The AI trade is no longer about chips alone — the next winners are the toll collectors
The AI debate is shifting from bubble-versus-not to a simpler question: who gets paid no matter which model, chip, or platform wins. In this tape, that favors the infrastructure and enablement layer — networking, cloud capacity, power, cooling, and buildout equipment — over treating pure compute leadership as the only trade that matters.
The AI trade is maturing, and the market is starting to act like it. With technology leadership wobbling and investors openly questioning whether AI spending can keep carrying mega-cap valuations, the better framing now is not whether AI is real, but where the most durable economics sit inside the capex cycle. Our view is that the next leg of the theme belongs less to "chips at any price" and more to the companies collecting tolls across the buildout: the networking vendors, cloud landlords, electrical suppliers, cooling specialists, and equipment makers that get paid as long as AI infrastructure keeps expanding. That matters because the physical AI stack is no longer optional, and unlike model leadership, it does not need a single winner to monetize.
Start with what changed in the tape. The first phase of the AI trade rewarded the cleanest compute exposure, and NVDA remains the benchmark for that logic. But this week’s debate around second-half spending, earnings delivery, and return on AI investment is a sign that investors are moving from admiration to underwriting. Once the question becomes who captures recurring economics from the buildout itself, the field broadens quickly beyond accelerators.
That is why the most useful comparison now is not chip leader versus chip laggard. It is pure compute beta versus the businesses that monetize the entire stack. Broadcom is the clearest example of the shift already happening in plain sight: it said Q1 AI revenue reached $8.4 billion, up 106% year over year, and guided Q2 revenue to roughly $22.0 billion, up 47%. That is not a speculative adjacency story. It is evidence that networking and connectivity are already becoming core AI revenue pools, not sidecars to the GPU trade.
The same logic applies to Oracle, where the market has been slower to give full credit. Oracle trades at a far lower 19.16x P/E and 6.26x sales despite revenue growth of 17.4%, and management is leaning harder into AI infrastructure with capex set above prior expectations and more financing to support that expansion. That is exactly what a toll-collector profile looks like: less dependent on winning the model race, more exposed to leasing compute, storage, and data-center capacity to whoever needs it. In a market asking for clearer monetization, that setup is easier to defend than paying up simply because a company sits closest to the training workload.
The physical bottlenecks make the case even stronger. AI is no longer just a semiconductor story; it is a power-density and thermal-management story. Vertiv sits right in that choke point, with 27.7% revenue growth and 164.4% EPS growth, because cooling and power systems become more valuable as racks get hotter and denser. Eaton and Quanta Services are less glamorous, but that is the point. If data-center demand keeps rising, somebody has to supply the electrical gear and somebody has to build the transmission and facility backbone. Those economics do not care which frontier model has the best benchmark this quarter.
A few comparisons show how the market is repricing that reality:
NVDA: 33.59x P/E, 18.77x sales, 63.0% net margin, +3.9% YTD
ORCL: 19.16x P/E, 6.26x sales, 25.4% net margin, -24.8% YTD
Yes, the obvious counter is that the chip leaders still own the best economics. NVDA is still growing revenue 65.5% with a remarkable 63.0% net margin, and bulls are right to say that remains an elite combination. But that argument increasingly misses the market’s new problem: not whether AI demand exists, but whether the most crowded leadership names can keep absorbing ever-higher expectations while the spending bill spreads outward across the stack. When the capex cycle broadens, the winner-take-most logic at the chip layer gives way to a winner-take-a-cut logic across infrastructure.
The second counter is also fair: some of these toll collectors are no bargain. VRT at 67.58x earnings and AMAT at 68.92x are not hidden value stories, and PWR has already had a huge run. But that does not invalidate the thesis; it sharpens it. This is stronger as a relative argument than a blanket cheapness call. We are not saying every infrastructure name is mispriced. We are saying the market is right to pay for businesses tied to the whole AI buildout, and wrong to keep framing the opportunity as if only the compute leaders matter.
That is also why Applied Materials belongs in the conversation even with a richer multiple. Its 152.7% YTD move looks extreme next to just 4.4% revenue growth and 0.3% EPS growth in the current snapshot, but the market is looking through that to the next bottleneck: advanced packaging and wafer-fab intensity needed to keep AI hardware scaling. In other words, even at the equipment layer, investors are paying for constraint ownership. The common thread across AVGO, ORCL, VRT, ETN, PWR, and AMAT is not that they are all cheap. It is that they all monetize the buildout whether one model family, one cloud platform, or one chip architecture wins outright.
The better AI trade now is not a vote against chips. It is a vote for the parts of the stack that keep getting paid as AI spending moves from enthusiasm to infrastructure. That is a more durable place to be when the market starts asking harder questions about ROI, earnings conversion, and how much of today’s demand is tied to one vendor versus the entire system being built around it.
What we are watching from here is simple: whether spending keeps broadening from accelerators into networking, cloud capacity, power, cooling, and construction. If that broadening stalls, or if enterprise AI demand proves too narrow to justify the physical buildout, the toll-collector thesis weakens fast. But if second-half results keep showing that the bottlenecks are downstream and the revenue capture is spreading, the market’s next winners are unlikely to be the same names that defined the first sprint.
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.
