Why Is Taiwan Semiconductor Manufacturing (TSM) Winning More 3nm AI Chip Work? - Yahoo Finance

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 making the AI revolution happen, you're in exactly the right place. Big thanks to our sponsors keeping the lights on here. Ada, which helps businesses integrate AI into their workflows without the headache. Ago Consulting, that's Ago, your go-to for silicon development from AI to so, and Zen Semiconductor, an absolute powerhouse building the processors and AI fabric behind modern data centers, with ventures like their Sierra RISC-V CPUs and Lume AI fabric, plus Ada for deploying AI at scale in any business. Today we've got a packed episode, we're talking about why TSMC keeps winning that premium 3nm AI chip work, Nvidia's surprisingly interesting move to make rival chips play nice together, the classic Nvidia vs AMD long-term investment debate, some overlooked companies quietly powering every major AI chip out there, and a wild smuggling story that you genuinely cannot make up. Let's get into it. All right, first up, Taiwan Semiconductor Manufacturing, better known as TSMC, is continuing to dominate the bleeding edge of AI chip manufacturing, and specifically, they're pulling in more and more 3nm work. And if you follow this space at all, this shouldn't be a complete surprise, but the degree to which it's happening is really worth talking about. So let's set the stage. When we talk about 3nm process nodes, we're talking about the absolute frontier of chip manufacturing. The nanometer number is a bit of a marketing label at this point. It doesn't literally mean every transistor feature is 3nm.

But what it does tell you is that you're getting incredibly dense, incredibly efficient transistor packing. We're talking billions of transistors crammed into a piece of silicon the size of your thumbnail. And right now, when you want to build a cutting edge AI accelerator, the kind that's training massive language models are powering inference at data center scale. You want the most advanced process node you can get. More transistors means more compute. Better efficiency means less power draw. And in the data center world, power is money. Like, literally every watt you save translates directly to operational cost savings. So why is TSMC winning this work? A few reasons. First, they have a massive head start. TSMC's 3nm node, which they call N3, is mature now. It's not experimental anymore. They've ramped production, they've worked out the yield issues, yield being the percentage of chips that come off the wafer actually working correctly, and they've got the capacity to handle enormous orders. Compare that to Samsung, who is TSMC's closest competitor in leading edge nodes. Samsung has had real struggles with yield at advanced nodes. There have been reports of customers quietly shifting work back to TSMC, because Samsung just couldn't hit the yield numbers needed to make production economically viable. When you're building chips that cost thousands of dollars each to the end customer, yield is everything. Intel Foundry is the other would-be competitor here. And look, Intel has huge ambitions with their 18a node, and I genuinely want to root for them because competition is good for everyone, but they're still proving themselves as a foundry.

They don't have the decades of trust and customer relationships that TSMC has built. And trust is a huge deal in this business. Think about it. You're handing over your most valuable intellectual property, your chip designs, to a manufacturer. You need to know they can execute. TSMC has a track record that's basically unmatched. Apple, Nvidia, AMD, Qualcomm, all the big names run their most important silicon through TSMC.

The AI chip boom specifically is supercharging TSMC's 3 nanometer business, because the demand for AI accelerators is just relentless. Every hyperscaler, your Googles, your Microsofts, your Amazons, your Metas, they're all designing custom silicon now. And all of those custom AI chips need a home. And increasingly, that home is TSMC's N3, or N3 E-Nodes. Nvidia's Blackwell architecture, for example, that's on TSMC 4 nanometer, and their next generation is going to push further down the curve. Apple's M series and A series chips are on TSMC 3 nanometer. Google's latest TPS are TSMC manufactured. The list just goes on. What does this mean for the broader industry? Well, TSMC's dominance at the leading edge creates this interesting geopolitical dynamic that you can't ignore. The vast majority of the world's most advanced chips are manufactured in Taiwan. That concentration creates risk, and it's why you see so much policy activity around building out semiconductor manufacturing in the US and Europe. The CHS Act in the US was specifically designed to incentivize companies like TSMC to build facilities stateside.

TSMC's Arizona fabs are under construction and coming online, though there have been delays and challenges getting those up to full speed. But even when they do come online, the capacity will be a fraction of what TSMC operates in Taiwan.