AI Euphoria, Real Risks: What Finance Leaders Should Do Before the Next Volatility Spike
The AI boom is pushing markets into a new risk cycle. Executives are paying premiums to buy capabilities and data they can't build fast enough, while investors focus on future profits over present fundamentals.
Matthew Danzig, managing director at Lazard, called AI the "number one topic of conversation" for both investors and operators. "Every company that's a potential target is figuring out their AI angle," he said-often by acquiring teams, models, or proprietary datasets to stay competitive.
The result: valuations at historic levels and capital needs that dwarf prior tech waves. McKinsey estimates roughly $7 trillion in capital by 2030 just to fund data centers. Meanwhile, investors are shrugging off rising leverage and thin near-term revenue to service that debt.
We're already seeing sentiment whiplash. Nvidia's shares slipped after blockbuster results and a soaring market cap, reigniting talk of an AI bubble and pulling other tech names lower.
Volatility Hits Faster, Fades Faster, Repeats More Often
Joanna Welsh, Chief Risk Officer at Citadel, said their models show modern markets amplify shocks. "Markets are just faster," she noted. "These volatility spikes and pulses, they hit harder, they fade faster, they repeat more often." Citadel, with $71 billion AUM, is positioned for drawdowns at any time.
Under the surface, credit risk is stacking. Welsh highlighted a surge in high-quality corporate issuance of 30- and 40-year bonds to fund assets with roughly four-year depreciation cycles. Translation: you could be servicing debt long after the asset is obsolete.
What this means for CFOs, Boards, and Strategy Leads
- Force capital discipline: Build vs. buy vs. partner should be a live decision, not a slogan. Tie spend to clear unit economics, payback windows, and specific outcomes (gross margin lift, cost per inference, cycle-time reduction).
- Stress-test the liability stack: Model revenue coverage against debt across rate paths and demand shocks. Set hurdle rates that exceed your weighted average cost of capital and require payback inside the asset's useful life.
- Match duration to reality: Funding 4-year tech with 30-year paper invites obsolescence risk. If you use long-dated debt, hedge duration or structure step-downs and call options that align with refresh cycles.
- Price M&A with obsolescence curves: Assume model decay and compute cost deflation. Use milestone-based earnouts tied to adoption, gross margin, and data rights durability-not just top-line promises.
- Interrogate vendor economics: Ask about gross margin after compute, access to GPUs, guaranteed capacity, and energy contracts. Validate data provenance and IP indemnities. If the moat is "we'll raise more," walk.
- Pick an infrastructure path you can fund: Own, lease, or colocate-run total cost of ownership across energy, cooling, networking, staffing, and refresh cadence. Capacity reservations can become stranded if demand slips.
- Upgrade risk playbooks: Expect repeated volatility pulses. Set liquidity buffers, predefine de-risk triggers, and run cross-asset shock scenarios (equity, credit, rates, commodities) hitting at once.
- Make strategy tangible: Publish 3-5 operating KPIs: time-to-value, model adoption by workflow, incremental gross margin, cost per model call, SLA reliability. Tie exec comp to those, not vanity metrics.
Portfolio and Treasury Moves to Consider
- Trim concentration in single-name AI winners; use pairs or factor hedges to reduce one-directional exposure.
- Watch credit: long-duration issuance rising against short-life assets is a red flag. Monitor spreads and refinancing windows.
- Term structure management: stagger maturities, preserve optionality, and avoid covenant traps tied to growth targets.
- Liquidity first: raise dry powder into strength; deploy into dislocations, not euphoria.
Leading Indicators Worth Tracking
- Announced data center capex versus realized revenue and utilization.
- Volume of 30-40 year bond issuance for assets with sub-5-year refresh cycles.
- GPU order backlogs versus delivery timelines and take-or-pay commitments.
- Energy availability, pricing, and permitting delays in key regions.
- Stock-bond correlation flips during earnings and macro prints.
- Earnings quality: reliance on stock-based comp and capitalization of R&D.
Boardroom Questions for Any AI Deal or Major Spend
- What must be true for this to pay back inside the asset life-and what if demand is 50% of plan?
- Where do we have proprietary data advantages that compound? Are rights and indemnities airtight?
- How do we exit or pivot if infra costs or access (compute, power) tighten?
- What's our plan if volatility knocks 30% off equity and widens credit by 200 bps next quarter?
The growth story is strong. The math has to be stronger. If markets are moving faster, your diligence, capital allocation, and risk controls have to move faster too.
Need structured ways to evaluate the stack of tools and vendors in finance? See a curated set of AI tools for finance to pressure-test use cases and ROI assumptions.
Your membership also unlocks:
