GlobalFoundries Reaches Supply Agreement with TSMC for AI Chip Co... - Industrial Info Resources

Skyler Monroe (0:10)

Hey, everyone. Welcome back to the AI Hardware Show. I'm your host, Skyler Monroe. And if you're into the chips, the silicon, and the infrastructure powering the AI revolution, you are absolutely in the right place. Big thanks to our sponsors today, Ada, helping businesses integrate AI into their workflows seamlessly. Ago Consulting, that's Ago, your go-to for silicon development consulting from AI all the way to Sozi Design. And Zen Semiconductor, 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, TSMC is everywhere, supply deals, Texas expansions, and demand they literally cannot keep up with. Microsoft is making a bold move to cut its Nvidia dependency. And we're going to talk about why soaring AI chip earnings might actually be a warning sign for some big name stocks. Let's get into it. All right. First up, GlobalFoundries and TSMC have reached a supply agreement focused on AI chip co-production. And this one is genuinely interesting because it's not a story you'd expect to hear.

These two companies are usually thought of as competitors, both operating in the foundry space. They make chips for other companies. So what's going on here?

Well, the short version is this. Demand for AI chips has gotten so enormous that even the biggest players in the world can't go it alone anymore. And that's kind of a wild thing to sit with for a second. We're talking about TSMC, the most advanced semiconductor manufacturer on the planet, essentially saying, we need a partner to help us keep up. Now GlobalFoundries and TSMC occupy pretty different parts of the market. TSMC is at the bleeding edge, we're talking 2nm-3nm processes. The kind of nodes that power your latest Nvidia GPS and Apple chips. GlobalFoundries on the other hand operates at more mature nodes, think 12-14nm and above. And for a long time, people kind of wrote off mature nodes as old news. But here's the thing, not every component in an AI system needs to be built on a bleeding edge process. Think of it like building a sports car. Yeah, you need a high-performance engine, that's your cutting-edge AI accelerator. But you also need brakes, seats, a steering wheel, components that don't need to be made from the most exotic materials. In the chip world, those are things like power management chips, IO controllers, interface chips. And those can absolutely be made on mature nodes. So what this agreement likely represents is a kind of division of labor. TSMC focuses its most advanced capacity on the highest-complexity AI chips, while GlobalFoundries handles some of the supporting silicon. It's a smart move, honestly, and it reflects a broader industry shift, where co-production and chiplet architectures are becoming the norm rather than the exception. The chiplet model is worth explaining here because it's central to why this deal makes sense.

Instead of building one massive monolithic chip that does everything, modern AI processors are increasingly built from multiple smaller chips, called chiplets, that are stitched together using advanced packaging. Each chiplet can be optimized and manufactured at the most appropriate process node for its function.

So you might have your compute die on TSMC's most advanced node, and your memory interface or IO die on a more mature GlobalFoundries process.

This is exactly what companies like AMD and Intel have been doing for years with their CPU and GPU designs. And now, with AI driving insane demand, the Foundry ecosystem is essentially being restructured around this multi-vendor, multi-node philosophy. From a geopolitical standpoint, this is also notable. GlobalFoundries has major fab capacity in the United States, specifically in upstate New York and Vermont, and in Europe. So a supply agreement like this isn't just a business decision. It has real implications for supply chain resilience and national semiconductor strategy.

The bottom line here, this isn't just two chip companies shaking hands. It's a signal that the AI hardware supply chain is maturing, diversifying, and recognizing that you need the entire ecosystem working together, not just the tip of the spear. And that's good news for the long-term stability of AI hardware supply.

Okay.

Story number two.

TSMC is exploring a major expansion into Texas. And the reason is pretty straightforward. AI chip demand is massively outpacing what they can currently produce. And look, if you've been following the semiconductor industry at all over the last couple years, this is a sentence you've heard a lot. But it keeps being true. And the scale keeps getting bigger. Let's put this in perspective. TSMC is already building in Arizona. They've committed to multiple fabs there, billions of dollars in investment, the whole deal. And now they're looking at Texas as another potential site. That tells you something about the magnitude of the capacity problem they're facing. Texas makes a lot of sense as a location, by the way. You've got a business-friendly regulatory environment, significant existing tech infrastructure. Samsung already has a fab in Taylor. Texas access to a large engineering talent pool. And the state has been very aggressive about attracting semiconductor investment. It's also got better water access in certain regions compared to the Arizona desert, which matters a lot for chip manufacturing. Now, why is demand so far ahead of supply? This is the core question, and it comes down to a few compounding factors. First, building a semiconductor fab or fab for short, is one of the most capital intensive and time consuming industrial projects in human history. We're talking about facilities that cost 10, 20, even 30 billion dollars to build, take three to five years to construct, and require extraordinarily precise environmental controls. You can't just spin one up in six months because demand spiked. Think of it like an oil refinery, but the tolerances are measured in atoms. The machines inside these fabs, the EUV lithography systems from ASML, cost around 200 million dollars each, take years to manufacture, and ASML can only produce a limited number per year. So the supply chain constraints are layered on top of each other at every level. Second factor, the AI compute build out is accelerating faster than almost anyone predicted. Every major cloud provider, Microsoft, Google, Amazon, Meta, is spending tens of billions of dollars on AI infrastructure. Every enterprise is trying to figure out their AI strategy. The demand signal coming into the semiconductor industry is unlike anything we've seen, maybe ever. And it's not just training large language models anymore. Inference actually running AI models to get answers is becoming a massive workload too, and it requires dedicated, always on hardware at scale.