Ex-OpenAI's Kass: AI Is Going to Make a Lot of Winners

Zack Kass, former OpenAI head of go-to-market, said at the Citi Macro Conference in Hong Kong that AI investing is still in the early days but will create many winners, signaling a continued secular opportunity in AI commercialization. His comments are bullish for companies that can effectively monetize AI, but they are high-level and likely directional rather than immediately market-moving.

Analysis

The broad bullish consensus that “AI creates many winners” understates where value will actually accrete: the hardware and cloud layers that capture nonstop, recurring spend (GPUs, interconnect, memory, power) and the integrators that turn models into operational workflows. Expect a multi-year capital cycle where foundry/OSAT capacity, GPU allocation and data‑center power buildouts drive bottlenecks — these are the choke points that determine which companies can monetize AI rather than who merely claims an AI roadmap.

Key downside catalysts that can quickly reverse optimism are regulatory and operational: an EU/US safety/regulation regime, a high‑profile model safety event, or a sudden pullback in enterprise capex if macro tightens. On timeframes, headlines (earnings, new model launches) will move equities in days–weeks, but durable winners are defined by 6–24 month hardware cycles and 2–5 year enterprise adoption curves; monitor TSMC/NVIDIA capacity signals and major cloud providers’ capex cadence as leading indicators.

A common miss in market positioning is concentration risk — revenue and pricing power will cluster at a few stack layers, producing a winner‑take‑most outcome rather than broad gains across all tech names. Underappreciated second‑order beneficiaries include data‑center REITs and regional grid/power utilities, while legacy outsourcing and low‑value software vendors face margin erosion. For banks, the near‑term opportunity is fee capture from AI M&A and financing, but balance‑sheet and venture exposure make pure banking plays asymmetric unless you size them as event‑driven, option‑style exposures.

Key Decisions for Investors

• Long NVDA via a 3–6 month diagonal/call spread (buy 3–6m 5–15% OTM calls, sell nearer‑dated calls) to capture continued GPU tightness while capping premium; target asymmetric payoff >2:1 if GPUs remain supply‑constrained, max loss = premium paid.
• Pair trade (6–12 months): Long MSFT (or GOOGL) vs Short IBM — size 1:1 dollar exposure. Rationale: cloud scale + model hosting capture vs legacy services risk of margin squeeze. Target 20–30% upside on the long leg vs 15–20% downside protection from the short; stop‑loss at 12% adverse move.
• Buy Digital Realty (DLR) or Equinix (EQIX) with a 12–24 month horizon to play data‑center capacity tightness. Position as 2–4% portfolio exposure; expect 15–25% upside if colo pricing/occupancy sustains, primary risk is accelerated capex compressing FCF.
• Small, optionality‑style long on C (Citigroup): buy 6–12 month calls (size <1% NAV) to capture potential incremental fee and financing upside from AI M&A activity. Risk/reward: limited premium outlay for event-driven upside; downside limited to premium if M&A/markets remain muted.