Fund Managers Flash a Red Light on AI Capex
Global fund managers just called out a rare warning: companies are overspending. Bank of America's Global Fund Manager Survey (Nov. 7-13) shows a net 20% of respondents say corporate investment is "too aggressive," the first time this has happened since 2005.
The trigger isn't mystery. It's the scale and financing of the AI buildout-data centers, chips, and the infrastructure behind them. The message to high-growth tech: slow down, hyperscalers.
The data you can't ignore
The survey covered 202 managers with $550 billion AUM. The shift in sentiment is directly tied to AI capex-both how big it's getting and how it's funded. That concern hit the tape as Nvidia shares slid and pulled global equities lower, amplifying the market's sensitivity to AI headlines.
Context matters: "Long Magnificent 7" remains the most crowded trade (54%). When the same names drive both performance and spending, concentration risk compounds.
AI bubble = top tail risk
Investors now rank an "AI bubble" as the biggest tail risk (45%), up sharply from 33% the prior month. Over half of respondents (53%) already think AI stocks are in a bubble. That doesn't mean near-term collapse, but it does mean positioning is stretched and patience for unproven capex is thinning.
Even well-respected voices are uneasy. Jeffrey Gundlach labeled the AI rush "a mania," and Aswath Damodoran warned the odds of a major correction are higher than at any time in two decades. He even suggested hedging with assets outside equities-including collectibles-if you actually know what you're doing.
Positioning is bullish, which is exactly why risk is high
Here's the paradox: sentiment is upbeat. Cash levels fell to 3.7%, the lowest since early 2025. Historically, when cash is at or below 4.0%, it's been a sell signal-BofA notes that in 20 out of 20 instances since 2002, stocks fell in the following one to three months.
If the Fed skips a December cut, the froth can bleed off fast. Keep an eye on the Fed's policy path and credit spreads; financing conditions are the fuse on capex plans.
What CFOs and portfolio managers should do now
- Tighten hurdle rates: Reprice projects with today's true cost of capital, not last year's deck. Stress-test ROI with realistic utilization, latency, and power constraints.
- Stage-gate big builds: Release funds in tranches tied to adoption, workload mix, and unit economics (e.g., revenue per inference/training hour).
- Fix the financing mix: Map debt maturities to asset life. Avoid short-term funding for long-lived infrastructure. Cap variable-rate exposure.
- Audit vendor concentration: Reduce single-supplier risk across GPUs, networking, and power. Negotiate take-or-pay clauses carefully.
- Measure productivity, not hype: Track cost-to-serve, gross margin impact, time-to-value, and payback periods. Kill projects that miss milestones.
- Scenario plan for a demand wobble: Model 20-30% revenue downside with delayed AI monetization. Pre-plan opex/capex cuts and hiring freezes.
- Portfolio risk: Trim crowded mega-cap exposures; rotate into names with positive FCF and disciplined capex. Add tail hedges (put spreads) into event risk.
- Liquidity discipline: With cash at 3.7%, build buffers. Avoid being a forced seller if volatility spikes.
Why this shift is happening
AI capex ramps are colliding with real-world bottlenecks: supply of top-tier chips, grid and substation delays, cooling, and skilled labor. Over-ordering to secure allocation looks smart-until demand normalizes and balance sheets wear the bill.
Markets are also sniffing out circular financing-equity gains funding more capex that aims to justify those gains. That loop works… until it doesn't. If monetization lags, the carry cost shows up in earnings and multiples compress together.
What to watch next
- Earnings quality from AI leaders: backlog vs. realized revenue, utilization of installed GPUs, and pricing power. Track updates on Nvidia's investor relations.
- Credit markets: HY issuance for data center buildouts, spreads, and covenant softness. Financing tells you where discipline breaks first.
- Capex guidance vs. FCF: Look for companies that fund growth from operations, not dilution.
- Hyperscaler commentary: Any pivot from "build at all costs" to "optimize ROI" is a tell for the next leg in this cycle.
The long game still looks promising-if you stay selective
Despite the warning on spend, a slim majority (53%) of managers already see productivity gains from AI. Looking out to 2026, broad AI-driven efficiency is seen as the top bullish catalyst (43%). The bear case remains inflation plus more Fed hikes (45%).
Translation for operators: focus on durable use cases that compound margin, not vanity projects. Translation for investors: own the cash generators with pricing power, and keep a plan if multiples mean-revert.
Practical resources
If you're pressure-testing where AI spend could actually improve P&L in financial workflows, you may find this useful: AI tools for finance.
No one's saying scrap AI. The call is to match ambition with discipline. Tighten the math, stagger the spend, and let results-not narratives-set the pace.
Your membership also unlocks:
