Three Waves of a Technology Cycle

The sequence

Robert F. Smith's sequencing frame describes how a general purpose technology diffuses and where the returns eventually land, laid out in three waves. The first wave is hardware vendors, silicon and systems, where he places most of the capital that has moved so far and names Nvidia as holding a unique position within it, noting that his firm has also been investing in edge silicon companies in anticipation of a different tier of chips used at the network edge. The second wave is hyperscalers building out infrastructure and capacity, some of which he suggests may be overvalued in the near term, with Google's fully integrated technology stack singled out as a particularly strong position within that wave. The third wave is application providers, software that actually serves businesses, which is where he argues the cycle is heading: "The application providers usually get the lion share of the economic rent long term, once the technology has been diffused into those markets."

Two things distinguish this from a generic commentary on technology-stack layers. It is a claim about time, not only about which layer captures value, since Smith presents it as the historical shape these cycles have always taken rather than an AI-specific prediction; capital concentrates in the enabling layers first, because that is where the visible constraint sits and where the story is simplest to tell, and value shows up in the applying layers last, because diffusion into actual workflows takes years. The gap between those two facts is what living inside a cycle feels like from the middle of it. The first two waves are also recast as utilities rather than as permanent winners: they ultimately provide a utility that application providers use to serve businesses, meaning the earlier waves do not just capture a shrinking share of rent over time, they become inputs, with their pricing converging toward cost, which is exactly what lets the third wave's margins improve. The corollary Smith is effectively selling is that the third wave is 97 to 99 percent private, meaning most of the sequence's eventual payoff sits outside public markets entirely.

Why it matters

This is a sequencing argument for a conclusion reached elsewhere on different grounds. One route gets there through switching costs, on the logic that it takes only a line of code to switch vendors at the infrastructure layer, and Brendan Foody gets there from where differentiated capability is actually manufactured.¹ Smith's route is diffusion history. Three independent paths converging on the same destination is a stronger position than any single one of them alone, and the diffusion-history route is the one most likely to survive if the switching-cost argument turns out to be wrong.

The framework also reframes the bubble question as a timing question. If the first two waves are supposed to overbuild and then convert into utilities, infrastructure overvaluation is not evidence that the broader cycle is fake; it is the mechanism by which the third wave eventually gets cheap inputs. That is a materially different reading of the same facts than a standard capex-bubble worry, though the two readings do not fully rule each other out, since an overbuild can subsidize the eventual survivors and destroy the capital that funded it at the same time. The framework also supplies the missing timing element other value-capture arguments leave implicit: diffusion in enterprise software is measured in years, which is precisely why Smith thinks the relevant assets are buyable now, ahead of that diffusion completing.

Tensions and open questions

The historical claim itself is asserted rather than evidenced against any specific prior cycle. The obvious historical candidates cut in different directions: personal computers rewarded both the application layer and the underlying chipmaker for two decades running, the internet rewarded the application layer but mostly through new entrants rather than the incumbents being diffused into, mobile let platform owners take their rent through app stores rather than through application providers themselves, and cloud computing is itself a second-wave business that has kept extraordinary margins for fifteen years with no sign of the predicted utility-style compression. Cloud in particular stands as a live counterexample to the utility-conversion step in the argument.

The term application providers is also ambiguous between incumbents and new entrants. Most historical examples reward the application layer generally, but more specifically reward new companies within it, while the framework's own examples are drawn from incumbent portfolios, quietly assuming existing vendors capture the third wave, which is precisely the assumption a rival argument, that software incumbents have no automatic right to exist through this transition, directly disputes. The three waves likely overlap rather than arrive strictly in sequence, since hyperscalers already ship applications, model providers already ship consumer products, and application companies already buy their own silicon, which makes treating them as a clean queue something of a simplification whose main effect is to place the speaker's own asset class at the front of the next wave. Finally, nothing in the framework specifies that the third wave's rent gets distributed evenly; it is equally consistent with applications broadly capturing the largest share while a small handful of them capture nearly all of it.