Meta strikes AI chip deal with AMD days after committing to deploy millions of Nvidia GPUs - CNBC

**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 that's actually powering the AI revolution, you are in exactly the right place.
Huge thanks to our sponsors today. AIEDA, helping clients integrate AI into their workflows. Ago Consulting, that's Oco, providing silicon development consulting from AI all the way to So-O and Zen Semiconductor.
And AI and Silicon Venture Group, building the processors and AI fabric behind modern data centers, with some seriously cool ventures like their Sierra RISC-V CPUs and Loom AI fabric.
All right, we've got a packed show today. Meta is hedging its GPU bets with a surprise AMD deal. TSMC just posted some jaw-dropping profit numbers. And there's a landmark $400 million financing deal that's changing how we think about AI chips as financial assets. Let's get into it.
Okay, so first up, Meta and AMD. And I love this story, because on the surface, it sounds almost contradictory.
Like, Meta just committed to deploying millions and I mean millions of Nvidia GPS, and then literally days later, they turn around and sign an AI chip deal with AMD.
So what's going on here? Is this a betrayal?
A backup plan?
Actually I'd argue it's neither. It's just smart supply chain strategy. And it tells us a lot about where the big hyperscalers are headed. So let's set the scene. Meta has been on a massive infrastructure spending spree.
We're talking about one of the largest AI compute build-outs in history.
They've committed to Nvidia's GPS H100s, and likely the next-gen Blackwell chips as the backbone of their AI training workloads. That makes sense. Nvidia's hardware is still the gold standard for large-scale model training.
CUDA is deeply embedded, the software ecosystem is mature, and if you're training billion parameter models, you want the best. But here's the thing, when you're operating at meta-scale, you cannot afford to be a single vendor shop. That's just risk management 101 Think about it like building a city's power grid. Sure, you might have one primary power source, but you'd be crazy not to have backup generators and alternative feeds. Because if something goes wrong with your primary supplier, whether that's supply constraints, pricing pressure, geopolitical issues, or just allocation problems, you need options. And Nvidia's chips are notoriously hard to get right now. The demand is absolutely insane, and even the biggest companies in the world have had to wait in line. So Meta cozying up to AMD makes a lot of sense strategically. AMD has been working hard on their Instinct series of GPUs. The MI300X in particular has gotten some real traction. Their ROCM software stack has improved significantly, and more importantly, for certain workloads, particularly inference, AMD's hardware is genuinely competitive. Now the details of this specific deal aren't fully public, so we don't know exactly what Meta is buying or at what scale. But the signal is loud and clear, Meta wants AMD in their ecosystem. And for AMD, this is huge. Landing Meta as a customer is a massive validation. It's like if you're a new restaurant and suddenly a Michelin star food critic gives you a thumbs up. Other customers start paying attention. It changes your credibility overnight. The broader implication here is that we're entering an era where the hyperscalers Meta, Google, Microsoft, Amazon, are all actively working to reduce their dependence on any single chip vendor. They're building their own custom silicon, they're diversifying across Nvidia and AMD, and they're pushing open ecosystems so they can switch more easily. That's a fundamental shift in the AI hardware landscape. And this Meta, AMD deal is a perfect example of it playing out in real time. All right, story too. And honestly, this one's kind of wild. TSMC just reported Q2 earnings and their profits surged 77% year over year, 77%.
Let that sink in for a second. That is not a typo. And the reason AI chip demand, pure and simple. So for those who might be newer to the show, let me give you a quick TSMC primer, because this company is absolutely central to everything we talk about here. TSMC, Taiwan Semiconductor Manufacturing Company, is the world's largest and most advanced contract chip manufacturer. They don't design chips. They make them for basically everyone. Apple, Nvidia, AMD, Qualcomm, you name it. If a company designs a cutting edge chip, there's a very good chance TSMC is the one actually fabricating it. Think of TSMC like the world's most sophisticated bakery. Companies like Nvidia come in with their recipe, their chip design and TSMC has the ovens, the ingredients, the precision equipment to actually bake that chip at scale. And right now, everyone wants TSMC's most advanced ovens, their 3 nanometer and 5 nanometer process nodes, because that's where the AI chips live. A 77% profit surge is extraordinary by any measure. For context, that's not revenue growth, that's profit. The bottom line, it means TSMC is not just selling more chips, they're selling them at premium margins. And that makes sense because AI chips are the most complex, most expensive chips they make. The yields are challenging, the process nodes are bleeding edge, and customers are willing to pay a premium because the demand for AI compute is so intense. TSMC has specifically called out AI as the primary driver of this growth. Their advanced packaging technology, COWOs, which stands for chip-on-wafer-on-substrate, is in particularly high demand because it's used to package the high bandwidth memory together with the GPU dies and chips like Nvidia's H100 and H200. And they are absolutely capacity constrained on that packaging right now. The record profit numbers also tell us something important about the staying power of AI hardware demand. There's been a lot of debate, is this a bubble? Is demand going to soften? And TSMC's numbers are like a temperature reading for the whole industry. When the foundry that makes the chips is posting record profits, it means orders are strong, capacity is booked out, and customers are paying full price. That's a healthy demand signal. Now, there is a flip side here, and this connects to the third TSMC story I want to touch on. Even with these incredible earnings, TSMC stock has been facing some new pressure. And this is where geopolitics comes roaring back into the conversation.