Stravoris
← Back to Insights

AI: The Broader Picture — Energy · Water · Cooling · Capital

·AI Industry Insights

Energy · Water · Cooling · Capital

When hyperscalers commit to massive AI infrastructure buildouts, they don't operate in isolation. Here's what's actually happening: AI datacenter demand is accelerating one of the largest infrastructure transformations in modern American history.

In last week's analysis of hyperscaler earnings calls, we found that Amazon, Microsoft, Google, and Meta are guiding toward approximately $650B in combined 2026 CapEx — unprecedented levels, driven primarily by AI datacenter infrastructure.

Looking at the details, what I found is dramatic. Constellation planning a $1.6B nuclear restart, Meta securing major power partnerships, and Google acquiring Intersect Power for $4.75B — these aren't isolated transactions. They're signals of a broader shift: energy, grid, and technology investment is being pulled forward and redirected to meet AI-era computing requirements — with totals reaching into the multi-trillions through 2030.

Let's break down the ecosystem response:

*PPA: Power Purchase Agreement

For every $1 hyperscalers spend on datacenters, the supporting infrastructure ecosystem is deploying $2–3.

One hard number: regulated utilities alone are projected to invest ~$1.4T (2025–2030) in grid and system upgrades. Layer in generation buildouts, storage, and data-center-adjacent supply chains, and you're in multi-trillion territory this decade.

Below, in the Addendum, is a structured summary of what I could validate online in the limited time I had. I've grouped the research into:

  • Power Generation
  • Grid Infrastructure
  • U.S. Federal Government
  • Cooling Technology
  • Semiconductor Efficiency
  • Water Management
  • Coal Site Conversions

This article is U.S.-only.

The Addendum is a long read — feel free to skim the section headers to get a sense of scale. I've included the links I used.


THE BOTTOM LINE

When we look at initiatives like:

  • $1.4T utility investment (double the previous decade)
  • ~250+ GW of new generation in development / interconnection queues
  • ~84 GW renewable PPAs contracted
  • ~18 GW battery storage under construction
  • 4 federal sites designated for AI + energy
  • 14 states modernizing grid regulations
  • ~$15B+ nuclear partnerships signed
  • 13-state geothermal initiative launched

…we realize this isn't a coordinated AI-specific mobilization — it's something perhaps more significant: AI demand is reshaping the largest infrastructure super-cycle in modern American history. The energy transition, grid modernization, and technology roadmap are being accelerated and redirected to support sustained AI growth.


What This Means for You

If you're building an AI company: this is opportunity. The infrastructure is being built. Power is being secured. Permitting is accelerating. The physical system is already adjusting to support long-term AI demand.

If you're an AI practitioner: AI isn't going away. Utilities, governments, and industrial operators are planning around it. That means demand for serious engineering, systems thinking, and execution will be there for years. Get moving.

If you're skeptical about the AI boom: follow the infrastructure capital. Utilities don't sign long-duration power agreements on speculation. When nuclear assets are restarted and developers tie generation directly to compute campuses, that's a multi-decade signal.

The bottom line: AI isn't just growing inside software stacks. It's being absorbed into the energy grid, the semiconductor roadmap, and national infrastructure planning. That's not how bubbles behave.


Addendum: Research and Links

1. POWER GENERATION: The $1.5 Trillion Energy Response

Nuclear Renaissance - $15B+ in Deals

We're witnessing something extraordinary: nuclear plants coming back from the dead.

  • Three Mile Island Unit 1 (shut down in 2019) targeting 2028 restart (835 MW for Microsoft)
  • Palisades (Michigan), a high-profile restart effort, is targeting a return in early 2026, after slipping from prior targets
  • Meta's 6.6 GW nuclear portfolio spans three providers (Oklo, Vistra, TerraPower)
  • Constellation achieving 98.8% operating rates across 21 reactors

Links:

Natural Gas Explosion - $35B+ Annually

The numbers are staggering: US gas-fired capacity proposals tripled in 2025 vs 2024

Link: Electric generators plan more natural gas-fired capacity after few additions in 2024 - U.S. Energy Information Administration (EIA)

Homer City, Pennsylvania: Plans call for converting a shuttered coal plant into a 4.5 GW natural gas facility (the largest in America) with a $10+ billion investment. Construction starts in 2025. Power production targeted for 2027. When lead times for turbines explode to 5-7 years and you're still building, that signals confidence, not caution.

Link: Homer City and Kiewit unveil plans for 4.5GW natural gas powered AI data center in Pennsylvania - DCD

Renewable PPAs

  • Amazon, Microsoft, Meta, and Google now dominate large-scale corporate renewable PPAs in the United States, accounting for the vast majority of contracts in the United States:
  • Google's Intersect acquisition: $4.75B for integrated renewable+datacenter development

Links:

Enhanced Geothermal - The Dark Horse

Here's what's fascinating: geothermal is going mainstream.

  • Meta-Sage Geosystems: 150 MW using proprietary tech, operational 2027
  • Google-Fervo Energy: 400 MW Cape Station (Utah), the world's largest Enhanced Geothermal System
  • DOE 13-State Initiative: Expanding geothermal beyond Nevada/Utah/California

The Department of Energy estimates geothermal could provide 120 GW by 2050—that's 16% of projected datacenter demand. MIT and Google X spinoff Malta Energy are converting old coal plants into molten salt thermal storage facilities.

Links:


2. GRID INFRASTRUCTURE: The $1.4 Trillion Modernization

Utility CapEx Super-Cycle

US electric utilities are entering what industry analysts call a "super-cycle":

  • $1.4 trillion in capital expenditure (2025-2030)
  • Double the previous decade
  • $208 billion in 2025 alone for grid upgrades

Link: EEI Projects Record $208 Billion in Utility Investments for 2025 — Lucasys

Let's put that in perspective: utilities are spending more in one year (2025) than they spent in any three-year period from 2010-2020.

Transmission Approvals Accelerating

  • Southwest Power Pool: $8.6 billion including the highest-voltage lines in US history
  • Several states have enacted Grid Enhancement Technology legislation
  • ~12 states piloting regulatory "sandboxes" that cut approval times from years to 18 months

Links:

The Load Flexibility Revolution

Duke University researchers estimate the grid could integrate substantial new datacenter demand—potentially 76-126 GW—with minimal capacity expansion under managed curtailment scenarios.

For context, peak US demand in July 2024 was ~760 GW. This study shows we could add 10-16% more capacity through intelligent load management alone. This is demand flexibility as infrastructure. And it's becoming a requirement for interconnection in multiple regions.

Link: Rethinking Load Growth: Assessing the Potential for Integration of Large Flexible Loads in US Power Systems | The Nicholas Institute for Energy, Environment & Sustainability


3. FEDERAL GOVERNMENT: The $50B+ Strategic Response

DOE Federal Land Program

The Department of Energy has designated 4 federal sites for AI datacenter and co-located power development:

  • Idaho National Laboratory
  • Paducah Gaseous Diffusion Plant (Kentucky)
  • Two additional sites (with nuclear infrastructure history)

Proposals due: January 30, 2026. That's next month. The speed matters. DOE moved through procurement stages in 2025 in months, not years—signaling this administration views AI infrastructure as a national priority comparable to defense.

Link: DOE Announces Site Selection for AI Data Center and Energy Infrastructure Development on Federal Lands | Department of Energy

Executive Action Cascade

Recent executive actions on accelerating federal permitting changed the game:

  • FAST-41 expansion to all qualifying datacenter projects (expedited review)
  • Categorical exclusions for datacenter infrastructure (streamlined environmental review)
  • Federal lands and brownfields available for development
  • EPA directive to modify Clean Air Act and Clean Water Act regulations

Commerce Department: Loans, grants, and tax incentives for projects requiring 100+ MW, provided sponsors have committed $500M+ in capital.

Link: Accelerating Federal Permitting of Data Center Infrastructure – The White House

DOE Funding Programs

The Loan Programs Office is financing: Nuclear reactors, Enhanced geothermal systems, Grid stability investments, AI/datacenter energy infrastructure. Link: Advanced Nuclear Energy Projects Solicitation Fact Sheet

Grid Deployment Office programs include: Transmission Facilitation Program, AI for Interconnection initiative, Interagency Transmission Authorizations. Link: Grid Deployment and Transmission | Department of Energy

These are billions in available capital with streamlined approval processes.


4. COOLING TECHNOLOGY: The $10B Innovation Sprint

Liquid Cooling Market Transformation

Data Center Liquid Cooling Market is set to expand massively, from USD 870 million in 2024 to USD 10.70 billion by 2030. That's over a 1000% growth. It is estimated that by 2027 50%+ of new hyperscale capacity will be liquid-cooled.

Link: Data Center Liquid Cooling Market Outlook Report 2025-2030

The reason is AI racks can exceed 50 kW, pushing facilities toward direct-to-chip and liquid loops.

Direct-to-Chip (D2C) Breakthroughs

  • Removes 70-80% of heat loads directly at the processor
  • 3.5x thermal efficiency vs air cooling
  • 40% power reduction possible

Link: Direct-to-Chip Liquid Cooling: Optimizing Data Center Efficiency

HP-Nvidia partnership: Silicon-based microfluidic cooling for next-gen GPUs, with $3.25M DOE support. Target: 0.01 K/W thermal resistance. Deployment: 2026-2028.

DOE National Labs Achievement

DOE's exascale computing facilities have achieved PUE of 1.03 (Power Usage Effectiveness). The Industry average is 1.56. This proves state-of-the-art efficiency is achievable at hyperscale. The technology exists—it's now about deployment velocity.

Links:

Cold Underground Thermal Energy Storage (Cold UTES)

NREL is leading a DOE-funded project exploring geothermal underground thermal energy storage specifically for datacenter cooling:

  • Long-duration energy storage + industrial-scale cooling
  • Reduces need for power plants dedicated solely to cooling loads
  • Partners: Lawrence Berkeley Lab, Princeton, University of Chicago

Link: Reducing Data Center Peak Cooling Demand and Energy Costs With Underground Thermal Energy Storage


5. SEMICONDUCTOR EFFICIENCY: The $100B+ R&D Engine

DOE Microelectronics Initiative

DOE's "Energy Efficiency Scaling for 2 Decades" program has one audacious goal:

Improve microelectronics energy efficiency by 1000x over 20 years.

Currently efficiency is doubling every 2.5-3 years. Energy efficiency per compute operation has improved by orders of magnitude over the past two decades. This matters because training a single large AI model can consume 50 GWh—enough to power San Francisco for three days. Chip efficiency directly impacts infrastructure requirements.

Link: Department of Energy Announces Pledges from 21 Organizations to Increase the Energy Efficiency of Semiconductors | Department of Energy

Advanced Packaging & Manufacturing

  • GPU thermal design power: 400W (2020) → 1,200W (2024)
  • Next-gen GPUs targeting improved power distribution, thermal interfaces, energy recovery
  • Memory chip production ramping (Samsung, SK Hynix)
  • Domestic fab expansion under CHIPS Act

The semiconductor efficiency race is an infrastructure multiplier: every 10% improvement in chip efficiency reduces datacenter power requirements by gigawatts.

Link: Thermal Management for AI Chips | Advanced Thermal Solutions


6. WATER MANAGEMENT: The Sustainability Imperative

The Challenge

66% of new US datacenters since 2022 were built in high water-stress areas. Academic research projects AI adoption could result in 4.2-6.6 billion cubic meters of additional water withdrawal by 2027. Link: How AI Demand Is Draining Local Water Supplies

One planned 1.2 GW natural gas plant in Texas is projected to consume 1.32 billion gallons annually—10x conventional datacenters. Link: Entergy proposes gas-fired power plants in Southeast Texas totaling 1.2 GW | Utility Dive

The Response

Closed-Loop Systems: Microsoft's Quincy Water Reuse Utility partnership, Microsoft-Schneider Electric non-water refrigerant designs, Saving millions of gallons vs evaporative cooling

Reclaimed Water Co-location: Datacenters co-locating with wastewater treatment facilities. Link: Data Centers and Water | Norton Rose Fulbright - July 2025

Advanced Technologies: AirJoule: Extracts water from waste heat using metal-organic frameworks. Thermal energy recovery systems. Link: How AirJoule plans to extract water from data center waste heat - DCD

Corporate Commitments: AWS, Google, Microsoft: "Water positive" strategies. 75% renewable-matched by 2025, 100% by 2030


7. COAL SITE CONVERSIONS: The $20B+ Repurposing Wave

America's shuttered coal plants are becoming AI infrastructure hubs. Coal sites have four critical assets:

  1. Existing grid interconnections (skip the 2-year queue)
  2. Water access
  3. Available land
  4. Local workforce

Major Conversions

Homer City, Pennsylvania: Former capacity: 2 GW coal. New capacity: 4.5 GW natural gas (largest in US). Investment: $10+ billion. Timeline: Construction 2025, operations 2027. Link: Former Homer City Plant to Become a $10 Billion AI Data Center Campus

Babcock & Wilcox - Denham Capital Partnership: Multi-site coal-to-gas conversion program across US and Europe. Leveraging B&W's conversion expertise with Denham's capital. Link: Babcock & Wilcox and Denham Capital Announce Strategic Partnership to Convert Existing Coal Plants to Power Data Centers

DOE Pacific Northwest National Lab: Leading "Coal-to-X" redevelopment campaign. Fact sheets, resources, technical assistance for developers. Link: Could old coal sites be recycled as new data center campuses?

Alternative Strategy - Thermal Storage: MIT research: Converting coal plants to molten salt storage facilities. 100-500+ MW per site. Preserves infrastructure, grid connections, jobs. Link: Could Aging Coal Plants Be Transformed into Renewable Data Center Energy Storage?

When you're converting infrastructure instead of building greenfield, you're demonstrating capital efficiency and speed-to-market urgency.