Intensifying global competition and 'personal agents': What to expect from artificial intelligence in 2026

Goldman Sachs CIO Marco Argenti warns that 2026 will be a step-change for AI as models evolve into goal-driven 'agents' and de facto operating systems that can ingest far larger context, browse the web, access files and execute multistep tasks; he predicts growth of 'agent-as-a-service,' large strategic partnerships creating a winner-takes-most industry dynamic, and an intensifying U.S.–China AI race. Argenti also cautions enterprises will face “token sticker shock” as production-scale deployments drive massive token consumption, forcing firms to prioritize high-value use cases and token optimization, while energy capacity—rather than funding—becomes the key scalability constraint. The net effect for investors and corporates: higher operating and infrastructure capex, renewed emphasis on strategic alliances and reskilling workers to adapt to agent-driven workflows, and concentrated competitive outcomes among large platform players.

Analysis

Goldman Sachs CIO Marco Argenti frames 2026 as a step-change for AI, arguing 2025 was a turning point and predicting models will act as goal-driven “agents” and effectively become operating systems that can ingest far larger context, browse the web, access files and execute multistep tasks. He expects models to reason across long-running conversations and document libraries, enabling users to set objectives while agents carry out the steps to achieve them.

Argenti anticipates a consolidation dynamic driven by scale: large strategic partnerships and “agent-as-a-service” offerings will create a winner-takes-most market, and the U.S.–China race will intensify with the capability gap narrowing. He also highlights workforce implications—firms will prize adaptability and reskilling as job effectiveness shifts toward employees who can work with agents rather than replicate old habits.

On economics and operations, Argenti warns of “token sticker shock” as production deployments massively increase token consumption and places token optimization at the center of enterprise AI strategy, while energy constraints—not funding—will become the principal scalability bottleneck. Market signals attached to the piece are moderately positive (sentiment_score 0.45) with a modest market impact estimate (0.55), reinforcing that investors should price both growth opportunities and rising infrastructure and energy capex into AI plays.

Key Decisions for Investors

• Increase exposure to scalable infrastructure providers—hyperscale cloud operators, data-center REITs, chipmakers and energy/infrastructure firms likely to benefit from higher compute and power capex
• Favor software and service vendors that emphasize token-efficiency, on‑prem/off‑cloud optimizations or offer agent-as-a-service products, and underweight companies with no clear token‑optimization roadmap
• Monitor corporate disclosures for token consumption metrics, AI production costs and incremental energy capex as early signals of margin pressure or operational risk
• Prioritize investments in firms forming deep strategic partnerships and large networks, as Argenti predicts a winner-takes-most dynamic that will concentrate returns among scale leaders
• Track workforce transition indicators—reskilling programs, adoption of agent tools and operating-model changes—and consider hedges for incumbents slow to adapt; also monitor U.S.–China AI policy and competition risk