Jensen Huang floated AI tokens as part of engineers' compensation. Sam Altman says they could one day be a form of universal basic income.

More than tripled: Chamath Palihapitiya said his firm's AI costs have 'more than tripled' since November 2025 and are now running into the millions, underscoring rapid, material cloud/inference cost pressure. Nvidia CEO Jensen Huang proposed bundling AI 'tokens' into engineer compensation at GTC 2026, arguing tokens raise productivity, while OpenAI's Sam Altman pitched tokens as a tradable 'universal basic compute' and core unit of AI monetization. Microsoft’s Satya Nadella warned that token generation must deliver societal value or risk losing social license, highlighting execution and infrastructure scaling risks for large AI data-center investments.

Analysis

Treat compute-as-a-tradable-asset as a structural re‑rating event for upstream hardware and custody providers, not end-user software. If compute becomes fungible and transferable, pricing power will migrate toward scarce physical inputs (leading-edge GPUs, custom ASICs, data‑center real estate) and to firms that provide liquidity, custody, and settlement; expect concentrated revenue capture among a handful of hardware incumbents and large asset managers over 12–36 months.

A predictable second‑order effect is demand volatility that outpaces capacity scaling: ephemeral spikes in token demand (product launches, model fine‑tuning cycles) will drive lumpy GPU procurement and quarterly margin swings for cloud providers; conversely, efficiency gains (quantization, sparsification, on‑device inference) are the primary deflationary threat and could compress token value within 18–30 months. Regulatory and energy constraints create asymmetric tail risks — a punitive tax or utility rationing event would crater token liquidity quickly, while gradual improvements in model efficiency would erode upside more slowly.

For positioning, think of three vector plays — hardware scarcity (long), financialization/custody (long), and mid‑cycle margin compression in large cloud operators (hedge/short). Size exposures as event‑driven: concentrated, time‑bounded option positions to capture convexity while limiting carry and regulatory execution risk.

Key Decisions for Investors

• Long NVDA via 9–12 month call‑spread (e.g., buy calls / sell higher strike calls) sized 2–4% NAV — Rationale: direct play on persistent GPU scarcity and higher ASP capture if compute tokens tighten; reward asymmetric if data‑center restocking accelerates, capped premium if market softens.
• Buy BLK 12–24 month exposure (outright or buy LEAP calls 18–24 months) sized 1–2% NAV — Rationale: optionality on new custody/ETF products that would monetize tradable compute; risk: regulatory pushback or slow product uptake; reward: recurring fee revenue if market grows.
• Hedge/short MSFT via 6–12 month put‑spread or pair trade (long NVDA call‑spread / short MSFT put‑spread) sized to cap portfolio delta — Rationale: protects against cloud margin compression and customer migration to alternative procurement models; risk: MSFT’s diversified revenue can mute short, so keep position size small and time‑boxed.