Telecom Bubble
The telecom bubble refers to the investment boom in phone and internet lines around the year 2000, which ended in a collapse of the industry. Today it serves as a point of comparison whenever the multi-billion-dollar buildout of AI data centers is discussed.
In the late 1990s, many companies believed the internet would soon be moving gigantic amounts of data. So, using borrowed money, they laid thousands of kilometers of fiber-optic cable, that is, lines made of thin glass fibers through which data travels as light signals. But actual demand grew far more slowly than planned. A large share of the cables lay unused in the ground while the loans still had to be repaid. Between 2000 and 2003, the share prices of telecommunications companies collapsed, and many went bankrupt. In hindsight, this pattern of overinvestment, overcapacity, and collapse is called the telecom bubble.
Why the cables from back then are a topic again today
The term has been popping up constantly in business news since 2023. The reason is the construction of data centers for artificial intelligence. Large technology companies together are pouring several hundred billion dollars per year into buildings, power connections, and specialized chips. Back then, too, it was infrastructure, meaning basic technical facilities, that was built on spec, in anticipation of future demand.
The question is: is this computing power actually needed? If demand for AI services grows more slowly than the buildout, empty data centers will emerge just as empty cables once did. Investors, analysts, and central banks are watching this closely. Anyone who calls an investment the “new telecom bubble” is saying: the technology is real, but expectations are too high.
It’s important to distinguish this from the more famous dot-com bubble of the same era. In that case, it was mainly internet startups without a business model that lost their stock market value. The telecom bubble, by contrast, involved solid, expensive construction projects undertaken by established corporations. That’s precisely why the comparison fits data centers better than the dot-com comparison does.
How optimism turns into overcapacity
It starts with a forecast that everyone believes. Around 2000, it was said that data traffic was doubling every three months. That figure was wrong, but it was passed on from company to company. Businesses used it as the basis for calculating how many lines would be worthwhile.
On top of that comes competitive pressure. Whoever builds first claims the market, so everyone builds at the same time. This is financed with debt, because the money is only expected to flow back years later. As long as optimism holds, banks are happy to hand out these loans. In the construction of fiber-optic networks, it was also underestimated that a single fiber could suddenly transmit far more data thanks to better technology. So capacity grew twice over: through new cables and through new devices at the endpoints.
The crash begins when a single provider cuts prices to fill its empty lines. Then prices fall for everyone, revenues collapse, and the debts remain. In the US, the price of data transmission fell by more than 90 percent within a few years. The company WorldCom filed what was then the largest corporate bankruptcy in US history in 2002.
The term in stock market news and chip reports
You’ll usually encounter the term as a warning. Typical sentences include: “Does the AI boom resemble the telecom bubble?” The trigger is often quarterly results from chipmakers or new expansion plans from cloud providers, that is, companies that rent out computing power over the internet. The comparison also comes up regarding power grids and power plants for data centers.
A common misconception is that a bubble means the technology was useless. The opposite is true. The surplus fiber optics from the 2000s were later sold off cheaply and, from around 2007 onward, carried YouTube, Netflix, and video calls. The investors lost their money; society got an inexpensive network.
For you as a reader, this means: the comparison says something about prices and timing, not about the usefulness of a technology. Whether AI data centers suffer the same fate remains an open question. One difference is often cited: so far, the major AI investors are paying mostly out of ongoing profits, not out of loans. A second difference: chips become obsolete within a few years, while fiber optics last for decades.