JIL’s Take on AI: Big AI Companies Have Become Power Companies

By John Ian Lau, Contributing Editor

In the relentless march of artificial intelligence, we’ve witnessed a profound transformation: the titans of tech—once synonymous with software and silicon—are now inextricably linked to the energy sector. Big AI companies like Microsoft, Google, Amazon, and Meta have effectively become power companies, voraciously consuming electricity to fuel their data-hungry models. This shift isn’t just metaphorical; it’s a stark reality reshaping infrastructure, policy, and investment landscapes. As we enter 2026, recent moves by the Trump administration underscore this evolution, while innovative (and audacious) ideas from figures like Elon Musk point to even wilder frontiers.

Auctioning Power from PJM

In a policy push that’s equal parts pragmatic and provocative, the Trump administration, alongside governors from Northeastern states, is urging PJM Interconnection—the largest U.S. grid operator serving 65 million people across 13 states and D.C.—to hold an “emergency” wholesale electricity auction. Announced in mid-January 2026, this one-time auction aims to secure at least $15 billion in 15-year contracts for new power generation, specifically to meet the surging demands of AI-driven data centers. Tech giants would bid on these contracts, funding the construction of new plants (potentially gas, nuclear, or renewables) even if they don’t ultimately use all the power—essentially fronting the bill for grid reliability amid skyrocketing electricity prices and blackout risks.

This isn’t mere rhetoric; it’s a direct response to the AI boom. PJM’s plan, unveiled just days ago, targets the explosive growth in data center demand, which has strained the grid to its limits. The auction, slated for completion by September 2026, could force companies to commit billions upfront, addressing immediate shortages while signaling a new era where Big Tech subsidizes national energy infrastructure.

For these companies, the implications are multifaceted—and costly. First, it accelerates their pivot from pure tech to energy barons. Microsoft, for instance, has poured billions into nuclear deals, including a 2025 pact to restart the Three Mile Island plant and investments in small modular reactors (SMRs) to power its Azure data centers.

Google is similarly aggressive, committing to carbon-free energy via offshore wind and geothermal, while partnering with utilities for dedicated power lines. Amazon’s AWS has inked massive renewable deals, including a $10 billion solar + storage project in 2025, but is also eyeing gas peakers for reliability. Meta, meanwhile, is building hyperscale facilities with on-site energy storage, and even OpenAI is part of the $100 billion “Stargate” venture with Microsoft to create AI-specific infrastructure.

In the big AI race, these strategies are about securing a competitive edge. NVIDIA, the chip kingpin, is indirectly fueling this through its energy-hungry GPUs, prompting a $100 billion fund with BlackRock and Microsoft for AI data centers and power grids. Collectively, Alphabet, Amazon, Microsoft, and Meta are projected to spend over $350 billion on data centers in 2025 alone, rising to $400 billion in 2026. This “all of the above” approach—mixing renewables, gas, and nuclear—reflects a scramble to mitigate blackouts and regulatory hurdles. But Trump’s auction adds pressure: It could lock them into long-term costs, potentially inflating operational expenses by 10-20% if overcapacity results.

The Energy Crunch: Build Times, Upgrades, and Consumption

Data centers are at the heart of this power play, but their lifecycle reveals inefficiencies that Trump’s policy aims to exploit. Construction typically spans 2-5 years, with costs ballooning from $7.7 million per megawatt (MW) in 2020 to $10.7 million in 2025—a 7% compound annual growth rate (CAGR). Once operational, their “utility” phase—before major upgrades—is often short-lived. AI hardware, like GPUs, refreshes every 18-36 months due to rapid advancements, rendering facilities obsolete faster than traditional infrastructure. Global construction hit record highs in 2025, with over $74 billion in North American pipelines alone.

Energy demands exacerbate this. U.S. data centers consumed 4% of total electricity in 2024, projected to double by 2030. Globally, they account for 1.5% of electricity use (415 terawatt-hours annually), with AI driving 5-15% of that—potentially rising to 35-50% by 2030. A single AI data center can guzzle as much power as 100,000 households, with hyperscale ones demanding up to 20 times more—pushing total U.S. data center share to 8.6% by 2035. When including AI delivery to users, this could hit 21% globally by 2030.

Elon Musk’s Cosmic Solution: Data Centers in Orbit?

Enter Elon Musk, who has long argued for off-world solutions. In November 2025, he stated: “In not more than 5 years, the best AI compute will be in space, because we have solar power there, 24/7.” Musk envisions adapting Starlink satellites for orbital data centers, powered by unlimited solar energy and cooled by the vacuum of space. He even floated lunar bases producing “solar-powered AI satellites” at scale, potentially generating 100 terawatts per year.

Musk’s track record with SpaceX—launching thousands of Starlinks—lends credibility, and concepts like orbital computing aren’t new (e.g., Amazon’s Jeff Bezos has echoed similar ideas). Possibilities are tantalizing: Near-zero energy costs, infinite scalability without land constraints. However, challenges abound—latency for real-time AI (though fine for batch processing), radiation hardening, and launch economics (Starship could drop costs to $190k per EFLOPS cluster). It’s futuristic, but if anyone can pull it off, it’s Musk. Still, Earth-bound grids will dominate for years.

A Smart Policy

Trump’s auction is an amazing policy stroke. It makes eminent sense: Data centers’ multi-year build times clash with AI’s rapid upgrade cycles, leading to stranded assets and grid strain. By compelling Big Tech to fund new capacity, it ensures reliability, creates jobs in energy construction, and counters inflation from rising utility bills. It also aligns with national security, preventing foreign dependencies on power for critical AI tech.

Yet, at its core, this is a stealth tax on these companies. Dressed as an auction, it shifts billions from Silicon Valley coffers to infrastructure without raising corporate rates outright. It’s clever politics—Big Tech foots the bill for the AI revolution they ignited, while taxpayers cheer. As investors, watch for ripple effects: Higher costs could pressure margins, but resilient players like Microsoft (with its energy hedges) stand to gain. In the end, AI’s power hunger is rewriting the rules—on Earth and, perhaps soon, beyond.