I keep hearing the AI datacentre boom compared to the 2000s telecoms crash. The parallels seem obvious - billions in infrastructure spending, concerns about overbuilding, warnings of an imminent bubble. But when I actually ran the numbers, the fundamentals look completely different.
I'm not here to predict whether there will or won't be a crash or correction. I just want to look at whether the comparison to telecoms actually holds up when you examine the history in a bit more detail.
What Actually Happened in the Telecoms Crash
Let me start with what the 2000s telecoms crash actually looked like, because the details matter. Firstly, there was massive capex - between 1995 and 2000 somewhere like $2 trillion was spent laying 80-90 million miles of fiber. Inflation adjusted, this is over $4trillion, or close to $1trillion/year in 2025 dollars.
By 2002 only 2.7% of this fibre was used.
How did this happen? A catastrophic supply and demand miscalculation past the pure securities fraud involved in many of the companies. Telecom CEOs claimed internet traffic was doubling every 3-4 months.
But in reality, traffic was doubling roughly every 12 months. That's a 4x overestimate of demand growth, which compounds each year. This false assumption drove massive debt-financed overbuilding. If you overestimate 4x a year for 3 years, by the end of your scenario you are 256x out.
Even worse for these companies, enormous strides were made on the optical transceivers, allowing the same fibre to carry 100,000x more traffic over the following decade. Just one example is WDM multiplexing, allowing multiple carriers to be multiplexed on the same physical fibre line. In 1995 state of the art was 4-8 carriers. By 2000, it was 128. This alone allowed a 64x increase in capacity with the same infrastructure. Combined with improvements in modulation techniques, error correction, and the bits per second each carrier could handle, the same physical fibre became exponentially more capable.
The key dynamic: supply improvements were exponential while demand was merely linear. While some physical infrastructure needed to be built, there was enormous overbuilding that could mostly be serviced by technology improvements on the same infrastructure.
AI Infrastructure: A Different Story
... continue reading