Bloomberg Masters in Business: Songyee Yoon (Podcast)

Apr 04, 2026: Barry Ritholtz interviewed Songyee Yoon, founder and managing partner of Principal Venture Partners, on Bloomberg's Masters in Business. The discussion focused on the firm’s strategy of investing in AI-native companies, how to distinguish startups that are truly native to AI versus those chasing the AI boom, and broader themes in the tech investment landscape and technological innovation.

Analysis

Distinguishing “AI-native” from “AI-adjacent” changes where value accrues: native firms own the dataset → model → inference stack and therefore capture recurring, usage-based revenue tied to inference throughput, not one-time licenses. That shifts margin pools away from legacy software sellers toward compute and data-infrastructure providers; expect concentration of economics in GPU/ASIC suppliers and data orchestration platforms over the next 12–36 months.

Second-order supply-chain winners are capital-intensive equipment and substrate vendors (foundries, EUV tools) because a small decline in model efficiency raises demand for more raw FLOPs. Conversely, large enterprise software vendors that rely on maintenance or seat-based pricing face downward pressure as customers prefer per-inference or per-outcome billing — a structural margin compression risk over 2–4 quarters for exposed names.

Tail risks: a rapid open-source model that reduces inference cost by 30%+ or a regulatory clamp on data use could collapse current revenue assumptions within months, not years. Key catalysts to watch are (1) cloud GPU utilization and spot instance pricing (real-time signal for demand), (2) next-gen inference silicon rollouts (TSMC/ASML cycle timing, 6–18 months), and (3) material ACV expansion reported by large SaaS customers (quarterly cadence) that validates monetization.

Valuation framing: the market is pricing optionality of future monopoly-like moats into many public names; prefer capital-efficient, convex exposures (long-dated calls, structured spreads) into infrastructure and cloud incumbents while shorting narrative-driven, negative-cash-flow public names. Maintain strict event-based exits — take-profits on hardware winners after 25–40% outperformance and re-evaluate shorts on every earnings print that fails to show AI-driven ARR/ACV expansion.

Key Decisions for Investors

• Long NVDA via 9–12 month call spread (buy Jan 2027 calls, sell a higher strike) sized ~2–3% of fund. Rationale: captures continued GPU demand with defined downside (premium paid). Risk: model efficiency gains or regulatory pressure could compress compute demand; target 2.5x payoff if NVIDIA outperforms broad market by 30–50% in 9–12 months.
• Long ASML (or equivalent EUV exposure) outright, 12–24 month horizon, size 1–2% of fund. Rationale: bottleneck in advanced lithography benefits from sustained foundry capex; payoff asymmetric if the chip cycle continues. Hedge with 3–6 month put protection sized to limit drawdown to ~8–10% of position value in a capex pullback.
• Pair trade: long MSFT (or AMZN AWS exposure) vs short a concentrated basket of small-cap, negative-cash-flow ‘AI’ stickers (or short ARKK via 6–9 month puts) sized net-neutral. Timeframe 6–12 months. Rationale: cloud providers monetize infrastructure/services while many small caps trade as narrative plays; expect cloud revenue per GPU to rise, compressing speculative valuations. Target 1.5–2x upside on the pair if cloud monetization continues; risk is broad market rotation into growth if rates fall sharply.
• Deploy $50–100m into late-stage secondaries (18–36 month hold) focused on AI-native verticals with >100% YoY revenue growth and >50% gross margins. Rationale: ability to buy at private-market discounts yields high asymmetric upside on proven go-to-market. Exit triggers: priced M&A at >6x revenue or IPO window; downside protected by revenue traction and data moats.