HPC & AI Research in Brazil
Speakers: Carla Osthoff (National Laboratory of Scientific Computing Brazil)
Session summary
Carla Osthoff, coordinator of the National Processing Center at Brazil's National Laboratory for Scientific Computing, surveys HPC and AI research infrastructure in Brazil. The country has had ten systems on the TOP500, with Santos Dumont the only one dedicated to research; a national program provides HPC machines to nine centers, complemented by roughly 17 systems above 50 teraflops at universities and around 100 HPC research centers. She outlines Brazil's four-year national AI plan launched in 2024, budgeted at about 23 billion reais (roughly 4 billion euros), organized around five pillars. Infrastructure plans include a new AI-dedicated machine four times larger than Santos Dumont, participation with the Barcelona Supercomputing Center in a RISC-V processor project for AI, sustainable AI infrastructure, and Portuguese-language models to reduce bias and strengthen sovereignty. A consortium with industry partners such as Petrobras is pursuing a reference design for national HPC node hardware, alongside collaborations with Portugal, Julich, and India. Other pillars cover education and training, a sovereign public-data cloud with a federated system across Brazilian supercomputers, industry-oriented software stack research, and AI governance guidelines. She also presents an alliance of advanced computing centers across Latin America and the Caribbean, currently spanning eleven sites, aiming for a federated regional infrastructure inspired by European models. In the Q&A she discusses data sovereignty, ongoing evaluation of open-source cloud software, and the importance of software sustainability for choices Brazil adopts.
Topics: national ai strategy · sovereign computing · risc-v processors · federated hpc infrastructure · latin american collaboration · portuguese language models
AI-generated summary of an auto-generated transcript (~1,472 words in full). Details may be imprecise — verify against the session recording.
Auto-generated captions from ISC 2026 session recordings · transcription errors likely, verify quotes against the video · timestamps are offsets into each recording · independent tool, not affiliated with ISC · a Radixia Labs experiment
