2026人工智能经济现状报告_66页_7mb

Summary of The State of the AI Economy (June 25, 2026)

Core Content

This report provides an in-depth analysis of the AI economy, focusing on real customer demand, economic impact, capital expenditures (CapEx), token dynamics, and the value stack in AI. It aims to offer a factual understanding of the AI sector's growth and challenges, free from hype and fear.

Main Points

1. Real, Big, and Fast Demand

• Real Revenue: AI demand is driven by real external customers, with the sector growing 3x faster than any prior IT wave.
• Deduplicated Revenue: The generative AI (GenAI) ecosystem revenue has reached $175 billion annually after deduplication.
• Compute Supercycle: AI demand has triggered a significant increase in compute usage, with a 10x growth in compute demand and new energy generation.
• Rapid Growth: The time to reach an additional $1 billion in cumulative revenue has decreased from 180 days in 2023 to less than 2 days in 2026.
• Backlog: Hyperscalers face contract backlogs, indicating strong demand that outpaces supply.

2. Economic Impact: Big is Still Small

• Revenue as a Rounding Error: AI revenue is still a small fraction of the overall economy, equivalent to 0.42% of US GDP.
• Consumer and Producer Surplus: AI benefits consumers through free services and producers through efficiency and revenue growth, though these gains are not fully captured in GDP.
• Early Adoption: Early adopters are outperforming peers, as seen in historical cases like automation and digital goods.
• AI Impact on Earnings: AI is increasingly mentioned on earnings calls, with 50–60% of claims now quantified, though the impact on the bottom line remains uncertain.

3. CapEx: The Largest Buildout in Tech History

• Cumulative CapEx: Hyperscalers and neoclouds have committed to $2 trillion in CapEx by 2026.
• External Financing: A growing share of AI infrastructure is being funded externally, indicating a shift in financing models.
• Depreciation Charges: The 2026E depreciation charge is expected to reach $111 billion, with current revenues covering about 19–32% of quarterly depreciation.
• Rental Rates: H100 GPU rental prices suggest that demand is absorbing existing supply, though overbuilding is a risk if depreciation is not offset by revenue growth.

4. Tokens: The Unit of Value?

• Token Growth: Token volumes have grown 14x annually, driven by agentic workloads and elastic demand.
• Elasticity: Token demand is highly elastic; a 10% price cut leads to 12–18% more usage.
• Token Pricing: Token-based pricing is likened to the 'pay-per-click' moment for AI, enabling better attribution to specific projects.
• Quality-Adjusted Tokens: These are the closest to a usable unit of value, as they reflect both volume and quality of output.

5. The Stack: Where Value is Captured

• Value Distribution: Revenue is concentrated in the hosting and foundation model layers, with apps and models gaining share.
• Stack Dynamics: The stack transforms capital and energy into cognition, with:
  • Chips: Convert energy to tokens.
  • Hosting: Convert chip CapEx to token OpEx.
  • Foundation Models: Convert tokens to intelligence.
  • Apps: Convert intelligence into consumer value.
• Competitive Pressure: Pricing power is influenced more by competitive pressure than stack position. For example, Nvidia holds the largest share, but AWS and Google may reduce its prominence through vertical integration.

Key Insights

• The AI economy is growing rapidly, with realized revenue surpassing $175 billion annually.
• CapEx is substantial, with a focus on compute and energy infrastructure, but revenues have yet to fully cover the depreciation costs.
• Token-based pricing is becoming the dominant model, and token demand is elastic, leading to increased usage as prices fall.
• Value is shifting up the stack, from chips to models and apps, as the industry matures.
• GDP does not fully capture AI's economic value, especially consumer benefits and producer surpluses that arise from efficiency and new capabilities.

Conclusion

The AI economy is at a pivotal stage, marked by real demand, significant CapEx investment, and token-based monetization. While the sector is still early, it is already exerting a substantial influence on the global economy, with compute and energy playing central roles. The value stack is evolving, and the ability to monetize depends on technical advancements and competitive dynamics.