The 6th Annual Global Technology Report, published by Bain & Company on September 23, 2025, projects that funding the computing power required to meet expected AI demand by 2030 would require about $2 trillion in annual revenue, and estimates that the world falls roughly $800 billion short of that figure each year even after AI-driven savings are reinvested. The report frames the $2 trillion target as achievable only if capital flows, efficiency gains, grid buildout, and demand realization all hold.
Summary of findings
Bain estimates that meeting projected AI demand by 2030 requires roughly 200 GW of incremental AI compute globally, with the United States accounting for about half of that incremental power. Building and operating that compute implies around $500 billion in annual capital expenditure, which Bain calculates would need to be justified by about $2 trillion in annual revenue. Even after accounting for AI-driven cost savings reinvested into capex, the report estimates an annual shortfall of about $800 billion against that revenue target.
The report attributes the gap to AI compute demand outpacing semiconductor efficiency gains (Moore's Law). Bain calculates that even if US companies shifted all on-premise IT budgets to cloud and reinvested AI efficiency savings into capex, the total would still fall short of the required amount.
Bain identifies three binding constraints on the buildout: the supply chain (chips, CoWoS packaging, and HBM), geopolitical competition for semiconductor dominance, and grid infrastructure that has not added capacity for decades.
Key projections (by 2030)
| Metric | Value |
|---|---|
| Global incremental AI compute requirement | 200 GW |
| US share of incremental power | ~50% |
| Annual capex required | ~$500B |
| Annual revenue required to justify capex | ~$2T |
| Annual shortfall after AI savings reinvested | ~$800B |
Reception and context
The $2 trillion and $800 billion figures became among the most-cited data points in the 2025–26 AI-bubble debate. They sit within a consistent cross-source picture alongside NERC 2025 Long-Term Reliability Assessment (LTRA) (224 GW grid stress), Epoch AI — How Much Power Will Frontier AI Training Demand in 2030? (4–16 GW per site), RAND — AI's Power Requirements Under Exponential Growth (2025), and Morgan Stanley's $3 trillion cumulative spend forecast.
The report is not itself a "bubble" call. Bain frames the $2 trillion as achievable if the right conditions hold — capital flows, efficiency gains, grid buildout, and demand realization. The framing aligns in order of magnitude with the "gigawatt-per-week" vision in Abundant Intelligence (Altman). It stands in tension with MIT NANDA — The GenAI Divide (State of AI in Business 2025), which reports that 95% of enterprise pilots yield no P&L impact, raising the question of where $2 trillion in annual revenue would originate.
Relationships
- supports: Abundant Intelligence (Altman) (same order of magnitude as the "gigawatt-per-week" vision).
- contradicts: MIT NANDA — The GenAI Divide (State of AI in Business 2025) — if 95% of enterprise pilots yield no P&L impact, where does $2T in revenue come from?
- depends-on: Semiconductor Supply Chain, Data Center Siting / AI Power Politics, AI Environmental Impact.
- related: Planned AI Bubble Debate, Circular Financing in AI.
Sources
- Primary: Bain Global Technology Report 2025