The Motley Fool Interviews NYU Professor Vasant Dhar: Thinking With Machines

NYU Professor Vasant Dhar discusses his new book Thinking With Machines and describes efforts to recreate Aswath Damodaran's valuation approach using modern LLMs — after simple fine‑tuning failed, his team decomposed Damodaran's quantitative 'Ginsu' spreadsheet and narrative framing to build a more reliable bot for scaling analysis and scenario work. He argues small probabilistic edges compound in markets and that AI can both amplify investor capabilities (e.g., systematic valuation, scenario stress-testing such as tariff impacts) and risk disempowering humans by becoming a gatekeeper, calling for consumer awareness and broader stakeholder involvement in governance.

Analysis

Market structure: AI accelerates concentration of economic rents to a few hardware and platform leaders (NVDA, major cloud vendors, exchanges capturing flow). Expect 6–18 month margin expansion for market-share leaders if GPU supply remains tight and ASPs hold; mid/late-cycle incumbents (commodity CPU vendors, small consultancies) face margin pressure. Higher trading volumes and quant activity imply positive volume/fee tailwinds for exchanges (NDAQ) and sell-side quant desks (benefitting MS-like franchises).

Risk assessment: Tail risks include rapid regulatory intervention (export controls, data/privacy rules) or an earnings shock that re-prices long-duration AI growth (>30% revenue CAGR priced in many names); those are low-probability but can cause 30–60% drawdowns in leaders within weeks. Short-term (days–weeks) will be volatility/news driven; medium-term (3–12 months) depends on supply (TSMC/Nvidia capacity) and model/product cadence; long-term (2–5 years) is structural adoption and possible concentration or fragmentation. Hidden dependencies: TSMC, ASML, cloud providers, and model-ops talent; catalyst set includes Nvidia earnings, US export announcements, and major model launches.

Trade implications: Direct play is concentrated long exposure to NVDA (prefer options to control downside) and modest long in MS and NDAQ to capture fee/flow upside; avoid broad consumer cyclicals where AI benefit is diffuse. Use 6–18 month call spreads or LEAPs to capture upside while limiting premium decay; trim on >30–50% rallies or IV collapse >25%. Entry on pullbacks of ~15% or after predictable IV sell-offs (post-earnings) is preferred.

Contrarian angles: Consensus underweights the probability of regulatory fragmentation that would favor specialized edge vendors and services over a single-stack winner; NVDA multiples may already price >30% CAGR — a regulatory or demand miss could be overdone. Historical parallel: 1999–2002 internet winners were few; expect a similar ‘winner-take-most’ outcome, but also larger winner collapses if policy or supply shocks hit. Unintended consequence: AI as gatekeeper could shift profit pools to platforms/exchanges, not application-level companies.

Key Decisions for Investors

• Consider establishing a 2–4% portfolio long in NVDA via a 9–15 month bull call spread (buy 12–18 month LEAP 30–40% OTM calls, sell nearer-term 10–20% OTM calls) to target ~+40% in 6–12 months; hard stop/hedge if NVDA falls >15% from entry or implied vol collapses >25%.
• Allocate 1–2% long to MS (Morgan Stanley) equity or buy 9–12 month ITM calls as a play on sell-side quant/AI flow monetization; take profits if the position rallies >30% or if equity trading volumes fall >20% sequentially.
• Add a 1% tactical position in NDAQ (long equity or 6–12 month calls) to capture higher trading/data-fee growth; offset cost by selling 1–2 month OTM covered calls on rallies >15%.
• Buy downside protection: allocate 0.5–1.0% portfolio notional to 3–6 month OTM puts on NVDA (or put spreads) as insurance against a regulatory/export-control shock; monitor US-China export control notices and Nvidia earnings/guide in next 30–60 days as triggers to re-evaluate.