As AI workloads explode, the world’s largest tech companies are no longer just grid customers—they are becoming power marketers, anchor tenants for multi-gigawatt generation, and, in some cases, de-facto utilities. For infrastructure CEOs, this isn’t a side story in Silicon Valley. It’s a structural shift in how generation, transmission and capital will be deployed over the next decade.
From social media to power marketing
In September 2025, Meta quietly crossed a line most digital companies never approach. Through a new subsidiary, Atem Energy LLC, it asked the U.S. Federal Energy Regulatory Commission (FERC) for permission to sell electricity, capacity and ancillary services at market-based rates—essentially the same authority that independent power marketers and trading desks use to operate in wholesale markets.
Atem does not own power plants today. But its parent does own something else: Hyperion, Meta’s new AI-optimised campus in Richland Parish, Louisiana, billed as its largest data centre project globally and estimated at around US$10 billion in direct investment. The facility will sit on more than 2,000 acres and is designed to draw gigawatts of power—large enough to justify its own new fleet of generation.
Louisiana regulators have already approved three new natural-gas power plants and associated grid upgrades from utility Entergy, justified in large part by the Hyperion load. Analysis by non-profit groups suggests that fuel and operating costs for those plants will largely flow through to Entergy’s wider customer base over decades, even though Meta’s initial commitments are nearer 15 years.
On paper, then, Meta is “just” adding data centre capacity. In practice, it is:
- Triggering multi-billion-dollar generation and transmission build-outs in a single U.S. state; and
- Standing up a power marketing arm that can buy and sell electricity like a utility, hedging price risk and potentially monetising surplus capacity.
For infrastructure executives, this is the new pattern: AI platforms are not only customers of grids and power plants—they’re increasingly co-architects and counterparties at system scale.
An electricity arms race among AI giants
Meta is not alone. Across the frontier AI landscape, access to reliable, large-scale power is quickly becoming the defining constraint—and the new competitive moat.
- OpenAI / Oracle / SoftBank – Stargate
The Stargate joint venture plans up to US$500 billion of AI infrastructure in the U.S. by 2029, with a target of 10 GW of data centre capacity. Recent announcements of new sites in Texas, the Midwest and other regions put the programme on track to secure the full 10-GW commitment by the end of 2025. - Microsoft – reviving nuclear for AI
In 2024, Microsoft signed its largest-ever power purchase agreement with Constellation to bring Unit 1 at Three Mile Island back into service, securing roughly 835 MW of carbon-free baseload under a 20-year PPA. The deal is backed by a US$1 billion federal loan and marks the first attempt to restart a previously shut U.S. nuclear reactor specifically to serve AI-driven data centre growth. - Google – the quiet energy trader
Google secured FERC market-based rate authority in 2010 via Google Energy LLC, enabling it to act as a wholesale power buyer and seller. Over the last decade it has become one of the world’s largest corporate clean-energy offtakers, and has used DeepMind’s control systems to cut data-centre cooling energy by around 40%, translating to a ~15% reduction in overall data-centre energy use. - xAI – turbines first, permits later
Elon Musk’s xAI has taken a more improvised route. In Memphis, Tennessee, the company has deployed around 35 gas turbines with a combined capacity of more than 400 MW to power its “Colossus” supercomputer, triggering investigations and lawsuits over operating without appropriate environmental permits. - Anthropic – reframing the scale of demand
In mid-2025, Anthropic warned that the U.S. alone will need at least 50 GW of additional electric capacity by 2028 to support frontier AI training and inference—roughly twice New York City’s peak electricity demand.
Individually these stories look like tech headlines. Taken together, they describe a structural re-rating of power as a strategic resource for AI—and therefore as a primary growth frontier for infrastructure.
What this really means for infrastructure leaders
For CEOs and C-suite executives in the infrastructure space—utility, IPP, EPC, transmission developer, fund manager—AI is not simply another large industrial customer. It is reshaping the way projects are conceived, financed and risk-allocated.
1. AI loads as anchor tenants for entire systems
Projects like Hyperion and Stargate are effectively behaving like multi-gigawatt anchor tenants, underpinned by long-dated corporate offtake and, in some cases, direct equity involvement from the tech companies themselves.
That changes the calculus for:
- Generation developers, who can now underwrite large gas, nuclear, hydro, solar and storage projects against a single concentrated offtaker rather than a diversified customer base.
- Transmission owners, who may need to build high-capacity corridors to effectively one or two hyper-scale campuses.
- Regulators and policymakers, who must balance the promise of jobs and tax revenue against long-term exposure of ratepayers if demand forecasts prove optimistic.
The opportunity is clear: AI can de-risk megaprojects that otherwise struggle to close. The risk is equally clear: systems become tightly coupled to a handful of digital tenants whose needs and business models can change faster than physical assets.
2. Project pipelines will skew towards firm capacity
AI workloads are highly sensitive to downtime and latency. That drives an outsized premium on firm, always-on power—even in regions with strong renewable resources.
- Nuclear restarts, life extensions and potential SMR (small modular reactor) deployments are suddenly back on the table, with Microsoft’s Three Mile Island deal as a proof-point.
- Gas-fired generation remains attractive as a fast-deployable, dispatchable backbone, as seen with Entergy’s Louisiana plants and xAI’s turbine farm.
That does not sideline renewables; rather, we should expect hybrid projects—gas + renewables + storage, or nuclear + renewables—structured around very high availability targets and stringent service-level agreements.
For infrastructure developers, portfolios that previously leaned towards merchant renewables may need to be rebalanced with firming assets, grid-scale storage and flexible thermal if they want to serve AI demand at scale.
3. Power trading and risk management become core capabilities
When Meta asks for market-based rate authority via Atem Energy, or when Google runs an in-house trading entity, they’re signalling a broader trend: large digital firms internalising energy trading and risk management rather than outsourcing everything to utilities.
For infrastructure companies, that has two implications:
- You may increasingly be negotiating with sophisticated in-house energy desks who understand locational marginal prices, congestion risk, capacity markets and hedging structures in detail.
- There is white space for new intermediaries—from flexibility aggregators to AI-driven power-portfolio managers—who can sit between AI firms, generators and grids.
Risk officers and CFOs in infrastructure businesses will need to treat these relationships more like structured commodity deals than traditional PPA sales.
4. Communities and ESG will be battle-fields, not footnotes
AI-driven energy demand is already colliding with local politics and ESG expectations:
- In Memphis, community groups and civil-rights organisations are challenging xAI’s turbines as a pollution burden on already vulnerable neighbourhoods.
- In Louisiana, critics argue that Entergy’s new gas plants will lock in fossil-fuel dependence for decades in order to serve a single corporate user.
The lesson for infrastructure leaders is simple: social licence is now a gating item, not a tick-box. AI projects bring global media attention and activist scrutiny. If your asset sits on the critical path to a headline data centre, expect higher levels of public engagement, political risk and, in some cases, litigation.
Boards will need robust answers to questions such as:
- How do we share economic upside with host communities, not just shareholders and AI clients?
- Can we structure contracts so that, if AI demand falls or policy tightens, we are not left with stranded fossil assets?
- What is our narrative about AI-driven power use in the context of net-zero commitments?
5. Supply chains and delivery models will be stress-tested
The AI power race also has direct implications for supply chains:
- Turbine OEMs, transformer manufacturers, HV cable suppliers, EPC contractors and specialist data-centre builders are facing clustered demand spikes tied to a handful of mega-projects.
- Permitting bottlenecks in grid infrastructure—already a global challenge—will be amplified by AI loads that require new high-voltage lines, substations and synchronous condensers, often on accelerated timelines.
This volatility rewards infrastructure platforms that can pivot quickly between technologies and regions, rather than those tied too tightly to a single jurisdiction or fuel.
Strategic questions for the boardroom
Given this landscape, CEOs and C-suite teams in the infrastructure sector should be asking themselves a few hard questions:
- Where do we want to sit in the AI–power value chain?
Are we content to be a capacity provider, or do we want to move upstream into structuring, trading, optimisation and even co-development of AI campuses? - How concentrated is our exposure to AI demand?
A portfolio with one or two very large AI offtakers can look attractive on day one—and dangerously concentrated if policy, technology or corporate strategy shifts. - Are our capabilities aligned with firm, high-availability power?
Do we have the right mix of assets (nuclear, hydro, gas, storage, grid services) and partnerships to deliver 24/7, low-latency power at scale? - Can we execute at AI speed without compromising governance?
Tech companies move in quarters, not decades. Infrastructure assets live for 30–60 years. Bridging that cultural gap without sacrificing safety, compliance and long-term value is a strategic challenge in its own right. - How are we positioning ourselves on ESG in the AI era?
Are we merely “keeping up with compliance”, or can we credibly argue that our projects enable an AI ecosystem that is cleaner, fairer and more resilient than the counterfactual?
Megawatts as the new moat
For the last decade, competitive advantage in digital markets has been framed in terms of algorithms, data and chips. Over the next decade, a different layer will matter just as much: who controls the electrons.
Meta’s Atem Energy filing, Microsoft’s nuclear revival, Google’s energy-trading arm, OpenAI’s 10-GW Stargate ambition and xAI’s turbine farm in Memphis are not disconnected curiosities. They are early signals of an AI economy in which power infrastructure is no longer a backdrop—it is centre stage.
For infrastructure companies, the choice is stark but exciting. You can treat AI as just another demanding load and respond tactically. Or you can recognise that a new class of digital utility is emerging—and position your organisation as a partner, co-architect and, in some cases, competitor in building the energy systems that will power it.
Either way, in the age of AI, your true moat may not be in miles of pipe or route-kilometres of cable, but in the gigawatts you can reliably deliver—and the sophistication with which you manage them.
