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 this is the podcast where we dig into the silicon, the systems and the infrastructure powering the AI revolution. 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 from AI chips to full-soil design. And Zen Semiconductor, the umbrella company building the processors and AI fabric behind modern data centers, with ventures like Sierra RISC-V CPUs and Loom AI Fabric, plus Ada for deploying AI at scale. Alright, we've got a packed episode today, TSMC is raising the bar on both spending and revenue forecasts. Google is quietly plotting to outbuild Nvidia on AI accelerators by 2028, and we're seeing Nvidia's Blackwell GPS pop up in some interesting new places.
Let's get into it. Okay, story one, TSMC is raising its capital expenditure and revenue forecast, and the headline reason is exactly what you'd expect, surging demand for AI chips. Now, if you're not deep in the semiconductor world, you might be wondering why TSMC raising its own budget is such a big deal. Let me paint the picture.
TSMC is essentially the world's most important chip factory. They fabricate silicon for Nvidia, Apple, AMD, Qualcomm, Google, and basically everyone else who designs cutting edge chips. They don't design chips themselves, they manufacture them. And when TSMC says, hey, we're spending more money and we expect to make more money, that is a seismograph reading for the entire semiconductor industry.
So, what does raising CAPEX actually mean? Capital expenditure, CAPEX, is the money TSMC pumps into building and equipping its fabs.
We're talking about extreme ultraviolet lithography machines, the clean rooms, the wafer handling systems, all of it. These machines cost hundreds of millions of dollars each.
So when TSMC says it's raising CAPEX, it's essentially saying, we're betting big that demand is going to remain strong, and we need more manufacturing capacity to meet it.
That's not a decision you make lightly. You don't just casually order a few more EUV machines like their office supplies. The revenue forecast increase is equally telling. TSMC had already given the market pretty optimistic numbers earlier in the year, and now they're revising upward again. That tells you demand isn't just holding steady, it's accelerating. And the culprit pretty clearly is AI, specifically, the insatiable appetite for advanced chips to train and run large language models and other AI workloads. Every major hyperscaler, your Googles, your Microsofts, your Amazons, they're all racing to build out AI infrastructure, and every one of those chips runs through TSMC's fabs. What's really interesting here is the timing. There's been a lot of noise in the market about whether AI investment is sustainable, whether we're in a bubble, whether the hyperscalers are going to pull back on spending. TSMC's updated forecast is basically a counter-argument to all of that pessimism. They're looking at their order books, they're talking to their customers, and they're saying, no, demand is real, it's growing, and we need to spend more to keep up. That's a pretty strong signal. Think of it this way.
TSMC is like the world's most exclusive bakery, and every major tech company is placing bigger and bigger orders for custom cakes. When the bakery says it's buying more ovens and hiring more staff, and also expects to bring in more revenue this year than last year, you know those orders are real. This isn't speculative capacity. TSMC is responding to actual demand signals from actual customers writing actual checks. And for anyone watching the AI hardware space, that's a very encouraging data point about where this industry is headed. All right, story two. And this one genuinely made me sit up straight when I read it. According to analyst research from Fubun Research, Google could build more AI accelerators than Nvidia cells in 2028 Let me just let that land for a second, Google. Building more AI accelerators than Nvidia cells. That is a wild claim, and I want to dig into why it's actually not as crazy as it sounds. So first, some context. Google has been designing its own AI chips called TPS, Tensor Processing Units, since around 2015
They've gone through multiple generations, and each generation has gotten more powerful and more purpose-built for machine learning workloads. TPU-US are custom AI accelerators, meaning they're not general purpose chips like a GPU, they're specifically optimized for the kinds of math that AI models need, things like matrix multiplications and tensor operations. Google uses them internally to train and serve its own models, and also offers them to external customers through Google Cloud. Now, the Fubon Research Analyst's claim is that by 2028, Google's TPU production volume could exceed the number of AI GPU us and Nvidia sells in the same year. And here's the thing, this is actually plausible when you think about the scale of Google's AI ambitions. Google has one of the largest AI research organizations in the world. It's deploying AI across Search, YouTube, Google Workspace, Google Cloud and its own model development with Gemini. The volume of inference that's just running AI models, not training them that Google needs to support its global user base is staggering.