Top AI investors say maybe it’s a bubble, but ‘bubbles are good for innovation’

At Fortune’s Brainstorm AI panel, VCs Steve Jang and Cathy Gao characterized the current AI financing boom as potentially a ‘‘bubble’’ but one that can accelerate innovation; Jang says capital should flow to AI infrastructure (chips, GPU marketplaces, specialized frontier models) while Gao favors enterprise application-layer companies that embed AI into complex workflows. Both warned of vulnerabilities: inflated valuations in pockets, a likely shakeout in consumer and humanoid robotics built on immature models, and persistent enterprise hurdles around trust and visibility that could slow adoption into 2026. The investment takeaway for allocators is a bias toward foundational infrastructure and deeply integrated enterprise software, and caution on AI-for-X consumer plays and robotics startups with long adoption cycles.

Analysis

Market structure: The biggest, durable winners are AI infrastructure owners — GPU/chip suppliers and hyperscalers — who retain pricing power while GPU supply constraints persist into 2025 (expect gross-margin support of +200–500bps for top providers). Enterprise workflow SaaS (CRM, WDAY) is a second-tier winner where AI embeds into high-ARPU workflows and can justify premium multiples; standalone “AI-for-X” apps and consumer/ humanoid robotics are losers facing long sales cycles, liability/regulatory drag and obsolescence risk.

Risk assessment: Tail risks include a regulatory shock (10–25% prob over 12–24 months) that raises compliance costs >3–5% ARR for enterprise sellers, and a model breakthrough that renders many startups obsolete (20–30% prob within 18 months). Near-term (days–weeks) expect sentiment-driven volatility; medium-term (3–12 months) expect re-rating of late-stage private comps and public mid-caps; long-term (2–5 years) market will concentrate into 3–5 platform winners.

Trade implications: Favor long exposure to CRM and WDAY and select infra (NVDA, GOOGL) while hedging through short exposure to pure “AI feature” small-caps with < $500m revenue and no >$100m ARR runway. Use 6–12 month call spreads on leaders to cap premium and buy 6–9 month puts on a small-cap AI basket to capture downside; rotate from alpha in apps into infra if CRM/WDAY outperforms by >20%.

Contrarian angles: The consensus underestimates persistent oligopoly risk — compute centralization will boost cloud/chip free cash flow even as app multiples compress. Robotics doom may be over-discounted for industrial automation (select industrial robot suppliers with >$200m revenue and 30% gross margins are attractively priced). Historical parallel: dot-com era produced dominant infrastructure survivors despite many failed apps — position accordingly.

Key Decisions for Investors

• Establish a 2% portfolio long position in CRM (Salesforce) funded by 6–9 month put protection; complement with a 12-month 10% OTM call spread to retain upside while capping cost (target IRR if CRM up 20%+ in 12 months).
• Allocate 1.5% to WDAY (Workday) via outright shares plus a 9–12 month 15% OTM call spread; trim if shares rally >30% or if enterprise deal signings miss consensus by >10% in next two quarters.
• Allocate 2% to AI infrastructure: 1.5% NVDA (or equivalent chip exposure) and 0.5% GOOGL (cloud) — use 9–12 month calls (or call spreads) to get directional exposure while limiting theta decay.
• Establish a 1% short/high-conviction hedge: buy 6–9 month puts on a basket of public "AI-for-X" small-caps (market cap < $5bn, revenue < $500m) sized to offset 30–50% of gross exposure; exit if any single name shows >50% revenue growth or secures a >$50m contract.
• Trigger-based action: If a major regulator (US/EC/UK) publishes binding AI rules within 90 days that imply >5% incremental compliance cost for enterprise vendors, reduce public app longs (CRM, WDAY) by 50% within one week and reallocate into infra (NVDA/GOOGL) cash positions.