AI: Bubble Watch or Just a Normal Reset? What Finance Pros Should Track in 2026
Three years after generative AI went mainstream, the money keeps coming - and so do the questions. The core tension: the scale of spending versus the pace of monetization. Recent pullbacks in Nvidia and Oracle, plus delays and wobbling sentiment around OpenAI-linked plays, hint at a shift from story to scorecard.
As one portfolio manager put it, we're at the "rubber meets the road" phase. Expectations haven't cooled - but acceleration is no longer a given.
What Could Pop It
AI's weak points are clear: unit economics, access to capital, and consumer willingness to pay. If any of those break, the big indices feel it. The S&P 500's multi-year run has leaned on AI-heavy platforms (Alphabet, Microsoft) and infrastructure plays (Nvidia, Broadcom, and power suppliers).
Sameer Bhasin said it plainly: these stocks don't fall because growth slows - they fall when growth stops accelerating. That's the fulcrum.
Access to Capital: The Stress Test
OpenAI plans to spend $1.4 trillion in coming years while expecting to burn roughly $115 billion through 2029 before turning cash generative in 2030. Fundraising hasn't been a problem so far, including $40 billion from SoftBank and a pledge of up to $100 billion from Nvidia, but circular financing fears are real. If the capital window tightens, that pressure ripples through counterparties like CoreWeave.
Eric Clark warned that when a crowded theme stumbles, exits get crowded fast. That risk increases when financing depends on confidence rather than cash flow.
Debt-Fueled Buildouts
Oracle's surge in AI-related bookings came with a bill: massive data center capex funded by tens of billions in bonds. Debt has a schedule. Equity can wait.
Shares slid after capex ran hotter than expected and cloud sales missed estimates, and reports of data center delays for OpenAI projects added fuel. Oracle's credit risk gauge touched its highest level since 2009 - a reminder that credit markets often spot trouble first.
Big Tech Spending: The Monetization Lag
Alphabet, Microsoft, Amazon, and Meta are on pace to spend north of $400 billion on capex over the next 12 months, mostly for data centers. Revenues tied to AI are growing, but they're still chasing the spend.
Depreciation is the silent drag. Alphabet, Microsoft, and Meta saw depreciation jump from about $10 billion in Q4 2023 to nearly $22 billion in the quarter ended September; ~$30 billion by this time next year is on the table. Expect pressure on buybacks and dividends. Projections show Meta and Microsoft running negative free cash flow after shareholder returns in 2026, with Alphabet roughly at breakeven.
Valuation Reality Check
This isn't the dot-com mania - not in aggregate. The Nasdaq 100 trades around 26x forward earnings versus ~80x at the 2000 peak. The difference: today's leaders have profits and scale.
Still, there are frothy pockets. Palantir trades above 180x estimated profits; Snowflake near 140x. Meanwhile, Nvidia, Alphabet, and Microsoft sit below 30x - expensive, but not panic territory. As Tony DeSpirito noted, exuberance exists, but it's not uniform.
If you want historical context on valuation blow-offs versus cycles, see an overview of the dot-com era here: Investopedia: Dot-Com Bubble.
Signals That Matter for 2026
- Capex-to-revenue convergence: Do AI revenues close the gap with data center spend by mid-2026?
- Gross margin trajectory in cloud: Watch AI VM pricing, discounts, and utilization; margin compression is the tell.
- Depreciation as a % of revenue: Rising lines here squeeze EPS and limit buybacks.
- Free cash flow after shareholder returns: Who can still retire shares and raise dividends without tapping debt?
- Credit spreads and CDS for AI infra names: Early warning system for funding strain.
- Energy costs and availability: Data center delays tied to power supply are first-order risk for project timelines.
- Customer willingness to pay: Enterprise AI seat pricing, usage caps, and renewals vs. pilot fatigue.
- Vendor financing and "circular money": Track where chipmakers are funding customers who then buy chips.
Where the Rotation Could Show Up
Growth deceleration is the trigger. If acceleration stalls, investors migrate from speculative AI beneficiaries to cash compounding franchises with clean balance sheets. Not a crash - a recalibration.
Bhasin expects the groupthink to crack. The path of least resistance is a rotation, not a collapse.
Portfolio Playbook
- Favor cash-first balance sheets: Net cash, consistent buybacks funded by operations, and transparent depreciation ramps.
- Underwrite with unit economics: Tie spend to payback periods, not press releases. Scrutinize AI gross profit per dollar of capex.
- Stress test funding: Assume a tighter credit window and higher spreads; avoid names reliant on perpetual external financing.
- Segment AI exposure: Core platforms (cloud/ads), infra (chips/power), applications (vertical SaaS). Allocate by monetization visibility.
- Use valuation discipline: Pay premium multiples only for accelerating revenue with improving FCF conversion.
- Watch depreciation inflection: Peak depreciation often precedes margin rebuild - time entries accordingly.
Scenarios to Price In
- Base case: Growth continues, but at a slower clip; multiples settle, leaders keep out-earning the field.
- Bear case: Funding tightens, data center delays spread, and key projects slip; rotation favors profitable, lower-multiple compounders.
- Bull case: Monetization from AI assistants and enterprise copilots scales faster than expected, offsetting depreciation drag.
Bottom Line
AI isn't broken. The financing model and the pace of monetization are the variables. If capex growth outpaces revenue for too long, multiples compress and buybacks fade.
Track acceleration, not headlines. And if you're building internal competency on AI to validate vendor claims and budget impact, explore curated resources for finance teams here: AI Tools for Finance.
Your membership also unlocks:
