As money rushes into AI chips, data centers, and fiber, one concern keeps coming up: are we replaying the 1990s telecoms bubble. Investors, executives, and workers want to know what comes next, and what to avoid.
“How similar is it to the 1990s telecoms bubble?”
The late‑90s surge in telecom spending ended in a brutal bust. That period saw heavy borrowing, an overbuild of capacity, and a wave of bankruptcies. Today’s buildout has a different core—AI and cloud—but some of the warning signs look familiar.
What the 1990s looked like
Telecom companies poured money into fiber networks, long‑haul cables, and switching gear during the dot‑com boom. Cheap credit and new competitors fueled the race. Many firms counted on boundless demand and rising prices.
Then the cycle turned. Demand grew, but not fast enough to fill the new pipes. Prices fell as rivals fought for customers. Firms with high debt loads were hit first. WorldCom and Global Crossing collapsed. Spending on gear and spectrum froze for years.
Why people see a parallel now
Today’s buildout centers on AI training, cloud services, and high‑capacity power and cooling. Big Tech is ordering advanced chips at record pace. Developers are leasing land for large data centers near major grids.
Some features echo the 1990s:
- Rapid capacity growth ahead of clear demand visibility.
- Heavy upfront costs financed with debt or long contracts.
- New entrants chasing scale advantages.
There is also a marketing cycle. Firms promise new use‑cases that will fill the machines. In the 1990s, it was long‑distance traffic and internet backbones. Now it is AI copilots, search, and automation across many sectors.
The case for key differences
Yet the current cycle does have important contrasts. A few massive buyers dominate orders and have strong cash flow. They can adjust plans faster than the fragmented carriers of the past. Pricing power in AI compute looks stronger than long‑distance calling ever was.
Much of the fiber installed in the 1990s still carries traffic today. That overbuild later became a base for streaming and cloud. A similar pattern could occur if new data‑center capacity gets repurposed as workloads catch up.
Regulation is also different. Telecom pricing and licenses shaped the last boom. Today’s issues center on power access, permitting, and chip supply. These factors may slow growth and reduce the risk of pure overbuild.
Signals to watch
Several indicators can help separate healthy growth from a bubble pattern:
- Utilization: Are GPUs and racks running near capacity, or sitting idle waiting for software and customers?
- Unit economics: Do AI services earn margins after power, cooling, and chip costs?
- Leverage: Are builders funding projects with sustainable debt and long contracts?
- Pricing trends: Are compute prices holding, or under pressure from fast supply growth?
- Customer diversity: Is demand broad, or dependent on a few buyers and pilot projects?
Industry impact and who wins or loses
Equipment makers and chip vendors benefit first in both cycles. In the 1990s, gear sales peaked before networks paid off. Today, chip and power suppliers could peak before software demand matures.
Utilities and grid planners face new stress. Data centers need steady power and transmission upgrades. Delays here could slow the cycle and favor regions with surplus capacity.
For startups, the risk is building for hype. If customer budgets tighten, providers with long‑term contracts and cash reserves will last longer. Smaller firms may need partnerships to access infrastructure without overcommitting.
What experts are debating
Analysts split on the core question. Some argue that AI demand will scale with new apps in health, finance, and media. They point to fast adoption of code assistants and image tools as early proof.
Skeptics say current spending assumes mass‑market use that is not here yet. They note that inference costs must fall, or pricing must rise, for profits to hold.
The lesson from the 1990s is clear. Capacity that earns a return survives. Capacity that relies on cheap money and heroic forecasts does not. The current buildout may avoid the worst outcomes if planners match projects to firm demand and manage debt.
For now, the question remains open. Watch utilization, pricing, and leverage in the quarters ahead. If they hold, the cycle could settle into steady growth. If they slip, the echoes of the telecoms bust will grow louder.
