Meta is no longer just a technology company buying electricity from utilities; it's becoming a quasi-utility itself, financing and building the entire power generation and transmission infrastructure needed to run its massive AI operations. On March 27, Entergy Louisiana announced that Meta will fund the construction of seven new natural gas power plants for its Hyperion AI data center campus in Richland Parish, Louisiana. Combined with three plants approved in August 2025, Meta now has 10 gas-fired plants in its pipeline, delivering 7.5 gigawatts of capacity . To put that in perspective, that's enough power to supply more than 5 million homes and represents a more than 30 percent increase to Louisiana's entire grid capacity.

Why Are Hyperscalers Building Their Own Power Plants?

The answer lies in the collision between explosive AI demand and the limits of existing power infrastructure. AI training and inference require enormous amounts of continuous, reliable electricity. Traditional utilities cannot expand fast enough to meet this demand, and interconnection queues for new power connections stretch years into the future. Meta's solution is direct: if the grid cannot deliver the power needed, build the infrastructure yourself .

The 10 plants are estimated to cost nearly $11 billion. Beyond the gas plants, Meta is also funding up to 2.5 gigawatts of renewable energy capacity including battery storage, 240 miles of new transmission lines connecting South Louisiana to North Louisiana and Arkansas, battery energy storage systems, and nuclear power uprates at existing Entergy facilities . Meta CEO Mark Zuckerberg has said Hyperion would cover "a significant part of the footprint of Manhattan."

This represents a fundamental shift in how hyperscalers approach infrastructure. Rather than treating power as a commodity to be purchased on the open market, companies like Meta are treating it as a core competency. The deal is structured so Meta "pays its full cost of service," and Entergy projects the agreement will deliver more than $2 billion in customer savings over 20 years .

How Are Venture Capitalists Rethinking Business Defensibility in the AI Era?

Meta's power infrastructure play reflects a broader reorientation in how investors evaluate what makes a business defensible. Michael Bloch, a partner at early-stage venture firm Quiet Capital, has articulated a framework that explains why physical infrastructure is becoming the ultimate competitive moat .

Bloch argues that artificial intelligence has permanently reordered what makes a business defensible. AI compresses the time it takes to do things, but it does not compress the time it takes for things to happen. Building software, maintaining integrations, and embedding a product deeply into a workflow were once meaningful barriers. They are becoming engineering problems measured in hours, not months. What remains are assets that required elapsed, real-world time to accumulate .

• Living, Compounding Proprietary Data: Not static datasets that were merely expensive to collect, but information generated continuously through operations that are themselves defensible. A competitor would need years to replicate the same dataset.
• Network Effects: Drivers, restaurants, and customers reinforce each other's value in ways that cannot be cloned overnight. The cold start problem may actually get harder as AI makes it trivial to build competing products.
• Regulatory Permission: Government timelines move at the speed of politics, not technology. Bank charters, FDA approvals, and defense procurement contracts are not compressible by intelligence.
• Capital at Scale: Chip fabs cost roughly $20 billion, nuclear plants around $10 billion, and satellite constellations billions more. The ability to finance and deploy capital at that scale depends on institutional trust and relationships that take decades to establish.
• Physical Infrastructure: Factories, power plants, battery networks, and data centers represent assets that no amount of AI can manufacture, transport, and interconnect faster than the laws of physics allow.

Bloch's framework aligns with a broader shift in how venture capital is orienting. OMERS Ventures noted in its 2026 VC outlook that tools like Claude Code and Cursor have made building software "astonishingly fast," pushing investors toward opportunities in spatial AI, robotics, autonomy, energy systems, and infrastructure . The firm wrote that moats are "now harder to find (and even harder to maintain)" in pure software.

"The closer a business model is to defensible infrastructure and real-world deployment, the easier it becomes to justify large cheques," noted Venionaire Capital in its North America VC 2026 analysis.

Venionaire Capital, North America VC 2026 Analysis

Lightspeed Venture Partners, which participated in Base Power's Series C funding round, cited forecasts that national power demand from AI alone will grow by over 100 gigawatts in the next five years and that Texas peak summer demand is projected to grow by 51 gigawatts, the equivalent of 51 nuclear reactors . The firm described distributed energy storage as an "essential component of the solution."

What Does This Mean for the Power Industry and Investors?

The implications are profound. Entergy stock rose 7 percent on the announcement, reaching a market cap of approximately $50 billion, up nearly 125 percent over the past two years . For gas turbine manufacturers, construction firms, and utilities with dedicated-build expertise, the demand signal is unambiguous. For investors, the key question is not whether AI needs power, but how the regulatory and political landscape will shape who pays for it and who captures the margin.

Meta initially announced a $10 billion investment in December 2024 for a 2,250-acre campus. Fortune reported in February that Meta quietly acquired an additional 1,400 acres. In October 2025, Meta entered a joint venture with Blue Owl Capital to finance, build, and operate the campus with up to $27 billion in total development costs . The expansion from 2.3 gigawatts (three plants) to 7.5 gigawatts (ten plants) in under a year reflects the speed at which AI compute demand is outstripping power availability.

The contractual terms are 15 years, which raises important questions about what happens if Meta's AI training needs evolve beyond this campus. Critics contend ratepayers could be stuck with the bill if Meta no longer requires the power after that span. Entergy argues the opposite: Meta financing the plants protects ratepayers from bearing the cost . The Louisiana Public Service Commission approved the first three plants in 2025 despite pushback from environmental groups. The seven new plants require fresh LPSC approval, and regulatory approval is not guaranteed, especially given the scale of adding 30 percent or more to a state's grid capacity for a single private customer.

What's clear is that the era of hyperscalers as passive consumers of utility power is over. Companies like Meta are now active builders of the energy infrastructure that underpins the AI revolution. This shift has profound implications for how we think about technology infrastructure, regulatory oversight, and the relationship between private companies and public utilities in the age of artificial intelligence.