AI Chip Demand Keeps TSMC Stock And Large Cap Tech Shares In Play - simplywall.st

Skyler Monroe (0:10)

Hey, everyone, welcome back to the AI Hardware Show. I'm your host, Skyler Monroe, and if you're into chips, silicon and the hardware actually powering the AI revolution, you are absolutely in the right place. Big thanks to our sponsors today, Ada, helping businesses integrate AI seamlessly into their workflows. Ago Consulting, that's Ago, your go-to for silicon development from AI to SoC, and Zen Semiconductor, an incredible umbrella company building the processors and AI fabric behind modern data centers, with ventures like their Sierra RISC-V CPUs and Loom AI fabric, plus Ada for deploying AI at scale in any business. Today, we've got a packed episode. We're talking TSMC and the AI chip demand wavekeeping markets hot, a wild HBM shortage hitting Chinese chip prices hard, some interesting gray market GPU deals popping up in Dubai of all places, a really cool research paper on making HBM more efficient for inference workloads, and the latest on whether Nvidia can even compete in China anymore. Let's get into it. Alright, kicking things off with a story that investors and hardware folks alike are watching closely. AI chip demand continues to keep TSMC's stock and large-cap tech shares very much in play. And honestly, if you've been paying any attention to the semiconductor market over the last couple of years, this is not a surprise, but the scale of it continues to be jaw-dropping. So let's break down why TSMC is essentially the backbone of this whole conversation. TSMC, Taiwan Semiconductor Manufacturing Company, is what we call a pure play foundry. They don't design their own chips. Instead, they manufacture chips for basically everyone who matters. Nvidia, Apple, AMD, Qualcomm, you name it. So when AI chip demand surges, TSMC is one of the first places that surge shows up, because almost every cutting-edge AI accelerator gets fabbed there. Think of TSMC as the world's most sophisticated bakery. They don't come up with the recipes that's Nvidia, AMD, and the hyperscalers, but they're the only kitchen advanced enough to actually bake the most complex ones, and right now, everyone wants their bread. The demand for AI accelerators, your H100s, your Blackwell GPS, custom ASICs from the hyperscalers like Google's TPS and Amazon's Tranium chips, all of that is flowing through TSMC's advanced nodes, particularly their 3nm and 4nm processes. And that's keeping TSMC's order books packed. Now, why does this ripple out to large cap tech shares more broadly? Because the companies spending the most on AI infrastructure, Microsoft, Google, Amazon, Meta, are also some of the biggest players in the stock market. When investors see that AI infrastructure spending is robust and that the underlying chip supply is there to support it, that gives confidence to the whole ecosystem. It's almost like a positive feedback loop. Chip demand stays high, foundry revenue stays high, hyperscaler buildouts continue, and the market stays bullish on the whole AI hardware stack. Now, there are some risks here, geopolitical concentration being a big one. TSMC is predominantly based in Taiwan, and that's a conversation that keeps coming up in boardrooms and in Washington. But for now, the demand signal is strong, and TSMC is investing heavily in capacity, including their fabs in Arizona and Japan. The bottom line here is that AI is not a flash in the pan demand driver. The infrastructure investment cycle we're in has real legs, and TSMC's continued strength is one of the clearest signals of that.

Okay, next story, and this one is really interesting from a supply chain and geopolitics angle. Chinese AI chip prices are surging, and the culprit is HBM shortages. So let's unpack this because there's a lot going on here. HBM stands for High Bandwidth Memory, and if you're not familiar with it, here's the quick explainer. Modern AI chips, your GPS and AI accelerators need to move absolutely massive amounts of data incredibly fast. Regular DRAM just can't keep up. HBM stacks multiple layers of memory dyes on top of each other, and connects them with thousands of tiny vertical connections called through silicon vias. The result is enormous memory bandwidth, we're talking terabytes per second, which is exactly what training and running large AI models demands.

Now, here's the problem for China. The primary manufacturers of HBM are SK Hynix, Samsung, and Micron.

SK Hynix, in particular, has a dominant position in the highest end HBM, they're the one supplying the HBM3E that goes into Nvidia's H200 and Blackwell chips. And because of US export restrictions, Chinese companies can't just buy whatever HBM they want from these suppliers. So you've got Chinese AI chip designers companies like Huawei with their Ascend line, and a bunch of emerging domestic players who are trying to build competitive AI accelerators, but they're hitting a wall when it comes to sourcing HBM. And when supply is constrained and demand is high, prices go up. It's economics 101, but the implications here are really significant. Think of it like this. Imagine you're trying to build a race car, and you figured out the engine design, but the only suppliers of the high-performance tires are under trade restrictions. You can maybe get some tires through other channels, but they're going to cost you a lot more, and you might not get the best ones. That's roughly the situation Chinese AI chip companies are in right now.