Conference Presentation, Panel
AI Infrastructure at Scale: Financing the Power-Constrained Grid | Global Conference 2026
Milken InstituteBrian Sullivan, Lucy Heintz, Douglas Kimmelman, Karl Kuchel, Rajit Nanda, Michael Zeltkevic, Doug Kimmel, Carl Cuscelli, Lucy Hines
- Global capital expenditures for the AI energy transition are projected to approach $1 trillion, driven by hyperscalers (Amazon, Meta, Google) possessing the strongest balance sheets in history to backstop infrastructure financing.
- Long-term power purchase agreements (term contracts) are becoming the primary mechanism to de-risk infrastructure projects, allowing developers to lock in fixed returns and attract capital despite the volatility of spot markets.
- In the United States, AI accounts for only 5% of total electricity demand growth, whereas the broader global economic equation in emerging markets is driven by industrialization, electrification, and cloud migration rather than AI alone.
- Saudi Arabia is leveraging its low-cost energy mix (solar, fossil fuels) and massive connectivity infrastructure (17 operational subsea landing stations, expanding to 24) to position itself as a global "center of gravity" for AI inference serving Europe, Africa, and South Asia with sub-40ms latency.
- Data sovereignty and latency are emerging as critical constraints, forcing a shift from a purely economic calculus to a strategy prioritizing national control over data and power generation capabilities.
- Electricity price increases in the U.S. (e.g., California at ~$0.55/kWh) are driven more by transmission and distribution infrastructure costs and regulatory mandates than by the marginal cost of generation, which has declined due to lower natural gas prices.
- The "race to AI" has pivoted from "speed to market" to "speed to power," with data center locations now dictated by the immediate availability of reliable, large-scale power rather than traditional connectivity hubs.
- Hyperscalers are expected to drive 30% of future data center demand, but current revenue models remain unclear, with some projections suggesting AI could absorb 10% of the global wage bill by 2030 to justify the required ~$1.5 trillion in ad-related revenue.
- Significant consolidation is anticipated in the AI stack (energy, data centers, and compute), particularly for non-hyperscalers, as smaller players face obsolescence risks and inability to secure capital compared to diversified tech giants.
- Supply chains for wind, solar, and storage from China allow for faster deployment of power infrastructure compared to the U.S., where natural gas turbine lead times have extended to five years and costs have tripled.
- Energy storage faces a financing hurdle due to a lack of 10-year operating histories for new technologies, leading lenders to avoid unproven battery solutions despite rapid cost reductions (37% drop last year, 34% forecast for this year).
- Investors are advised to mitigate obsolescence risk by structuring contracts with optionality to sell power to diverse off-takers (e.g., utilities, Walmart, EV charging) if AI-specific demand contracts fail to materialize.
- Government intervention is shifting toward regulatory and permitting acceleration rather than direct subsidies, which are currently being withdrawn in the U.S., though international projects (e.g., Philippines, Abu Dhabi) continue to leverage government-backed power purchase agreements.
- The panel concluded that while specific technologies (e.g., cold fusion, space-based data centers) remain speculative, the current investment cycle is essential for driving the innovation and infrastructure necessary to meet the exponential growth in compute demand.