Latest / Stock Picking with Fexingo: Individual Equities, Research, and Building a Concentrated Portfolio / Why Broadcom Is Winning the Custom AI Chip Race
Transcript
- Lucas: Broadcom is up in premarket trading this morning after a string of analyst upgrades, and I think it's worth asking — why is Broadcom, of all the semiconductor names, getting this kind of attention right now? Luna: Because everyone's still obsessed with Nvidia, but Broadcom is quietly becoming the go-to for custom AI chips. That's a different — and maybe stickier — business. Lucas: Exactly. Nvidia sells the general-purpose GPU that everyone wants. Broadcom sells custom ASICs — application-specific integrated circuits — to hyperscalers like Alphabet and Meta. And those companies are increasingly designing their own silicon to optimize for their specific workloads. Luna: If today's conversation helps you think about the semiconductor landscape in a new way, that's exactly what we're here for. And the way these episodes stay ad-free is listener support. You can find us at buy me a coffee dot com slash fexingo. Lucas: Yeah, that's the model. No ads, just direct support from people who find value in the show. So if that's you, we appreciate it. Luna: Alright, back to Broadcom. So the big question is — how big is this custom chip opportunity? Lucas: Broadcom's CEO Hock Tan has been talking about a $60 billion addressable market just in AI networking and custom compute. That's not small. And they already have three major hyperscaler customers — Alphabet, Meta, and another one they haven't named publicly. Each of those is designing chips with Broadcom's ASIC team. Luna: And that's completely different from Nvidia's model, where you buy a chip and use it as is. With Broadcom, you're co-designing something proprietary. Lucas: Right. And once you go down that path, you're locked in. The engineering investment is huge — it takes years to develop a custom chip — so you're not going to switch suppliers easily. That gives Broadcom recurring revenue from high-value, multi-year contracts. Luna: So it's not just about the chip itself. It's the ecosystem around it. Broadcom also provides the networking silicon that connects these chips in data centers. Lucas: That's the part I think gets underappreciated. Broadcom's Tomahawk and Jericho switches are the backbone of many large-scale AI clusters. When Alphabet or Meta builds out a new data center for training large language models, they're buying Broadcom's networking gear alongside those custom ASICs. Luna: So they basically own both the compute and the connectivity. That's a powerful combination. Lucas: It is. And it's why Broadcom's semiconductor revenue has been growing at double-digit percentages even while the broader chip industry has had some lumpy quarters. In their most recent fiscal quarter, infrastructure software revenue — that's largely from the VMware acquisition — also beat expectations. Luna: VMware. That's a whole other story. They bought VMware for about $69 billion in 2023, and the Street was skeptical they could pull off the integration. But Hock Tan has a track record with acquisitions. Lucas: He's known for buying companies with strong margins, cutting costs, and cross-selling. They've moved VMware to a subscription model, which initially hit revenue but is now starting to stabilize. And they're bundling VMware's virtualization software with Broadcom's networking hardware. Luna: So the software story is part of the investment thesis too. But let's stay on the custom chip angle for a second. How does this compare to what Marvell is doing? They also have a custom ASIC business. Lucas: Marvell is a competitor, but they're smaller. Broadcom's scale and existing relationships with hyperscalers give them an edge. Plus, Broadcom has a much broader portfolio — they're not just ASICs, they're also RF chips for smartphones, storage controllers, and of course networking. Luna: So if I'm an investor looking at the semiconductor space, I have to choose between Nvidia, which is the high-growth, high-valuation story, and Broadcom, which is more diversified and maybe more defensible. Lucas: That's the trade-off. Nvidia's revenue is still growing faster — they're seeing triple-digit growth in data center — but Broadcom's model might be more sustainable. The custom chip business has lower margins than selling GPUs, but it's less cyclical because hyperscalers plan their infrastructure years in advance. Luna: And Broadcom pays a decent dividend. They've been increasing it steadily. So you get income while you wait for the AI story to play out. Lucas: Exactly. The dividend yield is around 1.4 percent, which isn't huge, but it's growing. And they've been aggressive with buybacks too. In 2025, they returned over $20 billion to shareholders. Luna: That's the kind of capital allocation that value investors like. But the stock is not cheap — it trades at around 30 times forward earnings. Lucas: It's not a deep value play, no. But compared to Nvidia at 40-plus times earnings, it's more reasonable. And if you believe the AI infrastructure buildout has years to run, Broadcom is a core holding. Luna: One risk I think about is concentration. If one of those hyperscaler customers decides to take chip design in-house — like Alphabet with its TPU — does that hurt Broadcom? Lucas: It's a real risk. But Alphabet already designs its own TPU, and they still work with Broadcom on networking and some ASIC components. The relationship is more complex than just 'we design your chip'. Broadcom provides IP, manufacturing coordination, and integration services that are hard to replicate. Luna: So the moat is more about the full solution than the chip design itself. Lucas: Exactly. And Broadcom is also investing in new areas like optical interconnects — which are critical for connecting chips across data centers at high speeds. That's a growing need as AI models get larger. Luna: There's also the VMware angle — if that integration goes well, it could provide a nice tailwind to earnings for the next few years. Lucas: Right now, the market is focused on the AI hardware story, but the software side could be a catalyst once the subscription transition is fully lapped. Analysts expect VMware to contribute meaningful free cash flow starting in 2027. Luna: So for someone building a concentrated portfolio of tech stocks, Broadcom feels like the kind of name you pair with a growthier pick like Nvidia or AMD. Lucas: I think that's a smart approach. Broadcom gives you exposure to AI infrastructure with less volatility and a dividend. And it's diversified across end markets — telecom, enterprise, data center, and now software. Luna: We should note that the stock is up about 25 percent year to date, so it's not like you're catching it at a low. But the thesis is still intact. Lucas: No, it's had a good run. But the premarket move today suggests analysts think there's more upside. And given the secular trends, I'd agree — as long as hyperscaler spending doesn't collapse, Broadcom is well positioned. Luna: Alright, let's leave it there. For Fexingo Business, I'm Luna. Lucas: And I'm Lucas. Thanks for listening.