The Capacity Crunch.

**SPEAKER_1** (0:00)
Hi, everyone, and welcome back to Tech Insights The Chip-Observer, home to all things semiconductor related, covering industry trends, innovations, and key developments.
Today, we're looking at a story about capacity, how our current demands and world events affecting memory, raw materials, and even your favorite snack packaging. Recently, we organized a full webinar on the ongoing capacity crunch and its global effects. Posted by some of our experts, David McQueen, James Sanders, and Mike Walden. This episode walks us through some of the key insights affecting us, going deeper than just the headlines.
We're hearing a lot about chip shortages, factories running at full capacity, supply chains under pressure. But really, what's going on? If the chip shortage everyone's been talking about isn't actually a shortage at all, then what's happening inside the semiconductor industry right now is affecting everything. The devices you buy, the AI you use, and the global supply chains that power modern businesses.

**David McQueen** (1:00)
We're going to talk about one of the most pressing issues today, actually, the capacity issues in the semiconductor and supply chain.

**SPEAKER_1** (1:08)
So what do most people think the problem is? Most people assume the chip bottleneck is about logic, the processors that run AI workloads, laptops, and smartphones. And there's truth to that. Advanced node utilization at TSMC is running near capacity. But here's where the story gets more nuanced.

**David McQueen** (1:25)
Utilization for advanced nodes is running extremely high. So that's around about 90% this year. Projected to go up a little 94% in 2027, I think is the peak there.
And what's happening there is not only the mobile SoCs, which are high in volume, but then also HPC devices, AI devices, are beginning to put further strain on the available capacity.

**SPEAKER_1** (1:51)
So, what about TSMC's competition? Intel and Samsung. Their ability to pick up overflow capacity at the most advanced nodes matters a lot here.

**David McQueen** (1:59)
Intel, Samsung have, you know, as IDMs, they've kind of struggled a little bit, I think it's fair to say, in building up an external customer base for their, for their, uh, purely foundry side of their business. Um, but we have seen some very encouraging signs around 2 nanometer.

**SPEAKER_1** (2:18)
And that 2 nanometer breakthrough is significant. Intel shipped first, Samsung quickly followed. And for the first time in years, TSMC has credible competition at the cutting edge.

**David McQueen** (2:28)
Intel was first to market with 18A. Uh, that was the first, uh, 2 nanometer device launch was Panther Lake, which is, uh, actually again a mobile SOC, but in this case, laptop rather than, than smartphones. That's Intel using that capacity for its own manufacturing, so acting as an IDM there. But what we've seen since, that we've seen a lot of interest from customers in that process. And I think the successful launch of Panther Lake is a big proof point for Intel, that they can deliver a very, very advanced, literally the cutting edge process right now, I would say, uh, in volume, you know, on time.

**SPEAKER_1** (3:06)
So what is the real bottleneck? It seems that logic is tight, but that's not the real crisis. The real story is memory.

**David McQueen** (3:13)
The real big driver here is the need for HBM memory for those devices.

**SPEAKER_1** (3:17)
HBM stands for High Bandwidth Memory, and the demand for it has turned the entire memory market upside down.

**David McQueen** (3:24)
Again, if you just look at the units, it doesn't look like necessarily huge growth in that market.

**SPEAKER_1** (3:29)
The thing is that HBM devices have 16 or more layers of them. So the number of units coming out of the factory are not necessarily that high.
The wafers required to produce that actually are very high.

**David McQueen** (3:40)
The major memory manufacturers are chasing that market very, very hard. And fans are being re-engineered to serve HBM, basically looking at just under 100% capacity for legacy processes and running at 100% capacity for HBM.

**SPEAKER_1** (3:57)
So lead times now for memory are becoming particularly challenging. But what's the catch? Where's the trade-off? Memory manufacturers like SK Hynix, Micron and Samsung are spending heavily. CapEx is spiking dramatically.
But here's the counter-intuitive part. That investment isn't easing the crunch. It looks to be making things worse for everything except HBM.

**David McQueen** (4:17)
What we have seen this year, initially, we were projecting that CapEx would be slightly down. That was from the company's own financial statements, what they projected that CapEx would be in 2026 It was slightly down.
Now there's a massive spike up.

**SPEAKER_1** (4:35)
In particular, it seems to have been driven by the likes of SK Hynix, Micron and Samsung's Memory Division. They're very quickly re-engineering their fabs as fast as possible to chase that HBM market because there's such demand there.