Elea Data Centers will build BEL1, its first AI data center in the Amazon region, in Belem, Para, Brazil. As Data Center Dynamics reported, the site opens at 7.5 MW with expansion potential up to 100 MW, and operations start in Q2 2027. AXIA Energia supplies the power through a purchase agreement billed as 100 percent renewable.
A renewable PPA settles the power question. The thermal question stays open. Belem sits almost on the equator, deep in the humid Amazon basin, where ambient and wet-bulb temperatures run high every month of the year. That single fact governs how much heat BEL1 can shed and at what cost.
High wet-bulb air strips efficiency out of evaporative and air-side cooling. A cooling tower rejects heat by evaporating water, and in saturated tropical air it evaporates less per gallon, so the water draw climbs while the thermal margin shrinks. Operators in dry climates already wrestle with this tradeoff, a problem we have covered in the water and power tradeoff. In a rainforest, the math degrades further.
Scaling from 7.5 MW toward 100 MW means rack densities that air alone cannot hold. AI training halls running modern accelerators push past the point where chilled air keeps silicon in spec, which is why liquid cooling has become a physical requirement at high rack density. Cold plates and CDUs move the heat off the die, but the building still has to reject that heat to a hot, wet outdoors.
AXIA brings a portfolio of 81 renewable facilities: 47 hydroelectric plants, 33 wind facilities, and 1 solar plant, with Italo Freitas leading commercialization and energy solutions. Clean electrons answer the carbon question for the power feed. The condenser still answers to the second law of thermodynamics, and tropical air sets a punishing approach temperature there.
Alessandro Lombardi, who founded Elea in 2019 and now runs it under I Squared Capital ownership, is building near the Miramar high-voltage substation in a city hosting COP30. The design trade is direct: chiller-heavy mechanical plant to beat the wet-bulb, or liquid cooling carried close to the chip to cut the volume of air that needs conditioning. Either path raises cooling capex against a dry-climate baseline, and that heat-rejection penalty sets the real operating envelope for an AI campus in the Amazon.