TSMC raises capex and revenue forecast, highlighting growing AI chip demand - 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 powering the AI revolution, you are absolutely in the right place. Huge thanks to our sponsors keeping this show running. Ada, helping businesses integrate AI into their real world workflows. Ago Consulting, that's Ago, your go-to for silicon development from AI accelerators to full-sock design, and Zen Semiconductor, building the processors and AI fabric behind modern data centers, with ventures like their Sierra RISC vCPUs and Loom AI fabric, plus Ada for deploying AI at scale across any business. Today, we've got a packed episode TSMC is raising the roof on its forecasts. There's a wild smuggling case involving $300 million in Nvidia servers, Google's TPU ambitions are heating up. Someone ran a $500 and $502 billion parameter model at home, and we've got a spicy stock prediction for 2027

Let's get into it, all right? Kicking things off, TSMC has raised both its capital expenditure forecast and its revenue outlook, and the reason is pretty straightforward. AI chip demand is going absolutely bananas right now.

Now, if you're not deeply plugged into the semiconductor supply chain, you might be wondering why TSMC specifically matters so much. So let me give you the quick picture. TSMC, Taiwan Semiconductor Manufacturing Company, is essentially the world's most important chip factory. They don't design chips, they manufacture them. And almost every major AI chip you've heard of, Nvidia's H100s, H200s, the Blackwell series, Google's TPUs, Apple Silicon, they all flow through TSMC's fabs. So when TSMC raises its forecast, that's not just one company doing well, that's a seismic signal about the entire AI hardware ecosystem. Now let's talk CapEx capital expenditure. This is the money TSMC is putting into building and expanding its factories. Buying cutting-edge lithography equipment and generally scaling up its manufacturing capacity. These are eye-watering numbers we're talking about. We're already looking at a CapEx budget in the range of over $40 billion for this year, and they're raising that number? Think about what that means practically. Building a leading-edge semiconductor fab is like constructing a city. It takes years, costs billions, and requires some of the most precise engineering on the planet. The machines used EUV lithography tools from ASML cost around $200 million each. So when TSMC says it's spending more, it's not a rounding error. It's a fundamental bet on where demand is going. And the revenue forecast raise is equally telling. TSMC had already guided for strong growth, and they're now revising that upward. Why? Because their customers, Nvidia, AMD, Apple, Broadcom, and a growing list of custom silicon players are ordering more. AI training and inference workloads require enormous amounts of compute, and that compute needs to be manufactured somewhere. There is no AI without the fab. It's like the gold rush analogy people love to use. While everyone's fighting over who sells the picks and shovels, TSMC is the company that actually mines and refines the gold. What's really interesting here is the signal this sends about the longevity of the AI hardware buildout. There's been a lot of noise in the market about whether we're in an AI bubble, whether spending is going to pull back, whether hyperscalers are going to pump the brakes. And TSMC's revised forecast is essentially saying, no, the demand signal from our customers is strong and getting stronger. These companies plan years in advance. You don't raise capex guidance on a whim. This is forward-looking confidence backed by actual purchase orders and capacity commitments. So for anyone tracking the AI hardware space, and I know that's all of you, this TSMC update is one of the most important macro signals you can watch. When the world's most critical chip foundry is doubling down, you take notice. Okay, story two, and this one is genuinely wild. A California tech CEO has been arrested and is facing up to 20 years in federal prison for allegedly smuggling $300 million worth of Nvidia AI servers to China. And the way it was supposedly done is straight out of a spy thriller. Chips were routed through Malaysia and Singapore using falsified paperwork to obscure their final destination.

Federal prosecutors are involved, which means this is being treated as a serious national security matter, not just a trade violation. So let me give you some context here. Because this doesn't happen in a vacuum, the US government has placed strict export controls on advanced AI chips, specifically Nvidia's high-end data center GPS like the H100 and its successors. The concern is that these chips can be used to train and run military AI systems, and exporting them to adversarial nations like China could undermine US national security. These export controls have been tightening over the past few years, and Nvidia has even had to create watered-down versions of its chips like the H800 and A800, specifically designed to comply with export rules for the Chinese market by limiting certain performance characteristics. But $300 million in AI servers is not a rounding error. That's a massive quantity of hardware. To give you a sense of scale, an Nvidia H100 server, like a DGX H100, runs somewhere in the neighborhood of $200,000 per unit. So we're potentially talking about hundreds of individual systems. That's not someone slipping a couple of GPS in a suitcase. That is an organized, deliberate operation with supply chain logistics, shell companies and falsified trade documents. The routing through Malaysia and Singapore suggests the perpetrators were using those countries as transshipment hubs. Basically making it look like the final destination was somewhere in Southeast Asia when the goods were actually headed to China. This is a really important story for the industry, because it highlights just how valuable these chips have become and how the export control regime is creating significant black market incentives. When a piece of hardware is so restricted and so in demand that someone is willing to risk 20 years in federal prison to move it, that tells you something profound about the geopolitical weight that silicon now carries.