Emerging Loads

The electrification of end uses is a key driver of the energy transition, replacing fossil fuels with low-carbon electricity.  It reduces GHG emissions, improves energy efficiency, and accelerates the digitalization of industrial and economic activities.  It also creates synergies with distributed energy resources (DERs), battery energy storage systems (BESS), demand response, and smart grids, enabling more flexible & resilient power systems.

  

In Brazil, emerging electricity loads are gaining importance across strategic sectors.  In agribusiness, electric-powered center-pivot irrigation and agro-industrial processing are expanding opportunities for distributed generation.  In urban areas, EVs (electric vehicles) drive demand for charging infrastructure, while industrial heat pumps enable the electrification of low- and medium-temperature process heat, particularly in the food & pharmaceutical sectors.  Concurrently, green H2 electrolyzers are emerging as major new loads, while the rapid growth of AI-driven hyperscale data centers creates energy-intensive consumers requiring large volumes of clean, reliable power.  These facilities are becoming vital considerations for electricity planning and investment.

  

This page places particular emphasis on Data Centers, whose rapidly growing power demands are reshaping electricity infrastructure & investment priorities.

 

DATA CENTERS

 

Power-hungry data centers

  

Data centers (DCs) are emerging as a major new source of electricity demand, driven by cloud computing, digital services and the rapid expansion of AI.  According to the International Energy Agency (IEA), global DC´s electricity consumption reached around 415 TWh in 2024, or 1.5% of worldwide electricity use, and is projected to more than double to around 945 TWh by 2030 [IEA, 2024].  Unlike many other loads, DCs operate continuously, with a relatively flat load profile, potentially consuming ~80–85% of their rated power capacity.  Beyond the scale of this demand, its geographical concentration creates significant challenges for electricity grids, requiring substantial investment in generation, transmission, distribution, and reliable power supply.  For Brazil, accommodating this growth while leveraging its REN energy resources will be an important challenge in electricity system planning.

​

Evolution of DCs: from web servers to AI hyperscale

​​

DCs have evolved alongside the growth of digital technology.  Early facilities primarily hosted web servers & enterprise app.; later, HPC (high-performance computing) expanded demand for intensive scientific and engineering calculations.  Today, the rapid development of artificial intelligence, particularly the training and operation of Large Language Models (LLMs), is driving demand for specialized computing infra based on GPUs and other AI accelerators.  This evolution has also accelerated the growth of hyperscale DCs: very large facilities, operated by major cloud or tech companies, designed to support massive computing capacity, extensive storage, and highly scalable digital services.

​​

Components, IT loads, power density & PUE

​

Modern DCs rely on three main types of processors.  CPUs (Central Processing Units) are general-purpose processors that manage operating systems and a wide range of computing tasks.  GPUs (Graphics Processing Units) handle many calculations simultaneously, making them particularly effective for AI training and high-performance computing.  TPUs (Tensor Processing Units) are specialized processors designed to accelerate AI workloads, particularly neural networks.  As AI adoption expands, the growing use of GPUs & TPUs increases computing power requirements and, consequently, the electricity demand and heat generated by DCs.

   

Another important consideration is memory, which plays a critical role in DC´s performance, particularly in AI workloads.  DRAM is widely used as the main memory in servers, providing fast, temporary storage for data and programs being processed by CPUs.  HBM (High Bandwidth Memory), by contrast, is a specialized DRAM stacked vertically next to GPUs.  Its much higher data BW enables large volumes of data to move rapidly btw memory & processors, reducing delays during AI train-infer.  HBM is a high-performance, high-cost memory tech used in advanced AI accelerators.  The leading DRAM mfrs are Samsung (South Korea), SK hynix (South Korea), Micron (U.S.), and CXMT (China).  The  first three companies also supply HBM, with SK hynix holding the leading position in this specialized market, followed by Samsung, and Micron.

  

Nvidia (U.S.) leads the AI GPU market, supported not only by its hardware but also by its CUDA platform, a mature software ecosystem of programming tools, libraries and AI frameworks that enables developers to harness GPU computing efficiently.  This ecosystem provides a significant competitive advantage.  AMD is another major competitor, with its Instinct accelerators targeting AI and high-performance computing, while technology companies such as Google and Amazon are developing their own specialized AI chips to meet specific computing needs.

  

The growing use of these processors directly affects DCs’ IT loads & power density: DC’s IT load is the electricity consumed by servers, GPUs, storage systems, and networking equipment to process, store & transmit data.  It is measured in kW or MW and excludes cooling and other supporting infrastructure, which add to the facility’s total electricity demand.  According to Enerdata, IT power density has increased from approximately 2 kW/m² to more than 20 kW/m² over the past decade.  While conventional racks typically operate at single-digit or tens of kilowatts, advanced AI systems can require up to approximately 250 kW per rack, demanding high-capacity power distribution and advanced liquid-cooling solutions.  These trends make electrical infra & heat removal critical factors in DCs design [Enerdata, 2026].

  

Another key efficiency indicator is Power Usage Effectiveness (PUE), defined as the ratio of a data center’s total facility energy consumption to the energy consumed by its IT equipment.  A PUE of 1.0 represents the theoretical ideal, while a PUE of 1.3 means that for every 1 kWh consumed by IT equipment, the facility uses an additional 0.3 kWh for cooling, power distribution, and other supporting systems.  Lower PUE values indicate more efficient infrastructure, although the metric does not measure the computational efficiency of the IT equipment itself.

​

Cooling techs: from air to liquid cooling

  

DCs traditionally rely on air cooling, using fans, air-handling units and chillers to remove heat from servers. Hot- and cold-aisle containment improves airflow mgnt and reduces energy consumption.  For higher-density installations, hybrid cooling combines air w/ liquid-based systems, including rear-door heat exchangers that capture heat at the rack.  Direct-to-chip liquid cooling (D2C) circulates coolant thru cold plates attached to processors, while immersion cooling submerges servers in a non-conductive fluid to dissipate heat.

  

As AI workloads drive rack power densities beyond the practical limits of conventional air cooling, liquid cooling is becoming increasingly important in AI DCs.  Direct-to-chip solutions are gaining traction, while immersion cooling offers an alternative for particularly applications.  These systems circulate coolant in closed loops, minimizing water consumption; however, additional heat-rejection equipment, such as cooling towers, may require water replenishment.  The trend is toward high-density, AI-ready facilities that integrate advanced thermal mgmt, energy efficiency, and waste-heat recovery.  A TrendForce projection indicates that liquid cooling’s share of AI chips is expected to rise from ~33% in 2025 to 53% in 2026, reflecting the rapid adoption of this tech in high-density AI DCs worldwide.

​

Digital Sovereignty: beyond DC´s location

​

Digital sovereignty goes beyond the physical location of a data center.  Hosting infrastructure in Brazil does not automatically mean that Brazil controls the data processed within it.  Three dimensions must be distinguished: territorial sovereignty, whereby equipment is located within Brazilian territory; legal sovereignty, whereby operations are subject to applicable Brazilian laws, including the General Data Protection Law (LGPD), even when the controlling company is foreign; and operational & technological sovereignty, which concerns who controls the servers, SW, administrative credentials, encryption keys, and access to or transfer of data.  True digital sovereignty therefore depends not only on where data is stored, but also on who controls the infra, techn, and data governance.

​

Brazil’s Data Center Landscape

   

Brazil has approximately 200 DCs, compared with more than 5,400 in the U.S., highlighting the substantial difference in scale btw the two markets.  Nevertheless, Brazil is one of Latin America’s leading digital infrastructure hubs, supported by its large economy, expanding cloud computing demand, REN energy resources, and growing interest in artificial intelligence (AI).  These figures are based on Cloudscene data reported in Stanford University’s AI Index Report 2026 and may vary according to the definition and coverage of DCs [Stanford, 2026].

  

Located in São Paulo state, Ascenty’s Vinhedo DC Campus illustrates the scale of investment in Brazil’s digital infrastructure.  The first phase reportedly received BRL 500 million in investment, with around 40,000 m² of area and 27 MW of power capacity attributed to Vinhedo 1; an earlier company statement cited 40 MVA across five buildings.  Vinhedo 2, inaugurated in 2020, received a further USD 150 million investment.  Together, the two facilities provide about 46,000 m² of combined area and around 7,300 racks.  Ascenty has reported a combined capacity of 36 MW, although published figures vary across company materials.  Figure 1 shows Vinhedo's DC, located at São Paulo, while Figure 2 shows the geographic locations of Brazil’s DCs [DCMAP, 2026].

  

Ceará: Brazil’s emerging AI DC superhub

  

The Omnia Data  Centers  project, developed in partnership w/ Casa dos Ventos to serve ByteDance (TikTok), is under construction at the Pecém Industrial and Port Complex in Ceará. The initial phase is designed for 200 MW of IT capacity, with at least 40 MW expected to become operational by Q3 2027. The broader campus could eventually reach 1 GW.  Ceará offers two key advantages: Fortaleza handles roughly 95% of Brazil’s submarine fibre-optic connections, and the state has substantial wind & solar resources.  However, environmental licensing issues remain a potential risk to the operational timeline.

  

Brazil’s National High-Performance Computing Network (SINAPAD)

  

Brazil’s National System for High-Performance Computing (SINAPAD), under the Ministry of Science, Technology and Innovation (MCTI), connects specialized computing centers known as CENAPADs throughout the country.  Coordinated by the National Laboratory for Scientific Computing (LNCC), the network provides researchers and institutions with access to high-performance computing (HPC), data storage and advanced scientific computing resources.  These capabilities support computationally intensive applications in climate modeling, engineering, materials science, bioinformatics and AI.

​  

The network includes major academic and research institutions such as UNICAMP, UFRGS, UFMG, UFC, UFRJ, and INPE, as well as the LNCC, home to the Santos Dumont supercomputer, one of Brazil’s leading scientific computing systems.  In August 2026, MCTI also designated the Laboratory of Computing of Central Brazil at the Federal University of Goiás (UFG) as a CENAPAD, subject to a plan to meet the network’s full requirements.  UFG’s inclusion expands the potential for advanced computing services in Brazil’s Central-West region, supporting research and innovation in agribusiness, energy, environmental monitoring, and AI.​

  

Brazil’s Artificial Intelligence Plan - PBIA

  

As demand for AI and high-performance computing grows, DCs are becoming increasingly important to Brazil’s digital economy.  Their expansion will depend not only on investment and connectivity, but also on reliable power supply, energy efficiency and access to low-carbon electricity.

  

Against this backdrop, Brazil’s Artificial Intelligence Plan (PBIA) 2024–2028 foresees about BRL 23 billion in investments to develop national AI capabilities, expand high-performance computing infrastructure, train skilled professionals and promote technological sovereignty.  By encouraging domestic AI development and advanced computing infrastructure, the plan could strengthen Brazil’s position as a regional AI hub while increasing demand for DCs and reliable, low-carbon electricity [MCTI, 2026].