The repetition has a theory
The frenzy around AI is not the first of its kind. Railways, electricity, the automobile and the internet each ran through a burst of investment, a bubble and a crash. The person who named that repetition is Carlota Perez.
In her 2002 book, Perez set out that the technological revolutions of the past 250 years all traced a similar life cycle: the Industrial Revolution (1771), Steam and Railways (1829), Steel and Electricity (1875), Oil and the Automobile with mass production (1908), and Information and Telecommunications (1971). Each of the five surges unfolded in the same shape over roughly half a century. This is no string of coincidences. Perez's claim is that the character of capital changes from phase to phase, and that shift is what produces the recurring form. That dotcom was one instance of the same pattern is taken up separately in the dotcom supply chain as a mirror for the AI bubble.
The installation bubble is a function
Each surge splits in two. The first period is installation: a new technology irrupts, financial capital pours in and overbuilds infrastructure, and a frenzy forms. That frenzy collapses at the turning point. The second is deployment: production capital leads as the infrastructure spreads through society, opening a golden age of productivity.
Perez's central insight is that the installation frenzy is not an aberration but functional. Laying down canals, railways and fiber far faster than society yet needs them is not something cold calculation would do; what underwrites the excess is financial mania. The crash washes out the speculation but leaves the physical infrastructure behind. The frenzy of installation leaves the infrastructure of deployment. The bubble is closer to a toll paid for arriving early.
Dark fiber earned its keep only after the crash
The fiber laid and left idle in the dotcom years, the so-called dark fiber, did not disappear. In the late 2000s YouTube, Netflix and the cloud switched it on and used it. The bubble deflated, but the infrastructure became the fuel of deployment. That capacity was excess in installation and legacy in deployment.
The wiki's measurements point to installation
Where does AI stand in this cycle? It could be the deployment climax of the ICT revolution, or mid-installation of a sixth paradigm. Either way, the measurements of the companies this wiki tracks (its nodes) lean toward installation.
First, the epicenter posts the largest returns. NVIDIA rose +239% in 2023 and +171% in 2024. Yet the numbers grew larger further down the chain. Kioxia rose +1,082% in 2025, and Nanya Technology turned from −56% in 2024 to +985% the next year. Ibiden, holding the AI packaging bottleneck, rose +261% in 2025 and +192% in 2026. The order matters. As the epicenter slowed (NVIDIA +38.9% in 2025 and +5.1% in 2026), the layers below peaked that same year. The weight of the gains moved from top to bottom. That sequence of triple-digit returns running from the top of the chain to the bottom (whether the AI cycle propagated along the value chain) is the signature of financial capital running ahead of production.
Second, the capital forms a circuit. NVIDIA investing at once in its customers and suppliers, the circular financing, is in Perez's frame a classic sign of an installation phase where financial capital has outrun production capital. It is a long way from the calm diffusion of deployment.
The map separates froth from infrastructure in advance
The implication that matters for investors is separation. The turning point divides the froth of speculation from the substance of infrastructure. The compute, power, substrate and datacenters being overbuilt now are, even if a crash comes, the physical legacy that will remain like dark fiber to power the deployment. On valuation alone the two rise together and look equally expensive. Yet one, once it reverses, disappears without a trace, while the other stays on the ground after the reversal.
This is where a value-chain map earns its keep. The map's question is which phase we are in now, and the material for reading it is the wiki's measurement of how far down the layers the re-rating has descended. If the re-rating has run from the epicenter to the end of the chain, the installation phase is likely near its peak. Which nodes belong to the financial capital that will vanish with the bubble, and which are the infrastructure that will remain after a crash: the chain lets you tell them apart.
Verdict
Five surges took the same shape, and AI's measurements too lean toward installation. The triple-digit returns running from top to bottom of the chain and the circular financing are the grounds. Perez's proposition that the frenzy of installation leaves the infrastructure of deployment fits the present phase, at least. The map's worth is in drawing, before any crash arrives, the line between the froth that will vanish and the infrastructure that will remain.
That is the cycle, the axis of time. But there is more than one axis. The same chain splits along space as well as time, and sanctions and borders replicate the same value chain in two copies. In China's parallel ecosystem we open that axis of space.