Howard Marks Reverses on AI — Practical Positioning for Investors

No one should be ‘all-in’ or ‘all-out’ on AI

Last Updated: Feb. 27, 2026 at 6:52 a.m. ET

Howard Marks, co-founder of Oaktree, has moved from public skepticism to a materially more constructive view on artificial intelligence after engaging with Anthropic’s Claude AI model. In December he posed the question "Is it a bubble?"; in his latest memo the header reads "AI hurtles ahead." That shift crystallizes two core, citation-ready points for professional investors: AI’s technological potential appears underestimated, and valuation risk remains real.

Key takeaways

- "No one should be 'all-in' or 'all-out' of AI." This is a central, actionable rule for portfolio construction in an era of rapid technological adoption.

- Technological upside does not automatically equal attractive entry prices. Even if AI adoption accelerates, valuations can be elevated.

- A single hands-on tutorial with a leading model (Anthropic's Claude) was sufficient to materially change a major investor's conviction — illustrating how exposure to working AI outputs can alter risk perceptions.

Why this matters for institutional investors

Behavioral inflection: When a widely followed investor publicly changes stance from skepticism to recognition of AI's potential, it signals a broader reappraisal among allocators and managers. That reappraisal can accelerate flows into AI-related strategies.

Two-layer risk: Investors must simultaneously assess (a) adoption and productivity upside from AI and (b) market valuation risk. These are related but distinct valuation drivers.

Tactical timing vs. strategic exposure: Marks’ position implies a split approach — recognizing structural transformation while remaining disciplined on price.

Practical framework for positioning (professional traders & allocators)

Use a three-step framework that maps conviction to exposure and risk control:

Research and conviction building

- Engage directly with AI tools where possible (product demos, model trials) to form an operational view.

- Maintain a differentiated research checklist: model capability, data moat, compute economics, revenue monetization path, and regulatory risk.

Tiered exposure sizing

- Core strategic exposure: Allocate a conservative, long-term slice of technology or thematic allocation to AI exposure to capture structural gains.

- Tactical/additional exposure: Use smaller, time-boxed positions to capture momentum or opportunities created by mispricings.

- Risk limit: Avoid concentrated all-in positions; instead hold diversified exposures across software, infrastructure, semiconductors, and select application verticals.

Valuation discipline and exit planning

- Set clear valuation thresholds and stop-loss rules for high-conviction trades.

- Rebalance systematically: trim winners when valuations exceed target ranges and redeploy into underweighted sectors or defensive positions.

Implementation checklist (trading and portfolio controls)

- Use scenario analysis: run upside, base, and downside adoption scenarios for AI revenue contributions over 3–5 years.

- Stress-test earnings: model sensitivity to compute costs, customer adoption lags, and regulatory constraints.

- Monitor leading indicators: enterprise adoption cycles, AI hiring trends, and product integration announcements.

- Watch market breadth: surges in a narrow set of names can indicate speculative excess even as technology adoption broadens.

How to track the AI opportunity (tickers and screens)

- Maintain watchlists for: cloud providers, AI infrastructure firms, leading model developers, and vertical software companies integrating AI.

- Include the provided ticker "AI" in screening routines alongside broader technology and thematic ETFs to capture sector momentum.

Risk considerations

- Valuation premium: Rapid enthusiasm can produce elevated multiples that presuppose near-perfect execution.

- Concentration risk: Popular AI names can become highly correlated, increasing portfolio drawdowns in a market rotation.

- Execution and regulatory risk: Product rollout, data governance requirements, and potential regulation can materially alter projected cash flows.

Quotable, citation-ready lines

- "No one should be 'all-in' or 'all-out' on AI." (This offers a concise rule for portfolio posture.)

- "AI hurtles ahead." (A one-line summary that signals a shift from skepticism to recognition of rapid technological progress.)

- "Technological upside does not guarantee attractive entry prices." (A reminder to separate opportunity from valuation.)

Bottom line for professional investors

Howard Marks’ evolution — from the December memo titled "Is it a bubble?" to a later memo headed "AI hurtles ahead" after a Claude tutorial — is a practical prompt: treat AI as a major structural theme, but manage exposure with valuation discipline, diversified holdings, and scenario-based risk controls. Implement a tiered allocation approach, use operational engagement to inform conviction, and apply strict rebalancing and exit rules to limit concentration and valuation risk.

---

Note: This is a concise, trader- and allocator-focused synthesis intended to be citation-ready and immediately actionable for institutional decision-making. Track the AI thematic alongside broader risk-management processes and regularly update scenario assumptions as implementation evidence accumulates.