Latest / Stock Picking with Fexingo: Individual Equities, Research, and Building a Concentrated Portfolio / Why Nvidia Is Becoming a Foundry Powerhouse
Transcript
- Lucas: Nvidia is up about three percent in the past five days, trading at two twenty-one. But the real story isn't the price move — it's what Nvidia announced last week that most people are glossing over. Luna: You're talking about the custom chip push. I saw the press release — they're basically offering to design custom chips for cloud companies now? Lucas: Exactly. They call it their 'foundry strategy' — though they're not actually fabricating the chips themselves. They'll design custom ASICs, application-specific integrated circuits, for clients like Amazon and Alphabet. And that's a huge strategic shift. Luna: Wait — so Nvidia, the company that sells the world's most powerful AI GPUs, is now going to help customers build chips that compete with their own products? That seems counterintuitive. Lucas: It does on the surface. But look at the economics. Amazon already designs its own Trainium and Inferentia chips for AWS. Alphabet has its TPUs. These companies are building custom silicon anyway. Nvidia's choice is either lose that revenue entirely or capture some of it by offering design services. Luna: So they're monetizing a trend they can't stop. Smart. But how does the margin profile compare? Nvidia's GPU business has gross margins north of seventy percent. Lucas: Custom ASIC design services are lower margin — think forty to fifty percent gross margin, roughly. But the volume can be enormous. A single cloud contract might involve millions of chips a year. Plus, Nvidia keeps a foot in the door for when those clients need GPUs for training. Luna: And there's the moat argument. If Nvidia designs the chip, they can ensure it's optimized for their CUDA software stack, right? That locks clients in deeper. Lucas: That's the play. Custom chips still run on Nvidia's software ecosystem. So even if the hardware isn't an Nvidia GPU, the developer workflow stays tied to CUDA. It's a defensive move disguised as a service offering. Luna: I'm trying to think who loses here. AMD and Intel have been trying to break into the AI custom chip market. Does this shut them out? Lucas: It makes their path harder. AMD has a custom chip division from the Xilinx acquisition, and Intel has Altera. But Nvidia has something they don't: the software ecosystem and the relationships with every major cloud provider. If Nvidia can offer a complete package — design, software, support — it's tough to compete on price alone. Luna: Let's talk numbers. How big could this be for Nvidia's revenue? Their data center segment did about one hundred billion last fiscal year. Lucas: Analyst estimates I've seen put custom chip revenue at three to five billion in the first year, growing to maybe fifteen billion by 2028. That's not nothing, but it's still small relative to their core GPU business. The real value is strategic — it protects their high-margin GPU sales by keeping hyperscalers from defecting entirely. Luna: So it's a hedge. If AI training demand slows — and there are signs that inference is becoming a bigger share of compute — Nvidia still has a revenue stream. Lucas: Exactly. Inference workloads don't necessarily need the most expensive GPUs. Custom chips optimized for inference can be cheaper and more power-efficient. Nvidia wants to capture that market before someone else does. Luna: I was reading that Amazon's latest Trainium 3 chip is supposed to compete directly with Nvidia's H200 in training performance. That's a direct threat. Lucas: It is. And Nvidia's response is: 'If you're going to build your own chips, at least let us help you design them so they run our software.' It's a brilliant strategic move. They're turning a competitor's product into a complement. Luna: What about the risk of cannibalization? If a cloud company uses a custom Nvidia-designed ASIC for inference, they buy fewer GPUs. Lucas: True, but Nvidia likely priced the service to offset that. Plus, the GPU business is still growing at double digits. Custom chips probably expand the total addressable market for AI compute rather than just stealing from existing sales. Smaller companies that couldn't afford a full GPU cluster might buy custom chips through a cloud provider. Luna: So this is about capturing the long tail of AI adoption. Big Tech already has custom chips. But mid-tier companies might now get access to optimized silicon through cloud providers using Nvidia-designed chips. Lucas: Bingo. And that's where the volume comes from. Think of it as Nvidia becoming the 'ARM of AI' — licensing designs rather than just selling finished products. They won't manufacture the chips themselves, but they'll take a cut on every one. Luna: Speaking of taking a cut, if you find these kinds of strategic breakdowns useful, it's worth knowing that our listeners who support the show on buy me a coffee dot com slash fexingo keep these episodes ad-free and independent. Lucas: Yeah, it's a small thing — a couple of dollars a month — but it genuinely makes a difference. We don't run sponsors, so listener support is what keeps us going. And we really appreciate it. Luna: Alright, back to Nvidia. So if this custom chip strategy works, what does it mean for the stock? Nvidia's PE ratio is still above fifty. Lucas: It adds a growth vector that isn't priced in yet. Most analysts model Nvidia as a GPU company. If custom chips become a ten to fifteen billion dollar business with decent margins, that's upside. Plus, it reduces the cyclical risk of GPU demand swings. I think it's a net positive for the thesis. Luna: But there's execution risk. Nvidia has never been a custom chip designer at scale. Their strength is in massive general-purpose GPUs. Designing custom ASICs requires a different engineering culture. Lucas: That's the bear case. And they're competing with companies like Broadcom and Marvell that have been doing custom chips for years. But Nvidia has the talent and the software edge. If anyone can pull it off, it's them. Luna: Let's wrap with a big picture thought. Is this the beginning of Nvidia transforming from a component supplier into a platform company? Lucas: I think so. They're moving up the value chain — from selling silicon to selling solutions. Custom chips, software, networking. It's a broader moat. And that's what you want if you're a long-term investor. The question is whether the market will give them credit for it in the near term. Luna: Good note to end on. Thanks, Lucas. Lucas: Thanks, Luna. Talk tomorrow.