Key takeaways

  • First phase brings a 300 MW substation, energised Dec 2027, with a path to 1 GW.
  • IT load starts at 216 MW and can be expanded to 386 MW.
  • Construction cost tops $500 M (about 2.6 bn R$) for a 944,000 m² site; 115,000 m² will be built‑up.
  • Closed‑loop liquid cooling is used to cut electricity draw and eliminate water for processor cooling.
  • The project is expected to generate about 3,000 construction jobs at peak and may help Brazil become a more prominent AI‑compute hub in Latin America.

Terranova broke ground on Campus Campinas in August 2026, marking the company’s entry into AI‑focused hyperscale infrastructure in Brazil. The development sits on a former farm in the Innovation Hub for Sustainable Development (PIDS) and is being built under a municipal tax‑incentive regime. While the project is still in its first construction phase, the design targets global‑scale AI workloads and incorporates sustainability measures that differ from traditional air‑cooled data centres.

Project overview and timeline

Construction began in August 2026 and will progress through a series of phases until 2029. The first‑phase substation is sized for 300 MW and is slated for energisation in December 2027. The campus will initially host 216 MW of IT equipment, with provisions to raise that figure to 386 MW as demand grows. The overall power envelope can ultimately reach 1 GW, should the company decide to expand the site.

Power infrastructure and capacity

The electrical backbone consists of a dedicated substation rated at 300 MW, which will be the first point of grid connection. This capacity is designed to be scalable; the operator has indicated a long‑term ambition to lift the substation’s output to 1 GW. On the IT side, the first phase supplies 216 MW of compute power, a level that places the campus among the larger AI‑centric sites in the region. Future expansion modules could push the IT load to 386 MW, effectively more than doubling the initial compute envelope.

Cooling and sustainability

Terranova opted for a closed‑loop liquid‑cooling system that circulates water in a sealed circuit. According to the developer, this approach reduces the facility’s electricity consumption and eliminates the need for water directly on the processors. In Brazil, no operational data centre currently uses liquid cooling at this scale, making the campus a regional first for large‑scale adoption of the technology.

Financial scale and site layout

The initial phase commands an investment of over $500 M (about 2.6 bn R$). The campus occupies roughly 944,000 m² of land, of which 115,000 m² will be built‑up. The remaining area will stay green or be transferred to the municipality, accompanied by public‑road hand‑over and water‑sewage treatment infrastructure. The combined power and cooling choices aim to keep operating costs low for hyperscale tenants.

Regional impact and ecosystem

Beyond the hardware, the project is expected to generate about 3,000 construction jobs at its peak, later settling to roughly 650 permanent positions. The campus joins a growing corridor of hyperscale developers in São Paulo, including neighboring projects that together bring several hundred megawatts of AI‑ready capacity to the state. By securing grid access early—an ONS authorization for 300 MW was obtained in December 2025—Terranova has mitigated a common bottleneck in Brazil’s data‑center rollout.

MetricPhase 1 valueMaximum planned value
Substation power (MW)3001 000
IT load (MW)216386
Site area (m²)944,000
Built‑up area (m²)115,000
Capital spend (USD)> 500 M

Terranova’s Campus Campinas signals a concrete step toward building AI‑compute capacity in Latin America. The combination of a sizable power budget, liquid‑cooling efficiency and a sizable land parcel provides a platform that can accommodate the rapid growth of AI workloads while keeping energy intensity in check. How quickly hyperscale cloud providers fill the 216 MW of compute will determine the pace at which Brazil becomes a key node in the global AI infrastructure network.

Sources

This article was researched and fact-checked against the following sources: