Every conversation I have with readers about chip stocks eventually turns into a conversation about one company, and I understand why. But Nvidia’s dominance is also exactly why I think the more interesting risk-reward in semiconductors right now sits one layer down, in the companies building the custom silicon and networking gear that hyperscalers are increasingly buying alongside, not instead of, Nvidia’s GPUs.
I wrote recently about buying tech stock cautiously, favoring the infrastructure layer where revenue is already showing up over the application layer still running on promises. Semiconductors are the clearest example of that infrastructure layer, and Nvidia is not the only name in it worth owning.
Why the Custom Silicon Story Has Gotten Too Big to Ignore
Broadcom’s most recent quarter is the number that changed how I think about this trade. AI semiconductor revenue came in at $10.8 billion, up 143% year over year, with bookings for chips not yet shipped topping $30 billion. Management guided to roughly $16 billion in AI semiconductor revenue for the following quarter, more than 200% growth, and reiterated a full fiscal year target near $56 billion, up about 180% over the prior year. Those aren’t rounding errors next to Nvidia’s scale. They’re evidence that hyperscalers have decided custom, purpose-built silicon deserves a meaningful slice of their compute budget alongside general-purpose GPUs, not as a replacement for them.
The customer list behind that number is the part I find most convincing. Reporting around the quarter ties Broadcom’s custom-chip business to Google’s TPU program, Meta’s internal accelerator effort, and additional design wins tied to ByteDance, OpenAI, and Anthropic. When the same handful of hyperscalers spending hundreds of billions on Nvidia GPUs are simultaneously paying a chip designer to build them an alternative, that’s not a company hedging against Nvidia so much as a market deciding it needs both approaches running in parallel.
Reading the Stock Reactions Correctly Matters More Than the Headlines
Here’s where I’d point readers back to something I’ve written about before on the mechanics of reading post-earnings direction, because chip stocks this year have been a textbook case of headline numbers and stock reactions pulling in opposite directions. Broadcom’s own quarter tells both sides of that story within a few months of each other. One quarter’s soft near-term AI guidance triggered a hard selloff despite otherwise strong results. A subsequent quarter, with AI chip guidance beating expectations and a new customer disclosed, sent the stock up more than 9% to a record high in a single session, adding roughly $135 billion in market value. Same company, same underlying growth trajectory, wildly different market reaction depending on whether guidance cleared or missed the bar the market had already set.
That pattern is exactly why I don’t chase these stocks off a single print. I wait for the reaction, then check whether the move was about the trailing quarter or the forward guide, because with this group it is almost always the guide that decides the next several weeks of price action.
| Company | Recent AI-Related Growth | Forward P/E | My Read |
|---|---|---|---|
| Broadcom (AVGO) | AI semiconductor revenue +143% YoY, $30B+ bookings backlog | ~32x | Best risk-adjusted way to own the custom-silicon trend |
| AMD | Data center revenue +57% YoY | ~64x | Real progress, but valuation leaves little room for error |
| Marvell (MRVL) | Revenue +34.1%, custom ASIC wins at Amazon and Microsoft | ~50x | Smaller, higher-beta way to play the same trend as Broadcom |
| Arm Holdings | Revenue +22.8%, 97.5% gross margin | ~131x | Extraordinary business, priced for perfection I’m not willing to pay for yet |
Why Broadcom Specifically Earns the Larger Position
Valuation is the simplest way I can explain why Broadcom sits at the top of this list rather than AMD, despite AMD’s own real progress. AMD trades at a forward price to earnings ratio north of 63, compared with Broadcom closer to 32, even though Broadcom’s AI semiconductor growth rate has been running faster in percentage terms over the last two quarters. When two companies both show strong execution but one is priced at roughly double the other’s multiple, the more expensive one needs a cleaner, longer runway of flawless quarters to justify holding it at the same conviction level, and AMD is still fighting Nvidia directly for GPU market share rather than occupying the comparatively less contested custom-silicon lane Broadcom has built around.
Nvidia’s own most recent quarter, which I broke down in detail when the numbers came out, actually reinforces this rather than undercutting it. Nvidia is not standing still. Its upcoming platform architecture is designed to expand the revenue opportunity per gigawatt of deployed compute well beyond what the current generation captures, and that is a real competitive answer to Broadcom, Marvell, and everyone else in this space. I don’t own Broadcom because I think it beats Nvidia. I own it because the two are increasingly serving different parts of the same expanding budget, and Broadcom gives me exposure to that expansion at a valuation that still has room to work if execution holds.
The Names I’m Watching From the Sidelines For Now
Marvell is the name I keep almost buying and keep deciding to wait on. The growth rate and customer list, Amazon and Microsoft designing custom AI silicon with Marvell’s help, are genuinely comparable to Broadcom’s story at roughly a tenth of the market cap, which is exactly the kind of asymmetry I like. What holds me back is size and concentration. A smaller company with fewer diversified revenue lines is more exposed if any single hyperscaler customer decides to pull work in-house or shift it to a competitor, and I’d rather watch another two quarters of execution before sizing a position the way I have with Broadcom.
Arm is the hardest one for me to walk away from and the hardest one for me to buy. A 97.5% gross margin business collecting royalties on nearly every chip architecture in the world is about as close to a toll road as public markets offer. But at well over 100 times forward earnings, the stock needs growth to reaccelerate meaningfully from its current pace just to grow into the multiple it already trades at, and I’m not comfortable paying that price for a story that still needs to prove the reacceleration rather than having already shown it.
A Lower-Volatility Way to Stay Involved
For readers who like this sector’s growth story but don’t want single-stock semiconductor volatility in a portfolio, I’ve also written about generating income through selling cash-secured puts on a broad index rather than picking individual names, which is worth revisiting if concentrated chip exposure feels like more risk than you want to carry into an earnings-heavy stretch. It’s a different tool for a different risk appetite, and I think readers should have both approaches available depending on how much single-name risk they’re willing to underwrite.
I also cross-reference analyst estimate revisions on Seeking Alpha before adding to any of these positions, since a rising or falling revision trend in the weeks after a print tends to tell me more about where consensus is actually heading than the headline reaction on the day itself.
For readers who want to track valuation, momentum, and estimate revision factors across this whole group in one place rather than pulling data from five different sources, Quant Rating Tool to fold exactly this kind of comparison into a single score you can check before each earnings cycle.
Where I Land
Broadcom is my highest-conviction position in this group, sized accordingly. Marvell is next in line pending another quarter or two of proof. AMD I own smaller than the growth headline alone would suggest, purely on valuation discipline. Arm I’m content to watch from outside for now. None of that changes my broader view that this part of the market, the companies building what hyperscalers need to deploy AI compute at scale, remains a better place to be positioned than the application layer still trying to prove it can convert usage into durable revenue.
Gavin Thorne writes on technology sector positioning and macro-driven equity strategy. This article reflects his personal research process and is intended for informational purposes only. It does not constitute investment advice.