The next AI trade is edge compute, not another round of data-center winners
The market is still pricing AI like a single data-center semiconductor trade, even as the next monetization battleground shifts toward local inference on PCs, phones, and edge devices. Nvidia’s June 1 PC AI chip launch matters less as another win for **NVDA** than as proof that AI is moving closer to the user — and that broadens the winner set.
The market keeps trying to force AI into one familiar template: buy the biggest data-center winner and wait for the next hyperscale capex cycle. We think that framing is getting stale. Nvidia’s June 1 launch of a PC-focused AI chip is the clearest sign yet that the industry’s next fight is not just about training models in giant clusters, but about where inference actually lives. Once AI becomes an endpoint feature instead of a cloud-only service, the upside starts to spread beyond the obvious server champions and toward the companies that own device silicon, CPU relevance, and architecture at the edge.
That is why this week matters. Nvidia did not just announce another accelerator for the same buyers; it pushed AI capabilities directly into laptops and desktops and explicitly stepped into a field that includes AMD, INTC, QCOM, ARM, and AAPL. The significance is strategic, not cosmetic. Training built the first AI bull market, but inference is the workload that has to meet real-world constraints like latency, privacy, battery life, and cost. Those constraints favor compute that sits closer to the user. If AI agents are going to be persistent, personal, and always on, they will not all run from a distant data center.
The valuation setup already hints that the market has not fully digested that shift. NVDA still trades at 33.95x earnings with a $5.36 trillion market cap, which is not unreasonable given its 65.5% revenue growth and 63.0% net margin, but it also shows how mature the consensus AI trade has become. By contrast, QCOM changes hands at 25.34x earnings and 5.57x sales, a much lower multiple for a company sitting directly in the path of AI PCs and on-device processing. AAPL at 37.18x earnings is not cheap either, but its role in local inference is still being treated more as ecosystem optionality than as a core AI multiple driver. The point is not that Nvidia is over; it is that the next leg of upside may come from names the market still treats as secondary characters in an AI story they are increasingly central to.
The cleanest evidence is in the product cycle itself. Nvidia’s latest quarter was enormous, with $81.6 billion in revenue and $75.2 billion from data center alone, which is exactly why the old trade still has believers. Yes, bulls can fairly say the proven money remains in centralized infrastructure, and the recent enthusiasm around AI servers shows that capex is not rolling over. But that comparison misses the direction of travel. When even Nvidia is spending launch-day energy telling investors that AI should sit inside the PC, the message is that the company sees the next adoption wave broadening beyond hyperscalers. The market should listen to that signal, not just celebrate it as one more reason to own the incumbent winner.
That broadening also changes which semiconductor functions matter most. Edge inference is not only a GPU story; it is a CPU, modem, NPU, power-efficiency, and architecture story. That is why AMD and INTC are more relevant to the next phase than the market narrative often admits. AMD has already been pushing Ryzen AI processors and local inference chips, while Intel has argued that AI inference demand is reviving CPU importance. The market action reflects that possibility, even if fundamentals are uneven: AMD is up 122.4% year to date, while INTC is up 175.7%, despite Intel still posting a negative net margin of -5.9%. We would not confuse that with clean execution, but we would recognize what the tape is saying: investors are starting to price AI as something that can restore value to architectures outside the classic data-center GPU stack.
The architecture layer may be even more interesting. ARM trades at a rich 474.12x earnings and 86.34x sales, which is extreme by any standard, but the reason investors tolerate that kind of valuation is obvious: if AI inference spreads across billions of devices, instruction-set ownership and low-power design become more valuable, not less. That does not make ARM cheap; it makes the market’s instinct directionally understandable. The same logic supports QCOM, which looks far less demanding on valuation and is directly exposed to the idea that AI becomes a feature embedded in premium devices rather than a service rented from the cloud. If the next AI cycle is about attach rates in PCs and phones, not just another round of server rack expansion, those are the kinds of exposures that can rerate.
There is a useful historical analogy here. When computing moved from centralized systems toward the PC, value did not disappear from infrastructure, but it did spread to endpoint hardware, operating environments, and the architectures that made local computing practical. AI looks ready for a similar handoff. The first phase rewarded whoever sold the picks and shovels to giant model builders. The next phase should reward whoever makes AI usable, fast, private, and cheap enough to live on the device in front of the user. That is a different question from who wins the next hyperscale budget cycle.
Our view is simple: the market is late to the idea that AI is becoming an endpoint deployment story. Nvidia’s PC push does not weaken NVDA; it strengthens the case that the opportunity set is widening to QCOM, AMD, INTC, ARM, and AAPL in different ways. The next AI trade is not abandoning the data center. It is recognizing that inference economics, user experience, and product design all point toward the edge.
What we are watching now is whether this remains a launch-season narrative or turns into visible adoption across AI PCs, on-device agents, and local inference features. If endpoint AI stays mostly promotional while spending remains overwhelmingly concentrated in servers, the old leadership can keep dominating. But if device attach rates and architecture leverage start to matter even modestly, the market will have to stop treating AI like a one-ticker semiconductor story. On that shift, we think the re-rating opportunity sits closer to the edge than to another crowded chase into the same data-center winners.
Our take, not advice. This is opinion commentary — informational only, not personalized investment recommendations. Markets carry risk. Do your own research and consider your own situation before any trade.
▌The Daily Briefing · Free
A new stock idea, every evening.
One stock worth watching each weekday, plus the analysis behind it. Free, in your inbox.
