Nvidia vs Cerebras: which discounted AI chip stock offers better value right now - Crypto Briefing

**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 deep into the chips, silicon and systems powering the AI revolution. Big thanks to our sponsors, LimitLess AI, helping businesses integrate AI into their workflows, and to go consulting your go-to for silicon development expertise. Today, we've got a packed episode. We're talking chip stocks, TSMC breaking records, a wild V100 mod that'll blow your mind, H100s going to space, and SambaNova doing some very interesting things to older hardware. Let's get into it. Alright, kicking things off, we've got an interesting investor-focused question making the rounds. Nvidia vs Cerebras, which discounted AI chip stock, offers better value right now. Now I know some of you are pure hardware nerds, and the word stock makes your eyes glaze over. But stick with me because this is actually a really fascinating lens through which to understand where the market thinks the AI chip industry is heading. So let's set the stage. Nvidia, you all know Nvidia. They are the 800-pound gorilla of AI compute. H, Dens, B100s, the Blackwell architecture, they're basically the default answer when anyone says I need to train a large language model.
Their market cap has been astronomical, and even after some pullback from their peak, they're still one of the most valuable companies on the planet. Cerebras on the other hand is a very different animal. They went public relatively recently and represent what I'd call the radical rethink school of chip design. Their flagship product is the wafer scale engine, and when I say wafer scale, I mean literally the entire silicon wafer is the chip. Most chips are tiny little dies cut from a wafer. Cerebras said, nope, what if we just didn't cut it? The result is a chip the size of a dinner plate with over 4 trillion transistors and hundreds of thousands of AI cores. That's a fundamentally different architectural bet than what Nvidia is making. Nvidia's approach is essentially, make really powerful individual GPUs and connect a bunch of them together with high-speed interconnects like NVLink.
Cerebras says, what if you eliminate the need for that inner-chip communication entirely by making one massive unified compute surface? The trade-off is yield. When you're making a chip that uses an entire wafer, any defect anywhere is a much bigger problem. Cerebras has engineered around this with some clever redundancy tricks, but it's still a manufacturing challenge that's genuinely hard. Nvidia's approach, by comparison, is more manufacturable and more flexible. From a value perspective, the question analysts are wrestling with is, does Cerebras have a large enough total addressable market to justify its valuation, and can they scale fast enough to compete in a world where Nvidia keeps shipping new generations? Nvidia's moat isn't just the hardware, it's CUDA. It's the software ecosystem, it's years of developer tooling that makes switching genuinely painful. My take? Both have merit depending on your time horizon and risk tolerance.
Cerebras is a high conviction bet on a specific architectural approach.
Nvidia is more of a the AI wave keeps rising and they're the surfboard bet. For hardware enthusiasts, though, I'd say watch Cerebras closely. Their benchmarks on certain inference workloads are genuinely impressive, and architectural diversity in this space is healthy for everyone. What I find most exciting about this comparison is that it signals the market is maturing. We're not just asking will AI chips be important anymore, we're asking which approach to AI silicon actually wins. And that's a much more interesting question. Okay, moving on to story 2 And this one is all good news if you're a fan of AI hardware momentum. TSMC just posted a record quarter. And not just a little record, a full year growth outlook that's now been pushed past 40%.
40%? Annual growth for one of the largest semiconductor manufacturers on earth. That is a staggering number. For context, TSMC, the Taiwan Semiconductor Manufacturing Company, is essentially the world's most important chip factory. They don't design chips, they make them. Apple's A-Series chips, TSMC, NVIDIA's GPS, TSMC, AIM-NB's processors, TSMC, they are the foundry that the entire semiconductor industry depends on. And when TSMC has a record quarter, it's a direct reflection of what's happening in silicon demand globally, and what's driving it. AI?
Specifically, the insatiable appetite for advanced process node chips that power AI training and inference workloads. We're talking 3 nanometer and now 2 nanometer class chips, which are the most cutting edge processes TSMC offers. These are extraordinarily complex to manufacture, which is why almost nobody else can do it at scale. Samsung is trying, Intel is trying with their foundry ambitions, but TSMC has a lead that is very hard to close. Think of process nodes like this. The nanometer number roughly corresponds to how small the transistors are. Smaller transistors mean you can fit more of them in the same area, which means more compute power per watt. For AI workloads, which are essentially just billions of matrix multiplications happening over and over, more transistors per area equals better performance. So everyone wants the most advanced nodes TSMC can offer. The 40 plus percent growth outlook is also significant because it suggests the AI buildout isn't slowing down in any meaningful way.