Meta’s 550 MW Ohio Gas Plants Show the Real Energy Cost of AI

Meta is not waiting for the grid to catch up with its AI ambitions. The company is financing two dedicated natural gas power plants in Ohio: a 200 MW facility in New Albany approved on June 9, 2025, and a 350 MW plant in Middleton Township approved on February 3, 2026. Together they add 550 MW of off-grid generation capacity built exclusively to power Meta's data centers. Ohio legislation passed around 2025 made it possible to approve projects like these in 45 days, with no public hearings required.

What happened

Detail	Facts
Facility 1	Socrates South, 200 MW, New Albany, OH. Approved June 9, 2025.
Facility 2	Apollo, 350 MW, Middleton Township, OH. Approved February 3, 2026.
Combined capacity	550 MW, dedicated entirely to Meta’s AI infrastructure.
Builder	Subsidiaries of The Williams Companies.
Approval window	As little as 45 days under Ohio’s expedited permitting law.
Public hearings required	None.
Estimated CO2 output	Up to 5 million tonnes per year at full scale.
Target completion	Late 2026.

Both plants are built “behind the meter,” meaning they connect directly to Meta’s data centers rather than feeding power into the public utility grid. Meta is financing the projects and will consume all electricity generated. The Ohio Power Siting Board (the state body that reviews large energy facilities) approved both under legislation that skips the public hearing process entirely.

According to reporting on the Apollo project, residents near the Middleton Township site did not have access to draft air permits until after construction had already started. Both projects reportedly use shell entities for day-to-day operations, which makes it harder for local communities and journalists to trace who is actually responsible for each facility.

Why is Meta building its own gas plants instead of buying grid power?

The short answer is speed. A natural gas plant can be permitted, built, and running in a fraction of the time needed for a nuclear reactor or a large solar installation. That is the explicit purpose of Ohio’s expedited approval pathway, and Meta is using it to avoid the long queues and capacity constraints that come with relying on public utilities.

The behind-the-meter model also gives Meta direct control over power reliability for its AI workloads, which demand consistent, uninterrupted electricity. Grid outages or congestion become someone else’s problem.

Why it matters

The AI compute buildout is not an abstract story about software. It has a physical footprint: steel, concrete, and in this case, natural gas combustion at a scale that rivals mid-sized industrial facilities. Two plants at full output could add roughly 5 million tonnes of CO2 per year, according to estimates cited in the source.

For investors, Williams Companies and similar infrastructure contractors now have a new class of well-funded corporate customer. If the behind-the-meter model works for Meta, other large AI operators are likely to copy it, creating a durable pipeline of construction contracts.

For communities, the Ohio story is a preview of a policy tradeoff playing out across the US. States that want data center investment are loosening permitting rules. Residents near those sites are losing the public input mechanisms that traditionally existed. Environmental groups are already pushing back, and if opposition grows, states could tighten these rules just as fast as they relaxed them, creating regulatory risk for projects still in planning.

This connects to a broader pattern we have been tracking in AI infrastructure spending, where the capital commitments involved are enormous and the timelines are compressed in ways that leave little room for public scrutiny.

Our take

Meta’s Ohio move is rational from a business standpoint. Speed of deployment is a real competitive advantage when AI model training and inference costs are falling but compute demand keeps climbing. Ohio handed them a legal shortcut and they took it.

But the 45-day approval window with no public hearings is the part worth watching. It is not just an environmental issue. It is a governance question: who gets to weigh in when a single company’s infrastructure decision could add millions of tonnes of emissions and reshape local air quality? The answer in Ohio right now is “essentially nobody outside the regulator.” That is a precedent other states will look at when they compete for the next data center campus.

For businesses thinking about AI integration at any scale, the energy story is a useful reality check. The compute that powers large language models and inference APIs has a physical cost that is not going away, and it will increasingly show up in policy, carbon reporting, and supplier due diligence requirements.

What to do about it

1. Watch Ohio’s regulatory response over the next 12 months. If public pressure forces rule changes, projects targeting late 2026 completion face timeline risk.
2. Check your own AI vendor’s energy disclosures. If you use Meta’s AI products or APIs, this infrastructure is part of your indirect emissions footprint.
3. Monitor Williams Companies and similar infrastructure contractors if you have any interest in the AI buildout as an investment theme.
4. If you are in a state currently courting data center investment, note the permitting model Ohio used and watch whether your local representatives adopt similar language.

The energy cost of AI is no longer a footnote. It is a 550 MW construction project, and it is just getting started.

Source: Bing News · Meta AI