Paraguay's hydro surplus is not an interconnection agreement

Paraguay's treaty share of Itaipú, 7,000 MW, covers roughly 86% of the country's own demand, and December 2025 peak domestic demand ran about 5,280 MW against roughly 8,760 MW of available capacity, per figures compiled by Hashrate Index from public utility data. That gap between what the dam produces and what Paraguayans actually draw is now the entire pitch behind a reported $8 billion in prospective AI data center investment. It is a real number. It is also not the number a lender should be financing a data center against.

A national energy balance surplus and a financeable interconnection are different instruments, and Paraguay already ran a live test of what happens when a fast-growing load treats them as the same thing.

The balance sheet number and the interconnection number

A system-wide surplus is arithmetic: total generation entitlement minus total domestic consumption. It says nothing about what capacity sits at any one substation, at any one voltage level, on any one line someone still has to build to reach it. ANDE's own Maestro Plan, reported by Canal Solar, projects national peak demand growing 6.9% to 9.4% a year through 2027, reaching roughly 5.5 GW. A gigawatt of new data center load does not appear at the dam. It appears at a 220 kV or 500 kV connection point, and that point has to be engineered, permitted and energized before it means anything to a term sheet.

ANDE already ran this experiment

Crypto mining got there first, and it is the closest thing Paraguay has to a precedent for a price-sensitive industrial load scaling against the surplus story. ANDE initially pointed miners toward more capacity than it ultimately delivered. ABC Color reported that medium-voltage, 23 kV mining allocations were revised down from an expected 117 MW to 69.3 MW, and tariffs for the sector have risen since. None of that is a scandal. It is a state utility doing what state utilities do once a load class gets large enough to matter to its own revenue, which ABC separately put at roughly $350 million from mining in 2026, and large enough to compete with the households it answers to first.

A megawatt in a national energy balance is a planning assumption. A megawatt at a signed interconnection point is a financeable asset. Paraguay has an abundance of the first and is still building the second.

What the financing documents actually say

Reporting on the proposed AI data center pipeline notes that the preliminary connection documents now circulating do not obligate either party to execute a Contrato de Conexión y Suministro, the supply and connection contract that would actually make a project bankable. That is a normal stage for an early pipeline anywhere in the world. It becomes a problem the moment the surplus headline, rather than the contract, is what shows up in an investment memo as proof the power already exists.

The grid is being built, just not on the pipeline's clock

ANDE is not standing still. The IDB, EIB and EU are financing roughly $150 million of transmission strengthening centered on a new Emboscada substation and the reconfiguration of two 500 kV lines, work the IDB expects to benefit close to 2.7 million people by 2030. Separate tenders cover a 500 kV Yguazú–Valenzuela line and a Yacyretá–Villa Hayes corridor. These are the right projects. They also run on a multi-year construction timeline that has little to do with how fast an $8 billion pipeline wants to move, and everything to do with when a hyperscaler's load actually gets a place to plug in.

What we think follows

Treat Paraguay's surplus as real and largely uncontracted, which is a different thing from unavailable. The resource exists. What is missing is the paperwork that converts a planning number into a financeable one: a signed interconnection and supply agreement at a named node, tied to a construction date for the lines that reach it. Underwrite the agreement, not the dam.

Credit where due. A country funding its own grid upgrades through multilateral lenders while still sitting on a surplus most of the region would envy starts from a better position than most LATAM grid operators. Chile and Peru would each take Paraguay's version of this constraint over their own.

Our own pipeline sits in Peru, where the binding constraint runs the other direction, generation queued behind transmission rather than load queued behind it. ZFC's roughly 870 MW of projects in development, including Olmos Wind at approximately 135 MW with battery storage integrated into the design in Lambayeque, carries its own interconnection timeline into Ready-to-Build rather than a headline capacity figure. The discipline is the same regardless of which side of the meter the queue sits on: ask what is actually built and contracted, not what a balance sheet says is theoretically spare.

The open question

Does ANDE's transmission build and the connection-agreement process firm up into binding megawatts before the $8 billion pipeline tries to close financing against the surplus headline instead of a signed contract? The multilateral money is moving the right way. The mining precedent suggests ANDE will ration hard once a load class collides with its own revenue and supply priorities. We do not know which dynamic wins first, and we have not met anyone drafting a data center term sheet in Asunción who does either.

See the full active portfolio or read more Insights.