OpenAI builds a 3.2 GW campus in Georgia: the AI electric wall moves from text to foundations

3.2 gigawatts, one site - the announcement of the OpenAI campus in Georgia reminds us that in 2026, the binding constraint of AI is measured in megawatts, not parameters.

In plain terms

OpenAI has unveiled a new data center campus in Georgia, sized at approximately 3.2 gigawatts. At this scale, it's no longer a server room—it's an industrial city plugged into a regional power grid.

Context

Since the OpenAI/Stargate wave and its US-first mesh, each announced site is read on two levels: does the land exist, and can the network keep up? Georgia has been a discreet crossroads in this race. The southeastern United States remains one of the few corridors where you can still connect at the GW scale without immediately hitting a moratorium (see New York) or local uprising (see Ireland, where data centers already represent ~23% of national electricity).

The data

• Announced capacity: ~3.2 GW for a single campus.
• Comparison: the combined load of several medium-sized American cities; the equivalent, in order of magnitude, of a few modern nuclear power plants.
• Regional context: Georgia is already under pressure to connect due to the data center wave.

Analysis

Three things matter. (1) The electricity mix: at 3.2 GW, no combination of solar/wind PPAs alone holds up. Base gas will be needed, potentially modular nuclear in the medium term. (2) The grid schedule: interconnection is the real critical path—several years between announcement and delivery. (3) The cascade effect: each GW-scale announcement from a hyperscaler makes it harder to connect competing industrial projects (fabs, batteries, hydrogen).

Scenarios

• Base: phased delivery (2027-2030), with increasing regional usage conflict.
• High: acceleration if vertical energy integration (nuclear dedicated to OpenAI/Microsoft).
• Low: policy—a moratorium like New York or an interconnection queue that extends the schedule by 2-3 years.

Risks

Impact on regional wholesale prices, local political tensions, exposure to energy security risk in the event of a major climate event.

Under the hood

At 3.2 GW, we enter the class where the data center becomes a utility case—in the regulatory sense. Expect to see interconnection procedures become politicized (public hearings, industrial prioritization) as is already the case in Ohio and Virginia.

So what

For leaders betting on AI infrastructure: the question is no longer "will we have the GPUs?" but "will we have the megawatts, where, and at what locked-in price?" Any compute strategy that does not include a state-by-state grid reading is incomplete today.