The same slots boil twice
Dotcom is usually remembered as a bubble inflated by the narrative of a new technology, the internet. But the core of its drive was not narrative but supply-chain overload. Dot-coms and carriers bought gear anticipating a traffic explosion, and the money ran down the chain into equipment, optical components, chips and fiber. And the slots where that overload boiled are boiling again, almost unchanged, on today's AI chain. The structure of circular financing, in which a supplier invests in demand to manufacture its own revenue, stood in the same place twenty-five years ago.
Dotcom was supply-chain overload too
The dotcom chain was internet and telecom infrastructure. Dot-coms and carriers bought gear anticipating a traffic explosion, and the money flowed to network equipment (Cisco, Nortel, Lucent), optical components and fiber (Corning, JDS Uniphase, Ciena), chips (Broadcom, Qualcomm) and fiber laying. Wherever a pick was sold, benefit poured in. Cisco was briefly the world's most valuable company in 2000, and Qualcomm rose about 2,600% in 1999 alone.
When demand stalled, the chain unwound in the very order it had risen. The slots where the benefit had been gaudiest were carved out deepest: Cisco about -88%, and Nortel, JDSU and Corning into the high 90s. Propagation on the way up flipped into staged liquidation on the way down. At the level of mechanism, the dotcom and AI overloads share one skeleton under different technologies: the order in which the picks sold is the order in which they unwind.
The same slots boil now
The coordinates of the slots that boiled in dotcom land on today's AI chain unchanged. Corning, in fiber, appears in both bubbles: the company that fell into the 90s in the dotcom crash has risen again on fiber contracts with Meta, NVIDIA and Amazon, +60.6% in 2024, +87.8% in 2025, +161.3% in 2026. Where JDS Uniphase once stood in optical components now stand Coherent (+117.6% in 2024, +94.8% in 2025, +120.6% in 2026) and Lumentum (+339.1% in 2025, +133.9% in 2026). Ciena, the optical transport that links datacenters, ran +175.8% in 2025 and +107.2% in 2026. A single slot, optical components, has run hot twice, twenty-five years apart.
Vendor financing came back as circular financing
One of the weak links that broke dotcom was vendor financing. Lucent and Nortel lent carriers the money to buy their own gear, and so the supplier manufactured its own demand through credit; when a customer collapsed, revenue and receivable evaporated together. That structure has come back under a new name. NVIDIA invests $2B in Lumentum, $2B in CoreWeave, and takes a stake in Nebius too: CoreWeave is OpenAI's second-largest customer, and Nebius signed a $17-19B contract with Microsoft. This loop, the seller supplying capital to the buyer, is circular financing, and it is structurally the same as vendor financing.
Dark fiber came back as power and substrate overbuild
The other weak link was dark fiber. Only about 5% of laid fiber was ever lit; the rest stranded in the dark, because a demand forecast had hardened into physical plant. Today's counterpart is the physical overbuild that flips first when demand pauses a beat: power, substrate and memory capacity. As with fiber, once these are laid they cannot be reversed; they can only wait for demand.
What differs is the cash at the demand end
If the same slots boil the same way, is the ending the same too? One thing differs. Dotcom's demand was demand borrowed into being through vendor financing, and above it sat a layer of pure froth (Pets.com) that rose on narrative alone with no revenue or profit. Today's demand sits with cash-generating giants: Microsoft capex of $64.6B (FY2025, +45%), Alphabet $91.4B (+74%), Amazon $131.8B (+59%) and Meta $69.7B (+87%), all out of their own cash flow. What differs is the cash at the demand end. The slots are the same, but the floor of demand holding up the chain is firmer than twenty-five years ago.
A firm floor, however, is not the same as no excess. The very fact that the same slots have boiled twice leaves a question: once is chance, twice is a pattern. If this repetition is not chance, there is a theory that produces it. Carlota Perez's technology cycle reads the grammar of that repetition, why the frenzy of an installation phase leaves behind, every time, the infrastructure of a deployment phase.