Solaer Chile AI Data Center Solar-Plus-Storage Analysis — Atacama Desert Energy Infrastructure 2026
Overview of Solaer's AI Data Center Energy Vision
Israeli renewable energy developer Solaer has announced plans to enter the digital infrastructure sector with an ambitious integrated energy project in Chile's Atacama Desert: a dedicated AI data center campus powered by 197 MW of solar photovoltaic generation, 500 MWh of battery energy storage, and a seawater desalination system providing cooling water via approximately 400 kilometers of pipeline from the coast. The project, sited on approximately 2,000 hectares of long-term leased land, represents a vertically integrated approach to data center energy supply that departs fundamentally from the standard hyperscaler model of signing virtual power purchase agreements (VPPAs) with grid-connected renewable projects.
Solaer's Chilean subsidiary has secured a 36-month initial assessment period for the land lease, with an operational term of 25 years extendable to 40 years upon project sanction. The project leverages Solaer's existing presence in the Atacama region — the company has 129 MW of operating renewable assets in Chile, 1.4 GW under construction or in pre-construction, and a 1.5 GW development pipeline — as well as its ENAPAC seawater desalination project, which provides the water infrastructure backbone for the data center's cooling requirements. The integrated energy-water-computing campus model reflects an emerging recognition that AI-scale data centers — with power demands of 100-500 MW per campus and 24/7 load profiles — require fundamentally different energy infrastructure than the retail and wholesale electricity market PPA structures that have dominated renewable energy procurement for the past decade.
Why Dedicated On-Site Generation Matters for AI Data Centers
The standard renewable energy procurement model for data centers — a hyperscaler (Google, Microsoft, Amazon, Meta) signing a VPPA with a grid-connected wind or solar project in the same regional grid — works well for load profiles where temporal matching between generation and consumption is not critical. A hyperscaler consuming 100 MW of baseload-equivalent power from the grid and holding VPPAs for 150 MW of wind and solar capacity achieves annual net-zero energy on a MWh-accounting basis, but at any given hour, the data center's actual electrons may come from gas, coal, or nuclear plants while the VPPA project exports renewable electrons elsewhere on the grid.
For AI training workloads, this temporal mismatch creates both economic and reputational risk. AI training clusters — arrays of 10,000-100,000 GPUs running training jobs that can last weeks or months — operate at near-constant, high-utilization load profiles (90-95% capacity factor during active training runs), fundamentally different from the variable load profile of traditional cloud computing workloads that can be time-shifted or geographically relocated to match renewable generation availability. A 100 MW AI training cluster operating 24/7 at 90% capacity factor consumes 788 GWh annually, equivalent to the output of a 300 MW solar plant (at 30% capacity factor in a high-irradiance location) or a 250 MW wind farm (at 35% capacity factor). Achieving hourly-matched (24/7 carbon-free energy) for this load profile requires either massive overbuilding of renewable generation plus storage, or integration of firm clean resources (geothermal, nuclear, hydro) — and the Atacama Desert offers a unique combination of the world's highest solar irradiance, making the overbuilding-plus-storage pathway unusually cost-effective.
Solaer's approach — colocating generation, storage, and load on a single site behind a shared transmission interconnection — solves the temporal matching problem at its root. The 197 MW solar plant, oversized relative to the data center's steady-state load (estimated at 80-120 MW based on the solar-to-load ratio), charges the 500 MWh BESS during the 10-12 daily hours of peak solar generation. The BESS then discharges to cover the data center's load during evening, night, and early morning hours when solar generation is unavailable. The 500 MWh storage capacity relative to an estimated 80-120 MW load provides 4-6 hours of full-load coverage, sufficient to bridge the typical Atacama overnight period plus a buffer for cloudy days (though the Atacama averages fewer than 15 cloudy days per year at its driest points).
Technical Deep Dive: Engineering a 24/7 Solar-Plus-Storage Data Center
Designing a solar-plus-storage energy system to power a 24/7 AI data center requires solving a multi-variable optimization problem that is fundamentally more complex than either standalone solar design or standalone data center electrical design. The key engineering parameters and their interactions are as follows.
Solar resource and generation profile: The Atacama Desert receives the highest annual direct normal irradiance (DNI) of any location on Earth — approximately 3,000-3,500 kWh/m²/year at the best sites, versus 1,500-2,000 kWh/m²/year for typical utility-scale solar locations in the US Southwest or Southern Europe. At these irradiance levels, a single-axis tracking PV system can achieve capacity factors of 32-38% (vs 22-28% for fixed-tilt systems in good US locations), meaning a 197 MW DC installation would generate approximately 550-660 GWh annually. The daily generation profile is highly predictable — near-zero output from approximately 8 PM to 6 AM, a rapid morning ramp from 6 AM to 9 AM, peak output from 10 AM to 4 PM, and a gradual afternoon decline from 4 PM to 8 PM — with minimal seasonal variation due to the Atacama's equatorial proximity (latitude 24°S).
Storage sizing methodology: The 500 MWh BESS must be sized to cover the net load (data center load minus concurrent solar generation) during non-solar hours. The key design parameter is the "design day" — the 24-hour period with the lowest solar generation and/or highest load that the system must cover without resorting to backup generation. For the Atacama, the design day is defined by rare cloud events (less than 5% probability annually) rather than seasonal solar variability. The 500 MWh capacity at 80-120 MW data center load provides 4.2-6.25 hours of autonomy, which, combined with the predictable daily solar recharge, achieves a loss-of-load probability (LOLP) of approximately 0.1-0.5% — equivalent to 9-44 hours of unserved load per year. For an AI training data center, where training jobs can be paused and resumed, this level of reliability is operationally acceptable; for a colocation data center serving latency-sensitive enterprise workloads, backup generation or a second grid interconnection would be required.
Cooling-water synergy with desalination: AI GPU clusters generate immense heat — a single NVIDIA H100 GPU dissipates 700 W at full load, and a cluster of 50,000 H100s dissipates 35 MW of thermal load, requiring 10-15 MW of cooling system power (chillers, pumps, cooling towers) and significant water consumption for evaporative cooling or heat rejection. The Atacama Desert's zero-rainfall environment makes conventional cooling tower operation impossible without desalinated water — which is where Solaer's ENAPAC desalination project becomes critical infrastructure, not just an ancillary feature. The desalination plant powered by the same solar-plus-storage system creates a closed energy-water loop: solar electricity powers both the GPU compute load and the reverse osmosis (RO) desalination process (approximately 3-4 kWh/m³ of produced water for seawater RO), and the desalinated water cools the GPUs, with the waste heat potentially recovered for desalination preheating (improving RO membrane flux rates at elevated feedwater temperatures) or rejected to the atmosphere through dry coolers as a last resort.
Grid interconnection and islanding: While the Solaer project is designed for islanded operation (no grid connection required for normal operations), a grid interconnection is likely included for two reasons: (a) export of excess solar generation during periods when the BESS is fully charged and the data center load is below peak solar output (approximately 50-100 MW of midday excess generation), creating an additional revenue stream from merchant power sales into Chile's spot market, and (b) backup power supply during extended maintenance outages or unforeseen system failures. The interconnection capacity would be sized for the data center's full load (80-120 MW) rather than the solar plant's output, as the primary function is backup import, not primary export.
Atacama Desert: The World's Optimal Location for Solar-Powered Computing
The Atacama Desert's suitability for this project extends beyond solar irradiance to encompass a unique combination of co-location advantages that would be difficult to replicate anywhere else on Earth. First, the region is already a major mining hub — copper and lithium mining operations in the Antofagasta and Atacama regions consume gigawatt-hours of electricity and millions of cubic meters of water annually, creating an existing ecosystem of energy and water infrastructure, skilled technical workforce, and supply chain logistics that a greenfield data center project can leverage. The co-location of a data center with mining operations also creates opportunities for industrial symbiosis: waste heat from data center cooling could preheat mining process water or provide space heating for mining camp facilities, and the data center's high-quality, redundant power infrastructure could provide backup or supplemental power to mining operations during grid outages.
Second, Chile's stable regulatory environment, investment-grade sovereign credit rating, and free trade agreements with 65+ countries (including the US, EU, China, and most of Asia-Pacific) make it an attractive jurisdiction for long-term digital infrastructure investment. Chile's data center market is nascent — the country currently hosts fewer than 10 commercial data centers, predominantly in Santiago — but its combination of renewable energy resources, political stability, and geographic position (serving as a Latin American connectivity hub with multiple subsea cable landings) positions it for significant growth as AI computing demand expands beyond traditional North American and European data center clusters.
Third, the land availability economics are compelling: 2,000 hectares of Atacama Desert land at lease rates of $50-200/hectare/year represents an annual land cost of $100,000-400,000 — negligible in the context of a data center project where the GPU compute hardware alone may cost $500 million to $1 billion. The land cost for an equivalent solar-plus-storage site in Northern Virginia (the world's largest data center market) would be 50-100× higher, and solar irradiance would be 40-50% lower, making the Atacama configuration economically attractive even after accounting for the cost of the 400 km desalination water pipeline.
Industry Impact: Redefining AI Data Center Energy Procurement
Solaer's project, if successfully executed, would establish a new template for AI data center energy infrastructure that diverges from both the hyperscaler VPPA model and the behind-the-meter renewable model used by smaller colocation providers. The key innovation is the physical integration of generation, storage, and load on a single site, eliminating transmission dependency, wholesale market price risk, and grid interconnection queue delays — the three largest sources of schedule and cost uncertainty in traditional data center development.
For the broader energy storage industry, projects like Solaer's create a new demand category for BESS that is distinct from the three traditional storage market drivers: grid-scale ancillary services and capacity, renewable integration and curtailment reduction, and behind-the-meter commercial/industrial peak shaving. AI data center storage demand is characterized by a requirement for dedicated, island-capable, 4-8 hour duration storage — a specification that aligns well with the LFP battery technology that dominates current manufacturing capacity but may evolve toward longer-duration technologies (flow batteries, iron-air, compressed air) as AI training clusters scale to 500 MW+ and require 12-24 hour storage autonomy.
Future Outlook: The AI Energy Infrastructure Arms Race
The Solaer project is a single data point in what is rapidly becoming an AI energy infrastructure arms race. Global data center electricity consumption is projected to grow from approximately 460 TWh in 2025 to 800-1,000 TWh by 2030, driven predominantly by AI training and inference workloads. The seven largest hyperscalers (Amazon, Microsoft, Google, Meta, Apple, Oracle, and Tesla/xAI) have collectively contracted over 100 GW of renewable energy capacity globally, but the pipeline of contracted projects is increasingly insufficient to meet the 24/7 hourly-matched carbon-free energy commitments that these companies are making to investors, regulators, and customers.
The emerging solution set includes: (a) colocated solar-plus-storage like Solaer's Atacama project, (b) advanced nuclear — small modular reactors (SMRs) and microreactors specifically designed for data center co-location, with multiple hyperscalers and nuclear developers announcing partnerships in 2025-2026, (c) geothermal — next-generation enhanced geothermal systems (EGS) providing firm, 24/7 carbon-free power with minimal land footprint, and (d) long-duration storage paired with overbuilt renewables, using iron-air, flow battery, or compressed air storage to achieve 24-100 hour autonomy. Solaer's project, as the most advanced example of the colocated solar-plus-storage pathway for AI, will be closely watched as a proof-of-concept for whether dedicated, island-capable renewable-plus-storage campuses can compete with grid-connected, nuclear-backed data center models on both cost and reliability metrics.
For deeper analysis of energy storage solutions for data centers and industrial applications, explore our comprehensive energy storage resource center and solar-plus-storage system design guides.
