Digital Acceleration Part II: Powering Latin America's AI Future
- Frontera Capital
- Sep 8
- 13 min read
Updated: Sep 14
By: Yamir Hickey Morales, Amanda Perez
Introduction
The global surge in Artificial Intelligence (AI) adoption, particularly generative AI, is creating unprecedented demand for power-intensive data centers, highlighting the need for sufficient energy capacity to keep up with power consumption. Latin America and the Caribbean (LAC) is rapidly emerging as a pivotal region in this digital transformation, driven by a growing digital economy, increasing cloud adoption, and enhanced connectivity. The region's advantage of a predominantly clean energy matrix helps attract "green AI" investments, aligning with the sustainability goals of major technology firms.
However, this rapid expansion is not without complexities. While investment in data center construction is projected to grow by over 18% annually through 2030, significant challenges persist. Grid infrastructure limitations, including energy instability and lengthy connection queues, present a bottleneck to deployment. The region also requires a diversified energy strategy that leverages both rapidly deployable renewables and reliable natural gas for baseload power as long-term solutions like new nuclear builds enter the fray post 2030. Furthermore, while the immense water demands of AI data centers drive innovation in cooling technologies, they also raise environmental and social concerns, especially in key markets with water scarcity issues.
This report is the second in our Data Center series. For a comprehensive overview of the market drivers and key challenges, we recommend reading Part I: "Digital Acceleration: Mapping Latin America's Data Revolution".
1. Technological Demands of AI Data Centers in Latin America
AI applications, especially large-scale AI models, require increasingly intensive and stable energy infrastructure. This immense demand is directly linked to the computationally heavy processes of algorithm training, query processing, and the complex cooling systems necessary to maintain server operations. The core of demand lies in the energy-intensive Graphics Processing Units (GPUs). NVIDIA's latest GPU, Blackwell, is anticipated to have a peak rack density exceeding 100 kilowatts (kW). Industry experts foresee a clear path to rack densities of 1 megawatt (MW) within the next couple of NVIDIA GPU generations. This represents a dramatic increase from most operating data centers today, which typically range from 5-10 megawatts, to leading-edge AI data centers requiring 100-200 megawatts, and even planned superclusters exceeding 1 gigawatt of power requirements.
The distinction between AI development phases—training and inference—has significant implications for data center design and power usage. Training refers to the process of teaching an AI model using vast datasets, typically accomplished by clustering large amounts of compute capacity in centralized "superclusters." This phase is characterized by very high energy intensity and can be done in remote locations. In 2024, training accounted for an estimated 80% of the Generative AI market. In the long-term, this ratio will shift in favor of a predominant inference facility market.
Energy Efficiency Trends: PUE Targets and the Adoption of Advanced Cooling Solutions
Operators in the region are increasingly focusing on designing and building data centers with a Power Usage Effectiveness (PUE) of less than 1.5, achieved through the integration of renewable energy, free cooling techniques, and efficient IT infrastructure. Global industry standards are becoming more stringent; for instance, as of January 1, 2025, new data centers operating at full capacity in hot climates (relevant for much of Latin America) are required to meet an annual PUE target of 1.4. Historically, the average PUE of data centers worldwide has significantly improved, declining from 2.5 in 2007 to 1.55 in 2022, with leading technology giants striving for PUE values very close to the ideal 1.0. Liquid cooling is significantly more efficient than traditional air cooling and is favored for use with the latest generation of Nvidia's Blackwell. However, many facilities in Latin America are actively moving towards adopting efficient air-based cooling systems due to concerns over water scarcity in the region.
2. Energy Sources for Latin America's AI Data Centers: A Balanced Approach
LAC stands out globally for its clean energy matrix. In 2024, 65% of the region's electricity was generated from clean sources, significantly above the global average of 41%. Hydropower remains the primary clean energy source, contributing 41% of total generation, while wind and solar combined account for 17%. The region is actively accelerating its energy transition. By 2024, an impressive 79% of new installed capacity in LAC is projected to be renewable; the region aims for 80% renewable electricity generation by 2030 and 85% by 2050. This advantage allows the region to pursue a "powershoring" strategy, actively attracting hyperscalers and colocation providers who prioritize carbon-free energy, thereby influencing their site selection and investment decisions.
Renewable Energy
Renewable energy sources are a cornerstone of Latin America's strategy to power AI data centers, driven by compelling economic and environmental factors.
Advantages: | Challenges: |
Competitive Levelized Cost of Electricity (LCOE):Onshore wind and solar consistently rank at the bottom of the global cost curve for electricity generation. Their costs are projected to fall further by 2-11% in 2025. New wind and solar farms are already undercutting the production costs of new coal and gas plants in almost every market globally. The LCOE from solar PV has seen an 85% reduction since 2010, and wind power a 56% reduction, making them highly competitive.
Rapid Deployment: Renewable energy projects, particularly wind and solar, offer the shortest "time to power." They can be commissioned in as little as 18 months. This rapid deployment capability is a critical factor for the speed-sensitive AI infrastructure buildout.
Alignment with Sustainability Goals: Major technology companies like AWS, Google, and Microsoft have committed to aggressive carbon reduction initiatives, including 100% renewable energy targets. These companies are actively using renewable energy and improving energy efficiency in their Latin American facilities. Power Purchase Agreements (PPAs) have become a prevalent trend, exemplified by Scala Data Centers and Serena's significant 393 MW wind power deal in Bahia, Brazil, set to commence in 2025. Patria Investments recently launched Omnia, a new platform dedicated to developing over $1 billion in 100+ MW clean energy-powered data centers across LAC, specifically tailored for high-density AI workloads.
| Intermittency and Power Density: A primary challenge for renewables is their intermittent nature, as their generation depends on external factors like wind and sunlight. This, coupled with their lower power density compared to baseload sources like nuclear or natural gas, results in significantly lower utilization rates.
For instance, a typical nuclear plant can generate nearly 40 times the amount of electricity compared to a typical solar project. Consequently, wind and solar are often seen as complementary to baseload generation rather than standalone solutions for continuous, high-demand AI operations.
Speed Dependency on Interconnection Queues: Due to the plug and play packaged deployment of solar and wind solutions, if planning is not done correctly or bottlenecks exist from the transmission and distribution operators, installed units may sit waiting until interconnected to the energy grid.
|
Despite the inherent intermittency challenges, the compelling economic advantages and rapid deployment capabilities of renewable energy, coupled with the stringent sustainability mandates of major tech companies, are making renewables the preferred energy procurement strategy for AI data centers in Latin America, driving substantial PPA activity.
The widespread adoption of PPAs is a direct market response to these advantages, indicating that tech companies are actively seeking to green their energy supply in the region. This translates into lower energy costs and long-term price stability via PPAs, while enabling sustainable AI infrastructure.
Natural Gas
Despite Latin America's strong renewable energy base and decarbonization goals, natural gas remains a critical "bridge fuel" for AI data centers. Its baseload capabilities and significantly faster deployment times relative to nuclear power make it a pragmatic and necessary solution for meeting the immediate and surging power demands, especially for large-scale hyperscale facilities. The rapid construction timeline of large-scale gas plants directly addresses the "time to power" constraint, which is paramount for AI infrastructure deployment. This indicates that while tech companies prioritize green energy, a pragmatic approach to immediate power needs will involve natural gas, especially as grid infrastructure catches up.
Role as a Crucial Baseload Power Source | Natural gas is expected to play a significant role in meeting incremental AI electricity demand globally and in Latin America. It offers the advantage of generating large amounts of power consistently (baseload) and has a shorter development timeline compared to new-build nuclear plants.
|
Current and Projected Contribution | Natural gas currently generates 25% of Latin America's electricity; projected to maintain share until 2030. In January 2025, natural gas accounted for 25.9% of total generation in LAC, increased by 5.4 TWh compared to December 2024.
|
Investments in Infrastructure and Power Plants | The Latin American natural gas generator market is projected to grow from US$1.21 billion to US$2.08 billion by 2030 (CAGR 9.4%). The natural gas-fired electricity generation market is expected to reach US$1.69 billion by 2030 (CAGR 2.2%). Key developments include:
|
The rapid buildout of large-scale gas plants addresses the critical “time-to-power” challenge that underpins AI infrastructure deployment. This suggests that while technology companies continue to prioritize renewable energy sources, natural gas will remain an indispensable component of the energy mix in the short to medium term, particularly as regional grids adapt to demand and manage interconnection and improvement needs.
Nuclear Power
Nuclear power provides a high-capacity factor, carbon-free baseload generation source, making it a critical technology for Latin America's long-term energy strategy. Its operational characteristics are particularly well-suited for meeting the continuous, high-load demands of industrial processes and data centers. Argentina, Brazil, and Mexico possess established nuclear power programs, contributing approximately 5 GW to the regional grid, and El Salvador intends nuclear to provide 26% of its energy mix by 2050. This existing infrastructure and operational experience serve as a foundation for further expansion, and long-duration projects such as Atucha 2 in Argentina have provided valuable experience in large-scale nuclear construction and project management.
The sustained, non-intermittent output of nuclear generation is perfectly suited to support critical infrastructure with high power consumption, such as AI data centers. Nuclear power's ability to provide constant power without reliance on weather conditions positions it as a strategic necessity for ensuring grid stability and reliable energy supply in a rapidly decarbonizing grid.
The development of Small Modular Nuclear Reactors (SMNRs) represents a significant technological advancement. SMNRs offer key advantages in terms of capital cost reduction, reduced construction timelines, and increased deployment flexibility. Countries like Chile and Colombia are already evaluating them to enhance energy security, decarbonization efforts, and grid resilience.
According to Mark Davies, a Vice President of Cumming Group, while SMNRs are seen as a "long-term solution for AI power requirements," they are not without challenges. He points out that "there are potential bottlenecks in supply of units and the buildout of supporting infrastructure," and highlights that despite the modular nature of SMNR units, "supporting infrastructure may have to be adjusted to different counties regulations and restrictions as well as different environmental conditions which can add to the cost and timeline."
SMNRs offer a potential solution for providing a reliable, low-carbon power source to meet future energy demands, particularly for high-consumption applications like AI data centers. Their ability to deliver resilient, round-the-clock power makes them a valuable option for grid stability. Continued development and strategic deployment of SMNR technology are being pursued to address initial challenges and integrate this resource into the global energy mix.
Table 2: Energy Sources for Latin America Data Centers
Energy Source | Typical Plant Size (MW) | Capacity Factor (Utilization Rate) | Carbon Emissions (Pounds CO2 per kWh) | LCOE (USD/MWh, New Builds) | Time to Power (from decision to operation) | Carbon Emissions (Pounds CO2 per kWh) | LCOE (USD/MWh, New Builds) |
Renewables | |||||||
Solar | <300 | ~20% | 0 | $29.58 - $88.16 | 18 months | 0 | $29.58 - $88.16 |
Onshore Wind | <300 | ~35% | 0 | $31.86 - $58.54 | 18 months | 0 | $31.86 - $58.54 |
Natural Gas | |||||||
Combined Cycle | 500-1,000+ | >90% | ~1.0 | $48.78 - $81.45 | ~3 years | ~1.0 | $48.78 - $81.45 |
Nuclear | |||||||
Newbuild | 1,000+ | >90% | 0 | >$100 | >10 years | 0 | >$100 |
Source: Morningstar
Regional Spotlight: Brazil, The Regional AnchorBrazil's data center market, valued at USD 3.6 billion in 2024, is projected to reach USD 8.9 billion by 2033. The country hosts nearly 50% of all data centers in Latin America , with São Paulo absorbing 80% of the national IT load demand and boasting over 40 existing and 20 upcoming facilities. However, other regions are gaining traction, including Minas Gerais, with a planned 200MW hyperscale campus in Leopoldina and an AI data center in Uberlândia with an initial investment of R$6 billion. By mid-2025, São Paulo's occupied data center capacity is anticipated to reach 446 MW, up from 375 MW the previous year. The market is being shaped by major players and significant projects:
Energy Demands, Grid Challenges, and the Risks of Overconcentration The influx of data centers is creating immense pressure on Brazil's energy and water resources. Over 15 GW of data center connection requests have been submitted through 2035, a figure equivalent to 14% of the nation's projected electricity demand. The high-density requirements of AI workloads, which can exceed 50 kW per rack, are a major factor driving this surge. While Brazil boasts a clean energy grid (84-90% renewable), the grid's heavy reliance on hydroelectricity makes it vulnerable to droughts, which can lead to blackouts. Access to the grid is the primary bottleneck for new supply, with permitting and construction often taking five to fifteen years. In response, operators are increasingly building their own dedicated substations, such as Scala's 560 MW facility in Tamboré, which connects directly to the national grid. ANEEL, the national regulator, is also planning a major power transmission auction in late 2025 to add new infrastructure, with a BRL 1.21 billion (~US$216 million) lot for a new substation in São Paulo and a contract for new transmission lines awarded in Minas Gerais. Beyond energy, water consumption for cooling is a significant concern. A single large data center can consume as much water as a small town. The heat generated by AI chips necessitates advanced liquid cooling, which can triple water consumption per kilowatt. The concentration of this demand in São Paulo raises the risk of resource strain, particularly in a country where millions still lack basic electricity access. |
3. Infrastructure, Regulatory, and Environmental Considerations
The rapid expansion of AI data centers in Latin America, while promising significant economic growth, is simultaneously exposing and exacerbating critical challenges related to energy infrastructure, regulatory frameworks, and environmental sustainability. Addressing these considerations is paramount for the region to fully capitalize on its digital transformation. Meeting higher electricity demand hinges on robust transmission and distribution systems, necessitating increased investment in the electric grid, particularly new transmission capacity. In Latin America, power supply poses significant challenges for data center developers due to lengthening grid connection queues and reinforcement timescales of five or more years. Chronic grid instability and lengthy lead times for critical components like switchgear and transformers make secondary cities riskier for development, compelling operators to invest in additional backup systems.
For example, 2024 storms in Brazil damaged transmission lines, causing rolling outages. Brazil's national grid operator forecasts a 50% rise in distributed generation by 2029, which risks local overloading and voltage deviations if unmanaged. To meet projected demand, annual grid investment in Latin America needs to more than double to USD 30 billion by 2035 under the Stated Policies Scenario (STEPS), with almost USD 20 billion expected from the private sector. Some data centers are using behind-the-meter (BTM) power to get around grid limitations and speed up deployment. This approach puts electricity generation and consumption in the same place, reducing the need for large transmission and distribution infrastructure. The main benefit is a faster "time to power" in areas where grid upgrades are slow.
Globally, examples include Williams' $1.6 billion natural gas deal for a Meta data center and American Electric Power's 100 MW order with Bloom Energy. In Latin America, a BTM virtual power plant is already working in Puerto Rico. Siemens Energy offers BTM solutions for data centers, including energy management systems, renewables, and backup power. BTM solutions are becoming a strategic necessity for data center developers in Latin America, offering a practical way to deal with grid constraints. They speed up deployment and revenue, while reducing the need for costly grid upgrades.
Regulatory Landscape
Governments are proactively developing regulatory frameworks and incentives to attract data center investment, often linking it to stringent sustainability mandates. However, this proactive approach also involves a delicate balancing act between rapid deployment goals and addressing local community concerns regarding resource allocation.
Brazil's government is proposing tax exemptions on data center investments, covering federal taxes and import duties, to enhance its appeal as a digital infrastructure hub. This plan mandates 100% renewable energy use and reserves some server capacity for domestic needs, reinforcing Brazil's focus on sustainability and digital sovereignty. The Ministry of Science, Technology and Innovation has also committed $87.5 million in government investment to support "green" data centers.
Mexico's government has actively welcomed the data processing and storage industry, promoting a "three pillars" pro-industry approach involving government, academia, and industry. Mexico also has a voluntary standard (NMX-489) outlining requirements for building high-performance data centers with a focus on sustainability and energy efficiency.
Vulnerability to Natural Disasters
Latin America's susceptibility to natural disasters, such as severe storms and seismic activity, poses a considerable risk to data center operations. For instance, severe storms in 2024 damaged transmission lines in Brazil, leading to rolling outages that lasted several hours. This vulnerability necessitates robust design and redundant systems to ensure the continuous operation of data centers, adding to the complexity and cost of infrastructure development in the region.
4. Conclusions and Strategic Imperatives
Latin America is having an AI-driven digital transformation due to its burgeoning digital demand, strategic location, and clean energy matrix. Its strong commitment to renewables is a competitive advantage for tech companies aiming for decarbonization. However, realizing this potential requires addressing technological, infrastructural, and socio-environmental challenges. The immense power demands of AI, with increasing rack densities and the shift towards distributed inference workloads, necessitate innovation in power generation mixes and energy-efficient cooling solutions.
Critical areas include:
Integrated Energy Planning & Grid Modernization: Across jurisdictions, the grid is a bottleneck. Proactive planning is essential to ensure renewable generation can be transmitted reliably. This requires substantial investment in new transmission lines and grid upgrades, with collaboration between governments and the private sector to overcome long lead times.
Diversified & Resilient Energy Portfolio: The intermittency of renewables requires a mix. Natural gas will serve as a baseload "bridge fuel" in the short term. Data centers should also explore behind-the-meter solutions to enhance energy security. Nuclear power offers a key long-term solution and should be provided with reliable development support due to long-tail construction.
Sustainable Design & Resource Management: Escalating water consumption demands a focus on water-efficient cooling and recycling. Operators can prioritize air-based and liquid cooling systems and robust water reuse programs.
Proactive & Adaptive Regulatory Frameworks: Governments should create regulatory frameworks that balance economic incentives with environmental and social safeguards, including clear policies for land use and grid connection.
Disclaimer: This has been prepared by Frontera Capital Advisors, LLC. Information herein has been obtained from sources deemed reliable and while we have no reason to doubt its accuracy, we have not independently verified the information nor make any representation or warranty, either express or implied, as to its accuracy, completeness or reliability. This article has been prepared solely for informational purposes. The Company should not rely on the contents of this presentation or construe the contents as legal, tax, accounting or investment advice or a recommendation. The company should consult its own counsel and tax advisors as to legal and related matters concerning any transaction described herein.