Are we in an AI bubble? Practical clues for finance pros
The stock market keeps breaking highs, and AI is the headline. The $35 trillion question: is this current surge a durable shift or a bubble that ends the usual way? Bubbles are hard to spot in real time, but the clues tend to rhyme.
Here's a simple, finance-first way to check the signals without getting lost in buzzwords.
What economists watch for (and how to apply it)
- Valuations vs. cash flows: Are multiples running ahead of earnings and free cash flow? Track market-wide valuation gauges like the CAPE ratio to see if price is decoupling from fundamentals. CAPE on FRED
- Price acceleration vs. adoption: If prices go vertical while real adoption, margin expansion, or unit economics lag, that's classic froth.
- Leverage and risk appetite: Rising margin debt and easy credit feed bubbles. Watch exposure levels and how quickly investors buy dips. FINRA margin statistics
- Profitless growth: Revenue spikes without a clear path to durable profits end badly once capital costs bite. Look for improving FCF yield, not just sales.
- Hype-to-revenue ratio: Count "AI" mentions on earnings calls, then compare to actual revenue contribution. If the story grows faster than the P&L, be careful.
- Capital intensity and inputs: AI is compute- and power-hungry. Surges in data center capex, grid constraints, and electricity costs can cap returns. IEA: Data centre energy
- Crowding and copycats: When similar pitches raise big rounds with weak moats, excess creeps in. The later the stage, the harsher the reset.
What could pop the enthusiasm
- Rate or energy shocks that pressure discount rates and operating costs.
- Supply bottlenecks (compute, power, cooling) that delay deployments and revenues.
- Regulation or compliance drag that slows enterprise rollout.
- Commoditization from open models that squeezes pricing.
- Earnings misses leading to multiple compression for the bellwethers.
Should we worry?
Bubbles misallocate capital, but they also build useful infrastructure. Railroads, fiber, and now AI compute rarely go to zero. The concern isn't "AI is fake." It's paying any price for growth that takes longer and costs more than the pitch deck suggests.
Time horizon matters. Over 5-10 years, productivity gains can be real. Over 6-18 months, prices can wander far from fundamentals and snap back fast.
Action plan for investors
- Re-test position sizing: If one theme drives your P&L, cap exposure and predefine add/trim levels.
- Prefer cash-generators: Focus on positive FCF, improving gross margins, and realistic payback periods.
- Barbell smartly: Pair AI leaders with "picks-and-shovels" (power, grid upgrades, semis equipment, cooling). Review cyclicality and customer concentration.
- Underwrite energy risk: Model higher electricity prices and delayed data center ramps.
- Hedging and patience: Use staged entries, DCA, and simple hedges. Shorting hype without timing skill can be costly.
Action plan for CFOs and finance leads
- Quantify ROI, not theater: Tie AI spend to specific workflows, baseline metrics, and measured uplift. Kill projects that miss targets.
- Control unit costs: Track tokens/queries per outcome, inference hours, and storage. Renegotiate contracts quarterly as price/perf improves.
- Avoid lock-in traps: Favor portable architectures, usage-based deals with caps, and exit clauses.
- Power and capacity: Treat compute and electricity as strategic inputs. Consider procurement hedges and phased capacity commitments.
- Compliance overhead: Budget for data governance, audit, and monitoring. The cheapest model can be the most expensive later.
A simple dashboard to keep you honest
- Market CAPE vs. long-run average and your required return.
- FINRA margin debt trend vs. index levels.
- AI mentions in earnings vs. AI-derived revenue share.
- Data center capex vs. realized margin expansion.
- FCF yield trend for AI leaders vs. the rest of the index.
- Electricity rates in key build regions and grid interconnection timelines.
If the boom runs longer than you expect
- Keep trimming into strength and rebalance on schedule.
- Favor businesses with pricing power, sticky demand, and credible cost curves.
- Stage exposure to "infrastructure for AI" but watch for overbuild risk.
Bottom line
Call the bubble if you want, but don't bet your future on a label. Track cash flows, inputs, and adoption-then size positions to survive being early or late. Discipline wins either way.
Helpful links
Your membership also unlocks:
