Belo Monte power plant adopts local AI to streamline corporate processes

Norte Energia, the operator of the Belo Monte hydroelectric plant in Pará, has developed its own AI platform called IANE. The platform runs on Dell PowerEdge R7625 servers equipped with AMD EPYC 9354 processors and an AMD Instinct MI210 graphics accelerator. It automates administrative tasks, cybersecurity operations, and document analysis, and also converts audio instructions into searchable data.

Before developing its own solution, the company tested OpenAI models—the creators of ChatGPT—to help employees find data in contracts. However, the simultaneous increase in queries revealed limitations in processing power and accuracy, leading the company to opt for a proprietary architecture based on open-source software.

According to Júlio Freitas, Norte Energia’s financial director and head of investor relations, the goal was to go beyond simply answering questions.

The objective was not just to answer questions, but to create an intelligence layer capable of interpreting data, integrating corporate databases through orchestration, and automatically triggering operational steps, all while maintaining full control and security over sensitive information.

IANE can already detect changes in computer network protection, open tickets, and prepare justifications. Audio instruction processing routines are also converted into searchable data. However, the platform does not interfere with systems that control real-time power generation.

Economics and costs

In socio-environmental monitoring, the company estimates a time savings of two hours per day per professional, equivalent to approximately 500 hours per year. This time reduction has not yet been translated into a consolidated financial metric for return on investment.

The decision to maintain part of the local infrastructure was driven by the need for predictable costs. In models based solely on pay-per-use in the public cloud, expanding AI across the entire organization would generate increasing and unpredictable variable costs, explains Freitas. The chosen architecture is hybrid: public data and conventional applications remain in the cloud, while contract processing, strategic information, and operational records are kept in a dedicated environment.

Norte Energia does not disclose how much it spent monthly on OpenAI or the total cost of its current infrastructure.

Local infrastructure vs. cloud

Dell Technologies provided the PowerEdge R7625 server, equipped with two AMD EPYC 9354 processors and an AMD Instinct MI210 graphics accelerator. Luiz Antonio Valentim, Dell’s AI business development leader in Brazil, notes that the economic advantage depends on the type of workload, frequency of use, and intensity of operations.

There is no single cutoff point, because this equation depends on the type of workload, the model used, the intensity, and the frequency of use.

According to Valentim, continuous and high-utilization demands may favor local processing, while the cloud remains useful for experimentation and specialized services. The trend, he says, is to build hybrid environments where each workload is allocated to the most appropriate location.

In the second quarter of its 2027 fiscal year, which ended in July 2026, Dell recorded $16.4 billion in revenue from servers optimized for AI, a 100% increase over the previous year. The projection for the full fiscal year is $74 billion in this category.

Current results and next steps

In energy trading, IANE collects data from industry bodies to generate reports and summaries. In the socio-environmental area, the platform monitors websites and helps the team select content that requires analysis.

Integration with corporate databases is achieved through flow orchestrators such as n8n, which allow the AI to execute processes within defined permissions. The company retains ultimate responsibility, with specialists validating and being accountable for the results.

Future expansion still depends on an evaluation of operational results and limits of operation. Data integrity issues, security problems, or uncontrolled deviations in automations would lead to a temporary interruption of the affected flow for review.

Frequently Asked Questions

• What is Belo Monte and who operates the plant? Belo Monte is a hydroelectric plant in Pará, operated by Norte Energia, the company that developed the IANE platform.
• Does having your own AI mean training a model from scratch? Not necessarily. Norte Energia uses open-source models run locally and integrated into its systems.
• Does IANE control power generation? No. The platform supports corporate processes and does not alter real-time power generation control systems.
• Has Norte Energia abandoned the cloud? No. The architecture is hybrid: sensitive data remains in the local environment, while other applications continue in the cloud.
• What savings does the company attribute to the project? The company estimates a time savings of two hours per day per professional in socio-environmental monitoring, equivalent to 500 hours per year. There is still no consolidated financial metric for return on investment.