AI economy hits $175B run rate as power becomes the real constraint

AI compute growth accelerates to 80% CAGR, breaking 50-year infrastructure trend

Exponential View's State of the AI Economy report reveals global compute stock growth has jumped from a historical 66% CAGR to 80% starting in 2020 as AI accelerators scale—marking the first major deviation from a five-decade trend since the PC era. This signals sustained infrastructure investment and potential multi-year acceleration before reverting to historical means, with direct implications for chip demand, data center buildout, and the economics of AI model training.

OpenAI Codex usage reveals agentic AI already embedded in production workflows at scale

Analysis of Codex usage shows 29% of users managing five or more concurrent agents weekly, with top 1% percentile users running ~71 hours of agent turns per day—concrete evidence that the shift from chatbots to agentic delegation is already embedded in production workflows. Nearly 100% of OpenAI employees now use Codex weekly, and product leadership argues that AI has fundamentally inverted the traditional development process, with taste and judgment becoming the most valuable professional skills as building becomes cheaper. Together, these signals validate the move toward AI-native organizational structures with fewer human managers and suggest that agentic workflows are no longer theoretical but operational at scale.

The State of the AI Economy

Exponential View published a bottom-up analysis of global AI economy revenue and infrastructure economics. Main takeaways:

• Generative AI ecosystem generated $110B in deduplicated revenue over past 12 months; annualized run rate now $175B, growing ~200% YoY—3x faster than Internet or mobile waves
• Q1 2026 AI revenues ($25B) now exceed depreciation on data center/chip investments ($21B), suggesting infrastructure buildout is approaching breakeven on revenue basis
• Token price elasticity shows 10% price cuts drive 12-18% increase in token consumption, expanding total spend despite lower unit costs
• Methodology uses item-by-item P&L reconstruction across hyperscalers and AI companies with confidence-weighted modeling to avoid double-counting supply chain flows
• Power infrastructure constraints identified as critical limiting factor: fixed demand charges and interconnection queue delays (e.g., xAI's Colossus self-supplied 1.2GW due to grid queue delays) matter more than marginal electricity pricing

Why Humans Are AI's Biggest Bottleneck: Alexander Embiricos on Codex's 20x Growth and the Future of AI-Native Product Development

Lenny's Newsletter published an interview with Alexander Embiricos, who leads product on OpenAI's Codex coding agent. Main takeaways:

• Codex has grown 20x since August 2025 and now serves trillions of tokens weekly, with OpenAI building the Sora Android app in 18 days using Codex as the primary development tool
• The bottleneck to AGI-level productivity has shifted from model capability to human review speed; humans reviewing AI-generated code and work is now the constraint, not AI generation speed
• Coding is becoming a core competency for all AI agents because writing code is the most effective way for agents to interact with and control computer systems
• Product strategy centers on AI as a proactive teammate across the full development lifecycle, not a reactive tool; this requires rethinking how humans and AI collaborate on software engineering
• The shift from building to reviewing represents a fundamental change in software engineering workflows and productivity measurement