The $650 Billion Question: Is AI's Infrastructure Boom the Next Fiber Optic Bubble?
Big Tech will spend $650B on AI infrastructure in 2026 alone. The last time the tech industry built this aggressively, 96% of the fiber went dark. Here's why this time might — or might not — be different.
By Maya Lin Chen, Product & Strategy · Oct 3, 2025
Big Tech is spending $650 billion on AI infrastructure in 2026. This analysis compares the AI capex boom to the 1999 fiber optic bubble, examining whether the GPU buildout will end in overcapacity or a new computing paradigm.
Frequently Asked Questions
How much is Big Tech spending on AI infrastructure in 2026?
The six largest AI infrastructure spenders (Microsoft, Google, Amazon, Meta, Oracle, and Apple) are collectively projected to spend over $650 billion on AI infrastructure in 2026, according to Wedbush Securities. Microsoft alone plans roughly $80 billion in capex, with similar figures from Google and Amazon. This exceeds the total global spending on telecom infrastructure during the peak of the fiber optic buildout in 1999-2000, adjusted for inflation.
What was the fiber optic bubble?
The fiber optic bubble (1996-2001) saw telecom companies invest over $150 billion (roughly $300 billion inflation-adjusted) in fiber optic cable infrastructure, driven by projections that internet traffic would grow 1,000% annually. Companies like WorldCom, Global Crossing, and JDS Uniphase built massive fiber networks. When demand grew slower than projected, 96% of installed fiber went 'dark' (unused). The resulting crash destroyed $2 trillion in market value and bankrupted dozens of telecom companies. However, the infrastructure eventually became valuable — the fiber laid in 1999 powers today's internet.
Is AI in a bubble in 2026?
The AI infrastructure buildout shares structural similarities with the fiber optic bubble — massive capital expenditure driven by demand projections that may not materialize on the expected timeline. The bear case: AI application revenue ($50-100B) is a fraction of infrastructure investment ($650B), creating a 6-13x capex-to-revenue ratio that mirrors the fiber bubble's imbalance. The bull case: unlike fiber (a commodity), GPU infrastructure has multiple monetization paths (training, inference, fine-tuning), and the major spenders (Microsoft, Google, Amazon) are profitable companies, not leveraged startups.
Will AI infrastructure spending lead to a crash?
The most likely outcome is not a dramatic crash but a capex hangover — a period in 2027-2028 where spending slows as companies digest the infrastructure they've built. This mirrors what happened with cloud infrastructure: AWS, Azure, and GCP all went through periods of overbuilding followed by demand catching up. The key risk isn't that the infrastructure is worthless (it's not), but that the companies spending $650B in 2026 will earn returns on that investment more slowly than their projections assume, leading to earnings misses and stock price corrections rather than bankruptcies.
Related Articles
Topics: AI, Strategy, SaaS
Browse all articles | About Signal