Cisco in 2000 vs Nvidia today: the closest AI bubble comparison

Why is Cisco in 2000 the reference for Nvidia in the AI bubble?

Both companies sit one step behind the headline. Cisco sold network equipment to the companies that built the internet, and Nvidia sells computing hardware to the companies that build AI. In both cases the supplier earns when customers spend, so its sales follow the spending cycle of others. That shared position makes the pair a natural comparison.

The record gives one verified number for Cisco: its stock lost 80% of its value in the crash (Wikipedia, Dot-com bubble). Nothing on this page states an Nvidia price, because share prices change daily and this page only lists criteria. You can read the current figures in Nvidia’s own filings and apply the six criteria yourself.

Which six criteria compare Cisco in 2000 with Nvidia today?

Six criteria carry most of the comparison. Who the customers are. How they fund their orders. How fast orders can change. How concentrated sales are in a few buyers. How much competition is arriving. And how the valuation compares with earnings. Each criterion has an answer in public reports, so the comparison rests on documents instead of impressions.

Score each criterion for each company as supportive, mixed, or fragile. A supplier with 4 or more fragile answers depends on its customers’ funding. A supplier with 4 or more supportive answers has more room to absorb a correction. The scores carry no price target, and they will move as new reports arrive.

How did Cisco’s customers pay for orders, and how do Nvidia’s?

The March 2000 Barron’s warning said that internet companies were running out of cash fast. Many of Cisco’s customers were such companies, so their orders depended on new funding. When funding closed, orders followed. The lesson for any supplier is to trace the origin of its customers’ cash.

For Nvidia today, the same trace begins with its largest buyers. Ask which customers fund purchases from their own earnings, which fund them from debt, and which depend on investors. The EQOS Asia AI study records about $200B a year invested in AI (EQOS Asia AI study, April 2026), and the split of that sum between earnings and outside money is the number that decides the risk.

How fast can orders for Cisco-style hardware change in a correction?

Hardware orders can change faster than end demand. When a customer holds spare equipment, it can pause new orders for several quarters while its own users keep growing. A supplier then sees sales fall even though the technology is still spreading. This gap between use and orders is what hit equipment sellers in 2000.

To test the gap today, compare the growth of AI usage with the growth of AI hardware orders. If orders grow much faster than usage, stock is building somewhere in the chain. If both grow together, the build-out follows demand. Quarterly reports and customer disclosures let you track the comparison every 3 months.

Does customer concentration raise the risk for Nvidia as it did for Cisco?

A supplier with a few very large buyers sees its sales change sharply when one buyer changes plans. A supplier with thousands of buyers sees a smoother line. Companies report the share of revenue from their largest customers in their annual filings, so the figure is easy to find and to date.

Concentration works in both directions. Large buyers give stable volume while they expand, and they give sudden drops when they pause. Score the criterion as fragile if a handful of buyers account for most revenue and those buyers rely on outside funding. Score it as supportive if buyers are many and fund themselves from earnings.

What did the Cisco stock fall show about valuation and a boom?

Cisco’s stock lost 80% even though the internet kept growing, which shows that the price of a supplier can fall far while its end market expands. The fall came from the distance between the price and the earnings that could support it. The criterion to apply to any supplier is the same: how many years of current earnings does the price represent?

The study records AI valuations up ×4 in 18 months and scores the enterprise AI market at φ(S) = 0.337, in the fragile zone. Valuation alone settles nothing, since earnings can grow into a price. The practical approach is to track the ratio each quarter and watch whether earnings close the gap or whether the gap keeps widening.

How can you use the Cisco and Nvidia comparison in a decision?

Write the six criteria on one page, score Cisco in 2000 from the record and your supplier today from its filings, and compare the columns. Update the page every quarter. The EQOS Asia method page explains how such scoring becomes a recomputable measurement. This page is general information, not investment advice.

The Λ simulator returns a φ(S) estimate in 10 minutes, and a 30-minute pre-diagnostic applies the scoring to a decision in Singapore, ASEAN, or Hong Kong. You bring the decision, and the session returns a score and a ranked set of options.

Frequently asked questions

About the author

Stéphane Jambu

President, EQOS Forecast Asia

Stéphane Jambu leads EQOS Forecast Asia and works as a consultant in e-commerce, search visibility, and AI. LinkedIn profile