It is TACO Tuesday but our fearless leader is too busy Flooding the zone —kissing dictator’s rears while killing our election system and stage managing an invasion of the Capital - all to stop the Epstein-Epstein-Epstein noise. So we’ll let the Bulwark cover that and we’ll continue our discussion from last post, on AI and the improbability of its economics.
The AI revolution is being sold as the next electricity. But history whispers a harsher warning: it might just be the next telecom bust. Even more devastating than dot.bomb, the Telecom bust ended in huge scandals - Global Crossing and Enron as just two examples, and triggered a severe recession that took several years to recover from. As background: in the late 1990s, telcos laid oceans of fiber, spent half a trillion dollars on gear, and triggered a historic bubble. Today, AI is retracing those steps only faster, and with more money at stake. And with that, on with the show…
Goldman’s Red Flag: Build First, Profit Later (Maybe)
Goldman Sachs recently issued a sobering assessment: AI is still “squarely in build mode”—with billions funneled into servers, GPUs, and energy-hungry data centers, but precious little real revenue flowing back. Like telecom before it, we may be building networks no one’s ready to use.
The Spending Spree: Telecom Then vs. AI Now
The parallels are striking:
Telecom (1996–2001)
CapEx exploded from ~$62B to $135B (18% annual growth).
By 2000, spending peaked at $213B (2025 dollars).
More than $500B in equipment + $1T in M&A sunk into the sector.
When revenues lagged, the bubble burst—leaving “dark fiber” and bankruptcies in its wake.
AI (2024–2025)
CapEx rose from $246B in 2024 to $320B in 2025 across Amazon, Microsoft, Alphabet, and Meta.
Amazon alone budgeted $100B+, Meta up to $72B, Google $85B, Microsoft $30B in a single quarter.
Morgan Stanley estimates nearly $1T in AI investment between 2024–2026; FT warns the global AI infrastructure binge could top $3T by 2029.
The scale dwarfs telecom. And the risk, if demand doesn’t materialize, is just as large.
Dependency and Fragility: AI as Critical Infrastructure
This isn’t just an industry story. As I’ve covered in previous post, the AI buildout has juiced U.S. GDP itself. Pantheon Macroeconomics estimates AI CapEx added 0.5 percentage points to GDP growth in early 2025. Strip that away, and the economy would’ve limped along under 1%. That means the economy is now codependent on AI spending. If AI adoption stalls, Wall Street won’t just face a tech correction—it could drag down broader growth. The U.S. economy is effectively leaning on GPU orders and hyperscaler megaprojects for its momentum. That’s a wishin’ and a hopin’ of the highest order.
Titans on Shaky Ground
Microsoft and Nvidia (valued at over $3 trillion) now power the backbone of the global AI-driven economy. But the fragile dependencies they've built mean that the health of Microsoft's Azure-OpenAI alliance and Nvidia’s chip dominance now represent existential risk—not just to tech stocks, but to the broader U.S. economy.
Lessons from the Bust
When the telecom bubble burst, the fallout was catastrophic: job losses, bankruptcies, and years of stagnation before fiber was finally utilized. AI could follow the same script:
Massive overbuild: Data centers and GPU clusters that sit idle.
Profit drought: Few paying customers (Microsoft’s Copilot adoption remains weak), while costs skyrocket.
Stranded capital: Trillions locked in sunk costs that investors never recoup.
Unlike the dot-com bust, though, AI’s collapse would ripple wider: through energy markets, chip supply chains, cloud pricing, and even sovereign debt loads (given how central AI CapEx has become to GDP).
Patience or Peril?
Goldman counsels patience: “Betting on the long term in technology seems to have worked… but you do need some patience.” Maybe. But patience doesn’t pay interest on $1T worth of sunk investment.
The telecom bust left behind fiber we still use today—dark strands that only lit up a decade later. But if AI crashes, what’s left behind? Overheated data centers? Obsolete GPUs? The wreckage of an overhyped dream?
The AI boom is built on faith that use cases, subscriptions, and enterprise productivity will eventually arrive. But history says faith isn’t a business model.
Final Thought: When Infrastructure Becomes the Bubble
We’ve reached a paradox. AI is sold as the future, but its present is a drag on balance sheets. It props up GDP while offering little in consumer or enterprise revenue. It’s a bubble of infrastructure—and infrastructure bubbles, once they pop, leave behind scars far larger than failed apps or speculative coins.
The last time America built too much network too fast, it took down an industry. This time, it could take down the economy.
Smash the machines!