Brazil's AI plan will be decided on the demand side
Thesis
Brazil already has AI capacity, concentrated in health and biotechnology. The Brazilian Artificial Intelligence Plan (PBIA) succeeds if it turns that capacity into capability, and Brazil’s own record shows what decides the outcome: with the same instruments, the programmes that mobilised public demand delivered, and the one without it did not.
Bio-Manguinhos turned fifty this year, and a session titled “The future is now” asked what the Brazilian Artificial Intelligence Plan (PBIA) could achieve. My answer drew on the framework in Penna, Mazzucato and Santos, published this year in Policy Design and Practice. Capacity is a stock: research base, productive infrastructure, a technical-administrative apparatus, articulated demand. Capability is a flow: the direction, coordination, instrument design and learning that put the stock to work. Industrial policy delivers when both are present.
The INOVA subprogrammes of 2011–2016 show the difference. They ran on the same plumbing, in the same years, with the same institutions. PAISS Industrial, in bioethanol, put three second-generation ethanol plants into operation and produced the first borer-resistant transgenic sugarcane. Inova Saúde integrated SUS procurement and the productive development partnerships into its policy mix. Inova Petro reported no commercial or technological results by 2022; its themes, set too narrowly around Petrobras, did not scale. The first two had articulated demand; the third had none.
Brazilian AI capacity is not abstract. AI patents filed at INPI concentrate in health and medicine, followed by chemistry and biotechnology, and the PBIA’s applied portfolio follows that map: at the April 2026 monitoring, 38% of its impact actions were in health. The plan also bets on infrastructure, with sovereignty defined as decision-making capacity, over three horizons: operate the dominant stack now, diversify integration next, co-develop open hardware later.
The same framework names the ways the plan can fail. Supply alone does not produce outcomes, and the instruments that create demand (technology procurement, public procurement for innovative solutions, the health partnerships) have to be activated. Several ministries share a single budget cycle while the technology moves faster. Trained specialists can leave, and the capacity built at home leaves with them. Frameworks for AI in clinical use, agriculture and public services have to become operations.
The vision I proposed was concrete: by 2030, a researcher at Bio-Manguinhos designs a vaccine antigen with models trained in Brazil, on Brazilian compute, using clinical and genomic data hosted in Brazil. Whether that happens will be decided by who commits to using what the plan builds.
Written with Claude Opus 5.5; reviewed and validated by the author.