to be a tough market.

Andrew:

So from a tools and CapEx standpoint, what are you seeing?

Claus Aasholm:

If you look at semiconductor equipment spending, it’s grown overall — which is normal for the market — but the balance has shifted. We used to be close to 50/50 between logic and memory tools. Now we’re closer to one-third memory. That’s underinvestment. You can see memory CapEx lagging badly, about 7.2 billion per quarter, which is lower than the last peak. They’re behind the curve. And the truth is, they don’t care. This is a margin game now, not a volume game. They like tight markets; they like saying no.

As a distributor, you know that dynamic very well.

Andrew:

Exactly. When they tell customers “maybe in two and a half years we can ship,” that’s when they come to us.

Claus Aasholm:

Exactly. And that’s why I think this is a chance for you to warn customers. This isn’t the boy who cried wolf. You’ve got the data. Look at utilization, margins, and tool investment. It’s clear that the industry hasn’t invested enough to support this level of demand.

Andrew:

Let’s move to analog and the hybrids. Many believe analog makers kept capacity low after the over-inventory period of 2023 and 2024. Do you think they’re still running under capacity?

Claus Aasholm:

Yes. If you look at their property, plant, and equipment values — the financial proxy for capacity — these companies more than doubled their capacity earlier this decade, but utilization dropped hard during the downturn. Renesas, for example, was at 33% utilization in Q4 and recently climbed to around 50%. That’s still very low. Inventories remain high across the board. Some have even raised prices on mature lines — not because of fab shortages, but because of bottlenecks elsewhere. For example, some analog parts depend on memory. If memory’s constrained, certain analog projects stall. Unlike memory companies, analog and hybrid manufacturers will actually reduce capacity if needed. They’re profitable even in downturns. The memory guys don’t have that luxury. They just keep running full speed.

Andrew:

What’s happening at the foundry level?

Claus Aasholm:

Foundries are a different story. Total wafer capacity continues to grow, but utilization, which dropped during the downturn, has been climbing again. We’re now in the low 80s in terms of utilization, maybe 82%, depending on the source. TSMC is right up there. They’re prioritizing leading-edge, but the rest of the foundry landscape — the nodes that serve analog, MCU, and mixed-signal — is filling up again too.

We’re about three-quarters of the way to full capacity across the foundries. The difference is in mix. What the foundries produce for fabless and hybrid companies is closer to fully utilized. What the hybrids produce internally — things like analog and power — is still well below normal. That means pressure will stay uneven. Mature-node capacity will stay tighter than many expect. And then there’s China. SMIC and Hua Hong are major players there. SMIC in particular has strong quality and competitive pricing. It’s not about technical capability, it’s about geography. A lot of mature-node capacity is now in China. That’s fine technically, but politically it’s risky. Western companies are trying to isolate China, yet they still rely on that capacity. It’s like cutting your oxygen supply. You can’t do it cleanly. I’ve worked with companies that used to have production at SMIC. Fantastic service, excellent quality. But geopolitics made it too risky. Moving production wasn’t easy, and most didn’t want to but they had to. So yes, foundry tightness will continue, especially at mature nodes. Geography will play a big role in sourcing and compliance decisions going forward.

Andrew:

With all that in mind, do you think demand continues like this into 2027 and beyond or are we looking at a bubble?

Claus Aasholm:

I think we’re still in the early innings. What we’re seeing now is the first phase of a larger shift.

People compare this to the dot-com bubble, but back then there were no revenues, just valuations. This time, there’s real demand, real production, and real profits. We may see a stock bubble, sure. But a stock bubble isn’t an industry bubble. The structural growth in compute demand - especially for advanced processing and high-speed memory - is real.

Andrew:

Final question — you’re known for your skepticism of vendor “funnels” or sales forecasts. Why?

Claus Aasholm:

Because I’ve been inside those systems. A funnel is really just a project database. It’s supposed to help salespeople and engineers prioritize opportunities. The problem is when CEOs start using funnel size as a metric in earnings calls. Once that number has to keep going up, sales teams inflate it. Customers play along because it helps them negotiate better prices. Suddenly, opinion gets presented as data. I prefer to look at financials, utilization, inventory — real indicators. When companies justify overbuilding or overspending with “our funnel is strong,” I call it out. It’s not data; it’s a story. After one of my articles on ADI, two other CEOs called me to offer clarification. They didn’t want that same treatment. So yes, I’m tough but that’s because someone has to ask the hard questions.

Andrew:

That’s completely fair and refreshing to hear. Claus, thank you so much for taking the time to speak with us today.

Claus Aasholm:

My pleasure. It’s been good fun. I hope it’s been useful for you as well.

For ongoing insights into memory market dynamics, follow Claus Aasholm and Semiconductor Business Review on LinkedIn and Substack.