Interview, Fireside Chat, Conference Presentation
AI Exchanges: Power Problems?
- Hyperscalers' combined capital and R&D budgets for 2026 and 2027 are forecast to rise by over $300 billion, sustaining capital allocation at approximately 87% of operating cash flow plus R&D, a figure below the 120% peak of the shale revolution.
- Demand for AI tokens and compute is projected to exceed expectations by matching productivity gains with continued consumption, while agentic machine-to-machine traffic is predicted to eventually surpass human AI traffic, creating tight supply and demand conditions.
- Global data center power demand is assumed to grow by 220% by 2030 versus 2023, with inference activity becoming more energy-intensive per server, deviating from prior assumptions of lower energy requirements.
- The U.S. power sector requires an estimated 500,000 new jobs, including 300,000 for generation and 200,000 for transmission and distribution, necessitating a 20,000 to 25,000 increase in energy apprentices for skilled labor.
- Grid infrastructure is not expected to materialize for three to five years, driving near-term deployment of behind-the-meter simple cycle natural gas generators before combined cycle options come online around 2029 into the 2030s.
- The energy mix for data centers through 2030 is projected to comprise roughly 60% thermal sources (primarily natural gas), 40% renewable sources, and additional nuclear contributions, whereas the 2030s may shift significantly with the availability of new nuclear and advanced gas technologies.
- Nuclear power is expected to remain limited to demothballing projects until the 2030s or mid-2030s due to parts availability and lead times, after which it could play a major role alongside solar and battery storage expected to show surprising efficacy.
- Hyperscalers are anticipated to ring-fence power projects via take-or-pay commitments and absorb green energy costs estimated at $40 per megawatt hour above natural gas, resulting in only a 2.5% impact on 2030 EBITDA and less than a one percentage point impact on return metrics.
- Utilities face hesitation to lead large-scale nuclear investments due to technology risks and a lack of recent U.S. experience, prompting reliance on capital markets while hyperscalers maintain financial flexibility with minimal net debt to EBITDA.
- Political pressure and NIMBY concerns are expected to drive hyperscalers to internalize power costs to shield consumer prices, even as the share of energy and electricity costs relative to total construction expenses remains relatively low.