Argentine Patagonia is being marketed to AI data center developers, and the capacity figures being floated are large next to anything the country has actually built. Pampa Energía is hunting for investors for a facility of up to 500 MW in Neuquén province. Poland's Green Capital wants to break ground in Chubut at 300 MW and eventually reach 3,000 MW. OpenAI announced a project with Sur Energy in October that has still not been signed. Reuters reports that investor interest rests on the region's cool climate, hydropower, wind, shale gas from the Vaca Muerta formation and large tracts of open land.
The gap between the pitch and the demand is the part worth reading closely. Pampa is advertising up to 500 MW and plans to sign its first contracts by the end of 2026, but the parties circling the project want to start at 20 to 40 MW. That is under a tenth of the headline number. The 500 MW figure is what the site could hold, not what anyone has asked for.
The financing asymmetry follows from that. Building out the grid infrastructure for the full capacity runs to roughly $900 million — a bill sized for 500 MW of load, in front of customers who are asking for 20 to 40. Someone has to fund the difference between the buildout and the bookings, and nothing in the plans says who.
Green Capital's numbers have the same shape, stretched further. A 300 MW start is a real data center. A 3,000 MW target is national-scale infrastructure, roughly the output of a large generation fleet, and it exists at the moment as a long-term intention rather than a contract. Announcing both in the same sentence flattens the distance between them.
OpenAI's October announcement with Sur Energy is the clearest case. It was announced, and it is unsigned. An announced-but-unsigned project is a position in a market for future power, not a commitment to build.
Read together, what Patagonia is selling right now is optionality on electricity and land. Every advantage on the list — climate, hydro, wind, shale gas, space — is a resource endowment. None of them is a permit, a grid connection, a submarine cable or a signed offtake. Those are the things that have historically decided whether a data center gets built, and they are the things still outstanding here.
The absence of local opposition is being read as an asset, and that reading is too generous. Argentina has seen no meaningful resistance to server complexes of the kind that has stalled projects in the United States, but it also has almost no large facilities to resist. The test case is already visible: Pampa wants to site its project beside its Loma de la Lata power plant, in an area where Mapuche families live and have clashed with energy companies more than once. The country has not yet had this argument, which is not the same as having won it.
Then there is the clock. Investors want to know how stable business conditions will be after the 2027 presidential election, and the first contracts are targeted for the end of 2026. That sequencing is not incidental. The deals are meant to close inside the window before the political event that could change the terms — which tells you how much of this project's value the developers believe rests on who is in office, rather than on the wind, the water or the gas.