In 1998–2000, telecoms poured over $100B into fibre, betting that internet demand would explode.
They were right. Then they went bust.
At the bust, around 7% of that fibre was being used. Global Crossing and others collapsed. A decade later, that same “dark fibre” carried Netflix, YouTube and the cloud.
The builders were right about the destination and wrong about the timing.
Fast forward to 2026. Big Tech will spend roughly $725B on AI infrastructure this year alone.
Same thesis. Same FOMO. Same vendor financing.
Two big differences:
– AI data centres are running near capacity, not sitting dark.
– GPUs depreciate in 3–5 years. Fibre lasts decades.
So if there’s a correction, it’ll clear faster, but it won’t leave cheap infrastructure behind for decades.
The lesson from fibre isn’t “avoid the boom.” It’s this – the value went to those who built on the infrastructure, not those who owned it.
For enterprise leaders, the question isn’t whether AI is a bubble. It’s whether you’re ready to capture the value when compute gets cheap.
Are we building the next internet, or the next Global Crossing?