Who's most optimistic about AI — and who isn't, according to Anthropic

Anthropic surveyed ~81,000 Claude users across 159 countries: 18.8% sought "professional excellence" from AI and 32% cited productivity gains, while 22.3% named job displacement as their top worry. Users in Sub‑Saharan Africa, Latin America and Asia expressed 10–12% lower negative sentiment and higher entrepreneurial aspirations versus Western Europe/North America. Anthropic's Feb launch of Cowork and rising investment in agentic AI spurred broad selloffs in software and research stocks, while analysts warn the study skews toward active users and that AI may amplify inequalities without adequate infrastructure and safeguards.

Analysis

The next 12–24 months will be defined less by feature releases and more by who owns the transaction layer for autonomous agents: owners of large-scale compute, model-delivery pipelines, and embedded billing will capture the lion’s share of value. Expect incremental cloud/accelerator demand to grow materially (we model a 30–50% increase in high-end GPU-hours from agent workloads within 12–18 months), compressing margins for downstream professional services that can’t convert that usage into platform fees.

Optimism in lower-income geographies is a demand signal, not an immediate redistribution of profits: inexpensive agent tooling will unlock entrepreneurial activity and gig work, shifting a portion of service revenue from high-cost markets to thin-margin, high-volume markets. That creates durable upside for lightweight SaaS/marketplace platforms that can monetize transactions or take revenue share, but it also amplifies two risks — concentrated model-provider pricing power and widening data/compute infrastructure gaps — which will determine who actually retains surplus.

Capital markets will reprice winners and losers on evidence of agent adoption velocity rather than announcements. Short-term stock moves can be violent on product demos; medium-term (~6–18 months) real earnings pressure follows if firms lose billable hours or see higher churn. Regulatory, safety, or compute-supply shocks remain plausible reverse catalysts and can flip narratives within 1–6 quarters if they materially raise operating costs or constrain model deployment.

Key Decisions for Investors

• Long NVDA (12–18 months): buy-dated call spread to own core compute exposure to agent workloads. Rationale: captures >80% of incremental high-performance compute pricing power; risk: competitor silicon and a demand cooldown. Target 2:1 reward:risk if adopted agent volumes rise per model above.
• Pair trade — Long AMZN or MSFT cloud exposure / Short ACN (3–12 months): size 1.5:1 notional. Mechanism: platform owners monetize usage/marketplace fees while legacy consultancies face hour-loss. Close if consulting revenue shows sequential resilience or cloud pricing falls >15% y/y.
• Long BABA (9–12 months) via call spread or covered calls: thematic play on Chinese cloud + AI stack commercializing agent features domestically. Hedge with 6% notional in puts against regulatory/event risk. Risk/reward: asymmetric if Chinese cloud captures enterprise autopilot conversion; downside tied to regulatory/capex cycles.
• Tactical short — selective mid-cap professional services/SaaS vendors (6–12 months): target names with >50% revenue from hourly professional services and no direct platform monetization. Expect 10–30% downside if billable hours compress >10% from agent substitution; stop-loss if those firms announce platform monetization or >5% upward revision to margins.