Brazil launches Jaci supercomputer and advances its National AI Plan

Brazil's Ministry of Science, Technology and Innovation (MCTI) and INPE inaugurated the Jaci supercomputer in Cachoeira Paulista — the first milestone of the RISC project, which plans to modernize the institute's computing infrastructure by 2028. Jaci replaces the Tupã system and doubles available processing capacity for climate forecasting, research and AI.

The move is part of a larger picture. The Brazilian AI Plan (PBIA) 2024-2028, launched by MCTI, allocates over R$5 billion for AI infrastructure and development over four years, with a goal to rank among the five most powerful supercomputers in the world. In parallel, the LNCC (National Scientific Computing Laboratory) is upgrading Santos Dumont from 5.1 to 22–25 petaflops.

What is at stake

For Brazil, the race is not academic prestige — it is technological sovereignty. Whoever trains and runs inference on large models domestically keeps strategic data within borders, reduces dependence on foreign cloud and builds local value chains in AI.

Geopolitical context matters. With the US tightening export controls on chips and models (as seen with the global pull of Claude Fable 5 from the API), countries with their own infrastructure have more options. It is the difference between dependent access and real capability.

What is still missing

Jaci is for climatology and academic research — not corporate generative AI. For Brazil to have a competitive cluster for training large models, the needed step is different: GPUs at scale (H100/Rubin), liquid cooling in a modern datacenter and manufacturer partnerships.

The good news is that the ecosystem is assembling. Foreign datacenters are arriving with billion-dollar investments. The challenge is ensuring that infrastructure serves national interests, not just multinational demand.

The local perspective

10Dobro operates in São Luís, Maranhão — and we feel the cost of not having frontier cloud infrastructure in Brazil's North/Northeast. Training stays in the US. Inference goes to the nearest available region. Any AI project with sensitive data (health, government, agribusiness) faces the same question: where does this run safely at an acceptable cost?

The PBIA is a good step. But Brazil will not go from R$5 billion to world top-5 AI infrastructure in four years — we need to be honest about what is achievable in the timeframe.