Entergy Builds Dedicated Gas Fleet To Power Meta’s “Largest Data Center” In Louisiana
Entergy Louisiana has started construction on two natural gas power plants, with a third planned, to supply electricity for Meta’s planned $27 billion data center complex in Richland Parish and at the Waterford site. The three plants will provide about 2,200 MW of capacity, more than double the power needs of New Orleans, with expedited regulatory approval and a go-live target around 2028–2029.
Analysis:
This is what AI at scale actually looks like on the ground. Not GPUs on slides. Concrete, turbines, and gas contracts.
A single hyperscale data center getting its own 2.2 GW power stack is a clear signal: compute is now gated by electrons, not chips. You can buy H100s if you wait. You cannot fake 2+ gigawatts of reliable capacity in a tight grid.
Several key implications for AI infrastructure:
AI data centers are now driving bespoke generation
Entergy is effectively building a private power backbone for Meta that just happens to be grid connected. This is not “data center loads on existing capacity.” It is generation built to spec for a single anchor tenant. Expect more utilities to treat hyperscalers as quasi-industrial customers, with dedicated plants, firm contracts, and custom regulatory paths.
Fossil gas as the bridge fuel for AI
Despite all the “green AI” talk, when someone needs a couple of gigawatts by 2028, they are not waiting on transmission upgrades and utility-scale solar plus storage. They are building gas. That tells you how CIOs and site selection teams are actually de-risking AI capacity: reliability first, emissions second, PR later.
Regulatory fast lanes for AI power
The project got expedited approval in under a year, in a sector where these usually take years. Regulators and politicians are clearly treating AI data center projects as strategic economic infrastructure. Expect a split across states: those that accelerate permits for AI and power, and those that drag, effectively pricing themselves out of hyperscale and sovereign AI hosting.
Location strategy: cheap land, weak local demand, strong utility
Richland Parish is a sparsely populated farming area. The data center and power plants will dwarf local load. This is the new pattern: push AI campuses into rural or exurban zones with room for 1+ GW, minimal NIMBY resistance, and a utility willing to build dedicated assets, then backhaul traffic over fiber. Local communities get jobs and tax base, but also become single-customer company towns in power terms.
Grid integration vs isolation
The plants will connect to the broader grid but are sized around the data center. In practice, that means Meta gets highly prioritized, firm capacity. In a stressed grid scenario, other customers may effectively compete with a hyperscaler whose anchor contracts underwrite the asset. That power asymmetry will become a political issue in more regions as AI loads grow.
Enterprise takeaway: power is a first-class architecture constraint
For large enterprises trying to do meaningful AI on-prem or with neoclouds, this is the bar. If a single site needs “more than New Orleans” power, then your 5 MW or 20 MW AI cluster must be planned like industrial infrastructure. That means:
- Early utility engagement
- Long-term power purchase or capacity agreements
- Clear stance on fossil vs renewable vs nuclear-backed supply
- Understanding that lead time for power can exceed GPU lead time
The Big Picture:
This slot into several macro trends.
AI data center construction surge
We are now in the gigawatt era of data center design. A single hyperscale AI campus can justify its own generation fleet. That is a step change from the older model of riding spare capacity in commercial or colocation sites. Expect more “power-first” campuses where the generator and substation decisions drive everything else, including where GPUs live and how big clusters can be.
Energy constraints as the new hard limit
Meta’s move is an admission that public grids are not ready to carry future AI loads without custom builds. Everyone in the AI hardware arms race is discovering the same thing: GPUs arrive in months, substations and plants arrive in years. That will:
- Extend project timelines
- Create regional “AI power hubs” and “AI deserts”
- Make power contracts as strategic as chip supply contracts
Vendor ecosystem dynamics
Utilities like Entergy are becoming critical players in the AI value chain. They are now de facto infrastructure partners to hyperscalers, alongside NVIDIA, Broadcom, Dell, etc. Over time you should expect:
- More joint announcements: “Cloud X + Utility Y” in state Z
- Utilities spinning up dedicated “data center / AI” business units
- Neoclouds and colos needing to partner deeply with local utilities to compete with hyperscalers’ bespoke builds
NIMBY vs YIMBY for AI
Rural Louisiana appears to be in YIMBY mode for data centers and power plants, helped by jobs and tax revenue. But 2.2 GW of mostly gas for social media / AI workloads will trigger pushback elsewhere. Environmental groups and consumer advocates will question:
- Who pays for these plants and grid upgrades long term
- Who bears the pollution and resource risk
- Whether local ratepayers are subsidizing hyperscalers
This will split regions into:
Pro-AI YIMBY states fast-tracking gas, nuclear, and transmission for data centers
Slower, more restrictive states that implicitly push AI builds to “friendlier” grids
Sovereign AI and neocloud positioning
If you are a country or large enterprise thinking about “sovereign AI” or cloud repatriation, this is a warning shot. Owning your data and compute is not just about GPU clusters and private racks. It is about:
- Long-term access to 50–500+ MW of reliable power
- Willing regulators
- Utilities willing to sign tailored contracts and build fast
Neoclouds that can secure power in the right regions become very interesting alternatives to public cloud AI services. The hyperscalers are showing the pattern. Smaller players will copy it at lower scale, tying up 50–200 MW pockets near flexible utilities.
Signal Strength: High
