Why this strategist is getting cautious on the AI trade
AI has driven a big chunk of equity returns over the past year. Yet Hirtle Callaghan CIO Brad Conger is tapping the brakes. His stance: keep exposure, but scale it back. The reason is simple-too many open questions, too much perfection baked into prices.
The case for caution
AI is still sorting out its winning business models. Who captures the economics-model providers, hyperscalers, chipmakers, or software firms that bundle AI as a feature? Where the compute happens (centralized vs. edge) changes the margin structure and who keeps the dollars.
At current prices, the market feels like it's underwriting multi-trillion revenue runways with very little friction. That can work-until it doesn't. If the next wave of models delivers less uplift than expected, or if open-source breakthroughs (think DeepSeek-style leaps) reset costs and pricing, momentum can stall fast.
What could crack the story
- Model progress slows: future iterations (e.g., a hypothetical "GPT-6") don't deliver step-change gains.
- Open-source compression: community models improve, costs fall, and premium pricing gets harder.
- Capital intensity: capex outpaces monetization, dragging returns on invested capital.
- Unit economics: inference costs stay high while customers treat AI as a bundled feature, not a budget line.
- Data constraints or regulation increase friction and timelines.
Positioning: trim, don't torch
Conger's team is underweight the AI complex and mega-caps. If AI-related names plus the Magnificent Seven sit near 40% of the market, they run closer to 27-28%. That's a clear tilt without taking career-risk extremes. You keep core exposure and cut the concentration risk.
- Set position caps and pre-plan trims into strength.
- Favor cash flow yield and balance-sheet quality over pure narrative beta.
- Use options or pair trades to hedge the fattest tails.
Where to be overweight: Europe on valuation
His offset is Europe. The valuation gap vs. the U.S. is wide by historical standards. Europe doesn't need U.S.-style multiples; it just needs "less bad" plus stable earnings. Recent prints have been solid, which helps sentiment and the multiple.
- Focus on quality cyclicals, select financials with capital return, and industrials with backlog visibility.
- Mind FX: decide whether to hedge EUR/GBP based on your base currency and policy.
- Simple vehicles: the MSCI Europe Index or a STOXX 600 tracker; layer active sleeves for idiosyncratic alpha.
What to watch next quarter
- Hyperscaler capex guides vs. AI service revenue attribution.
- Unit economics: inference cost per query, gross margin trend on AI workloads, and payback periods.
- GPU supply/demand, lead times, and pricing.
- Model efficiency benchmarks and any plateauing in real-world uplift.
- Earnings breadth and revisions in Europe; any narrowing of the U.S.-EU multiple gap.
Practical notes for finance teams
Treat AI like a product investment. Define the job to be done, run time-boxed pilots, measure cash impact, and sunset what doesn't clear hurdle rates. Beware vendor lock-in; keep optionality across providers.
- Set a simple ROI scorecard: adoption, time saved, error rates, gross margin effect.
- Standardize data access and prompts; audit outputs before scaling.
- Keep contracts short and modular until the economics are proven.
If you're evaluating tools for FP&A, audit, or research workflows, this curated list can save time: AI tools for finance.
Bottom line
The AI story is real, but expectations are doing a lot of heavy lifting. Conger's playbook-dial down concentration, add Europe on valuation, and demand proof on unit economics-is a sensible way to keep upside while respecting risk. In markets like this, sizing beats heroics.
Your membership also unlocks:
