In the AI Arms Race, Bet on the Neutral Arms Dealers

Peter Lynch's wisdom holds true: during the Gold Rush, would-be miners went broke, but the people selling picks and shovels made consistent profits. Today's AI boom follows the same logic. Instead of predicting whether OpenAI, Anthropic, or Google's Gemini wins the model wars, investors can own the infrastructure that all of them must buy.

AI workloads demand exponentially more power than traditional computing. Data center power consumption is projected to surge from 55 gigawatts globally today to 122 gigawatts by 2030, according to Goldman Sachs. The investment bank forecasts a 165% increase in power demand from data centers by the end of the decade, requiring an estimated $720 billion in grid infrastructure spending through 2030. In 2023, data centers consumed 26% of Virginia's total electricity. This concentration illustrates the power bottleneck that infrastructure providers must solve.

A single generative AI prompt uses 10 to 100 times more electricity than a standard internet search. Modern AI-optimised GPU racks draw 30 to 80 kilowatts per rack, up to tenfold more than traditional server infrastructure, necessitating liquid cooling systems. Corporate responses demonstrate the severity. In September 2024, Microsoft signed a 20-year power purchase agreement with Constellation Energy to restart the 837-megawatt Three Mile Island reactor. In June 2025, Meta signed a similar 20-year deal with Constellation for 1,121 megawatts from the Clinton Clean Energy Center. These multi-decade commitments reveal how seriously hyperscalers view power availability as the critical constraint on AI expansion.

Picks and Shovels of the AI Gold Rush

The AI infrastructure buildout creates three investment categories: power generation, data center infrastructure, and semiconductor equipment. Below are some of the listed players to watchlist.

Constellation Energy (CEG) operates the United States' largest nuclear fleet with 21 reactors generating more than 19,000 megawatts of nuclear capacity. Six years ago, Constellation shut down Three Mile Island due to poor economics. Today, Microsoft guarantees premium pricing through a 20-year contract, illustrating how AI demand has transformed power economics. Following the Calpine acquisition, the combined entity will operate nearly 60 gigawatts of total capacity across all generation types.

NextEra Energy (NEE) is the world's largest renewable energy producer. The company signed power contracts totalling over 3 gigawatts in Q2 2024, including an 860-megawatt Google deal. The development portfolio includes approximately 7 gigawatts of projects, predominantly serving data center customers. Data center power purchase agreements offer 15 to 20-year terms, providing exceptional revenue visibility.

Digital Realty (DLR) reported that 30% of megawatts signed in Q4 2024 came from AI-related workloads. This suggests AI bookings could exceed 50% in quarters with stronger hyperscaler activity. The company's global platform spans key markets across North America, Europe, and Asia.

Equinix (EQIX) announced a $15 billion joint venture in September 2024 to develop xScale hyperscale data centers across the United States, targeting 1.5 gigawatts or more of new capacity by 2029. The company is deploying direct-to-chip liquid cooling infrastructure globally, including a $260 million AI-ready facility in Singapore.

Vertiv (VRT) derives approximately 80% of revenue from data center customers. The company provides power systems, cooling infrastructure, and integrated rack solutions for high-density AI deployments. The global data center power market is projected to grow from $35.14 billion in 2025 to $50.51 billion by 2030, a 7.5% compound annual growth rate.

Applied Materials (AMAT) reported record fiscal 2025 revenue of $28.37 billion, up 4% year-over-year. Advanced packaging revenue, critical for high-bandwidth memory in AI accelerators, grew 25% year-over-year in fiscal 2025. Management expects leading-edge logic and advanced packaging to approximately double in the second half of calendar 2026. Modern AI chips require 2x denser interconnects to achieve needed memory bandwidth. Applied Materials' tools enable chipmakers to stack High Bandwidth Memory directly onto GPU dies, driving equipment intensity per wafer.

Policy Catalysts

Government industrial policy is accelerating infrastructure investment. The CHIPS and Science Act allocated $52.7 billion for semiconductor manufacturing, catalysing over $300 billion in announced private sector investment. This represents a 6 to 1 leverage ratio. Every dollar of government funding triggered six dollars of private capital deployment.

Later this month, President Trump is expected to sign an executive order launching the "Genesis Mission," a comprehensive national AI initiative. Department of Energy Chief of Staff Carl Coe described it as "equivalent to the Manhattan Project or space race". The initiative will establish AI data centers on federal lands including Oak Ridge, Oak Ridge Reservation, Idaho National Laboratory, and Hanford Site.

These locations enable developers to bypass lengthy grid interconnection queues by co-locating with existing power infrastructure. This policy innovation could accelerate project timelines by 18 to 36 months compared to conventional development. If Genesis Mission funding approaches CHIPS Act scale with similar leverage effects, it could catalyze an additional $300 to $500 billion in infrastructure investment over the next five years.

Why Infrastructure Wins

Betting on infrastructure companies instead of AI model developers offers four advantages:

Revenue visibility: Power purchase agreements span 15 to 20 years with investment-grade counterparties. Data center leases run 10 to 15 years. Equipment upgrade cycles occur every 2 to 3 years. Compare this to AI software companies where competitive moats remain unproven.

Technology neutrality: Infrastructure providers win regardless of which AI architecture dominates. Constellation doesn't care if Microsoft runs GPT-7 or another model. They care that Microsoft pays for 837 megawatts for twenty years.

Capital intensity as a moat: Building nuclear reactors, fabricating advanced semiconductors, and developing hyperscale data centers requires billions in upfront capital and years of construction. These barriers protect incumbents from disruption by well-funded startups.

Diversified customer base: Vertiv sells to Amazon, Microsoft, Google, Meta, and hundreds of enterprises. Applied Materials supplies TSMC, Samsung, Intel, and every major foundry. No single hyperscaler can dominate vendor economics.

Conclusion

The AI revolution will create enormous value, but capturing it requires choosing where to take risks. Infrastructure companies participate in AI upside through volume growth and pricing power while avoiding winner-take-all dynamics inherent in AI model competition.

When the gold rush began, most miners went bust. The merchants selling shovels built enduring businesses. As AI reshapes the global economy, smart investors would place their bets on the infrastructure that makes it all possible.

To know more about how to get started with global investing and US ETFs, click here.

The article has been written by Subho Moulik, Founder & CEO, Appreciate.

Note to the Reader: This article is part of Mint's promotional consumer connect initiative and is independently created by the brand. Mint assumes no editorial responsibility for the content.