Four ways it could pop, and what grows afterwards

Who is at risk, who could grow

Tap a scenario for the reasons and today's numbers. Red numbers mark net debt, spending beyond cash flow, losses, falling sales or rich valuations.

Scenario: The money stops

Trigger　Money stops flowing to companies building data centres on debt, and to private AI labs that have promised to buy far more compute than they earn. Once bond buyers or backers hesitate, the equipment orders that had been promised disappear.

In 2000The junk bond market shut and the carriers that had laid fibre on debt were cut off. Global Crossing (January 2002) and WorldCom (July 2002) went bankrupt, and Lucent, which had lent to its customers, booked $3.5bn of bad debts in 2001–02.

Related warning signs today

• Credit tighteningOff
• Spending beyond earningsOff

Cash-rich buyersPower, grid and buildings

Scenario: The spending stops

Trigger　The giants decide AI revenue does not justify the spending and cut capex. Orders for GPUs, memory and networking stop at once, leaving excess inventory and capacity.

In 2000When the carriers stopped spending in 2001, Cisco's sales fell by 30% in a single quarter. In April it announced a $2.25bn inventory write-down and 8,500 job cuts.

Related warning signs today

• Spending beyond earningsOff
• Chip stocks breakingOff
• The leader stallsOff

Users of cheap AICash-rich buyers

Scenario: Intelligence gets cheap

Trigger　Models and techniques that match today's performance on far less compute spread, and the assumption that there can never be enough GPUs breaks. On 27 January 2025, DeepSeek's model sent NVIDIA down 17% in a day, erasing $589bn of market value. That may have been the trailer.

In 2000In the late 1990s wavelength multiplexing (DWDM) multiplied what a single strand of fibre could carry by dozens of times. Installed capacity ballooned, bandwidth prices fell more than 90% in two or three years, and many of the companies selling it could not survive.

Related warning signs today

• The leader stallsOff
• Chip stocks breakingOff

Users of cheap AIAI on the device

Scenario: Rates cool it

Trigger　Rising power and component costs, or an overheating economy, bring inflation back and the Fed turns to hiking. Higher rates shrink today's value of future profits, and the most richly priced shares fall hardest.

In 2000The Fed raised rates by 1.75 points between June 1999 and May 2000, and the Nasdaq peaked in March 2000, midway through. After the fall it cut 11 times in 2001 alone, from 6.5% to 1.75%.

Related warning signs today

• Rate hikesOff
• Credit tighteningOff

Cash-rich buyersAI on the device

Area: Power, grid and buildings

Grows afterThe money stops

Assets that outlast a bust. Power plants, grid lines and substations last decades and take years to add. Even if overbuilt, latecomers put them to use.

In 2000The fibre ruined its builders, but it stayed and carried the video and cloud era a decade later. Data centre operator Equinix listed in August 2000, was caught in the crash and had to restructure its debt, yet survived to become one of the world's largest landlords.

Right after a bust, cancelled orders and lower power-demand forecasts could hit these shares hard. The growth comes later.

Area: Cash-rich buyers

Grows afterThe money stops, The spending stops, Rates cool it

Companies that earn from their core business and hold cash can buy the assets, people and technology that get cheap after a bust.

In 2000Verizon bought WorldCom's successor MCI for $8.5bn, Oracle bought Sun for $7.4bn, and a group including Apple and Microsoft bought Nortel's patents for $4.5bn.

Today's giants are also the biggest spenders. If they keep investing beyond their earnings, they end up holding the assets that lose value rather than buying them. Watch the capex ÷ cash flow column.

Area: Users of cheap AI

Grows afterThe spending stops, Intelligence gets cheap

The cheaper compute gets, the lower the costs and the wider the reach of companies building AI into their products.

In 2000Google (listed 2004), YouTube (2005) and Netflix streaming (2007) grew on the bandwidth and servers the bust made cheap. The winners of the next decade were the users, not the builders.

The winners may still be small or private (Google was founded in 1998, Facebook in 2004). These are examples of listed companies building AI into their products today, not predicted winners.

Area: AI on the device

Grows afterIntelligence gets cheap, Rates cool it

The smaller and cheaper models get, the more they can run on phones and PCs instead of in the cloud, and the more device makers can earn without owning data centres.

In 2000Apple launched the iPod in October 2001, mid-recession, followed by the iTunes Store in 2003 and the iPhone in 2007. It packaged cheaper parts and bandwidth into devices and a store.

Device upgrades depend on the economy, and it is not yet clear that AI will be a reason to upgrade.

How to read this

The scenarios are not predictions and are not mutually exclusive; in 2000–02 several happened at once. The companies are examples of listed firms in each area, with the same numbers and flags as the Survivors page (latest four quarters reported to the SEC). Even in an area that grows after a bust, a company with weak finances may not last long enough to benefit. Foreign and private companies are not included.

This is not investment advice. Survivors: the five tests →

Sources

• Cisco's 2001 inventory charge and job cuts (10-Q, SEC)
• What went wrong at Cisco in 2001 (CIO)
• NVIDIA sheds almost $600bn on DeepSeek (CNBC)
• Boom and bust in telecommunications (Richmond Fed)
• Telecoms crash (Wikipedia)
• Equinix's early years (Equinix blog)
• Fed rate changes (Federal Reserve)
• Apple presents iPod (Apple, Oct 2001)

For the 2000 facts also used on other pages (Global Crossing, Lucent, Verizon, Oracle, Nortel), see the sources on the Dot-com page.