The Federal Government revealed an allocation of R$ 2.3 billion designated for the development of a national artificial intelligence infrastructure. This initiative is part of a new phase of the Brazilian Artificial Intelligence Plan (PBIA) and was formalized during a ceremony held in Rio Grande do Norte, attended by President Luiz Inácio Lula da Silva (PT).
The outlined strategy includes resources allocated to distinct projects involving corporations from both the United States and China. According to an official government statement released this Thursday (08/20), the main purpose is to ensure national autonomy over Brazilian data, whether generated by private entities or public services, thereby preventing excessive dependence on any single company, nation, or technology.
The PBIA, in its entirety, foresees a total of R$ 23 billion in investments. The funds earmarked for this new infrastructure will be provided by the National Fund for Scientific and Technological Development (FNDCT) and will be released in stages over the coming years.
The government has organized the implementation of this strategy into different 'routes.' A little over half of the total amount, approximately R$ 1.3 billion, will be channeled into a supercomputing project based in Rio de Janeiro. This project will be executed in collaboration with Chinese companies Huawei and iFlytek over five years.
This technological structure is scheduled to begin operations in July 2027 and will focus on advancing large language models (LLMs), applying AI solutions in specific industrial sectors, and qualifying about 3,000 professionals.
In another area, the Federal Government will reserve R$ 1 billion, through a bidding process, to acquire a supercomputer that should rank among the ten most powerful AI processing machines globally. The installation of this equipment will take place at the Augusto Severo Technology Park (PAX), located in the municipality of Macaíba, Rio Grande do Norte. The management of this resource will be shared by the Federal University of Rio Grande do Norte (UFRN) and the National Laboratory of Scientific Computing (LNCC).
It is estimated that the bidding process will take between three and six months. The Reuters agency points out that the American company Nvidia has the highest probability of winning the competition. If there are no judicial challenges, the expectation is that the supercomputer will be operational within 12 to 18 months.
The equipment destined for Rio Grande do Norte will have a processing capacity of 7.2 thousand Petaflops. This represents an increase of almost twenty times compared to Santos Dumont, the current Brazilian supercomputer located at LNCC in Petrópolis (RJ), which offers about 27 Petaflops.
In practice, this technical disparity generates significant time savings. The new machine will be capable of training a large-scale language model comparable to GPT-4 in just one month. To perform the same task, the supercomputer located in the state of Rio de Janeiro would require two to three years.
This infrastructure will be accessible for governmental use, as well as for companies, universities, and institutions dedicated to scientific research.
The plan also includes a third line of action focused on the long term: a partnership with Spain aimed at the open-source RISC-V architecture. Scheduled to be implemented within fifteen years, this project aims to position Brazil as a producer of essential semiconductors and components.
Additionally, the government strategy includes plans to establish a national cloud computing service and ensure that data storage occurs entirely within national territory. To safeguard the security and ethics of these new technologies, the National Center for Algorithmic Transparency and Trustworthy AI has been established.
This new body, operated by the Applied AI Research Center of the Federal University of Minas Gerais (UFMG), will be responsible for auditing systems, measuring risks, mitigating biases, and developing transparency methodologies. The core will also support the implementation of public policies, such as the Digital ECA, although it does not replace the competent authorities responsible for oversight and application of legal sanctions.