AI Is Accelerating Discovery for Millions Living With Dementia

The Consortium for Biomedical Research and AI in Neurodegeneration (C-BRAIN), working with NVIDIA, is building an AI system to help Alzheimer’s researchers make faster discoveries from data that has never been accessible to AI tools or most human researchers. The consortium launched in November 2025 and is led by Dr. Randall Bateman, the Charles F. and Joanne Knight Distinguished Professor of Neurology at the Washington University School of Medicine in St. Louis. It brings together 18 member institutions across academic, philanthropic, pharma and research organizations.

The effort targets a problem that is not about a shortage of effort. There are 57 million people worldwide living with dementia today, and more than 10 million new dementia cases are diagnosed each year. National Institutes of Health funding for Alzheimer’s research has increased 7x in the last 15 years, reaching $3.9 billion annually, yet no treatment has been found to stop, prevent or reverse the disease.

The bottleneck is synthesis, not data collection

C-BRAIN describes the core bottlenecks as too few high-impact scientific findings, too much data locked in institutional silos, and a pace of synthesis too slow to keep up with published research. Much of the most valuable material — unpublished findings, negative results and governed proprietary data held by member institutions — has never been exposed to AI tools or outside researchers.

“A lot of life sciences data doesn’t make its way into the frontier models,” said Dr. Ade Ojewole, chief technology officer of C-BRAIN. “The consequence of that is we’re leaving a lot of intelligence on the table.”

C-BRAIN calls this “dark data.” Some of it is deposited directly into secure computer systems for AI tools to access. To use the rest without compromising patient privacy or institutional governance, the consortium relies on a hybrid centralized and federated approach: AI models learn from data at each institution without that data ever leaving its walls. Building and scaling that kind of infrastructure is technically demanding, which is part of what NVIDIA contributes.

Inside the compute setup

C-BRAIN currently runs 64 NVIDIA data center GPUs on premises alongside cloud-based nodes on Microsoft Azure. Additional NVIDIA GPU capacity, including NVIDIA Blackwell systems, is planned as the consortium’s ambitions scale.

The collaboration with NVIDIA incorporates technologies including the NVIDIA BioNeMo open developer platform and the MONAI framework for C-BRAIN’s research pipelines.

“The C-BRAIN collaboration with NVIDIA aims to unlock those models in a way that they can deliver something quite impactful and important,” Bateman said.

Ojewole framed the compute work as an ongoing optimization problem. “A great deal of optimization is going to be needed as we push the boundaries of what today’s spatial omics machine learning architectures can model,” he said. “NVIDIA software is definitely going to come into play there.”

Two tools already in researchers’ hands

C-BRAIN researchers have built two tools so far. OpenScientist, built with Lawrence Berkeley National Laboratory in an effort led by research scientist Justin Reese, is an open source agentic AI co-scientist. A scientist supplies a research question and dataset, and the tool semi-autonomously generates hypotheses, runs analyses and produces verifiable, auditable insights.

InsightEngine, built with 387 Labs, is an open source research platform that uses agentic AI to test hypotheses against the scientific literature. It analyzes user-supplied datasets, explores a biomedical knowledge graph and checks every claim against evidence to catch hallucinations before researchers commit to costly experiments.

In published case studies, OpenScientist completed analyses in minutes that domain scientists estimated would take weeks to months, at a cost of under $10 per run. At C-BRAIN’s current reach of more than 200 research groups, that capability could free substantial research capacity — potentially exceeding one million hours annually at scale.

What comes next

C-BRAIN’s longer-term goal is a multimodal biological foundation model trained on the full breadth of its data: neuroimaging, multi-omics, fluid biomarkers and longitudinal clinical measures. By year three, the consortium aims to discover new causal mechanisms of brain aging and Alzheimer’s disease, moving from correlation toward testable scientific principles using the full scope of available research and data.

Bateman described the underlying bet plainly. “We have a window of opportunity with artificial intelligence to do research fundamentally different than we’ve ever done before, in a way that accelerates discoveries to help patients,” he said, describing the consortium’s work as “AI tools built by scientists, for scientists.” He added that AI “that can think and reason has the potential to accelerate discoveries that will lead to powerful and implementable preventions and treatments for the entire population.”

For readers outside the lab, the practical significance is twofold. First, the project is a test case for whether federated learning can unlock sensitive biomedical data that institutions have been unwilling or unable to share — a governance question as much as a technical one. Second, the cost and time figures reported for OpenScientist suggest that agentic AI tools are moving from demonstration projects into routine research workflows, where the measure of success is not a benchmark score but hours returned to scientists and experiments avoided. Whether C-BRAIN’s year-three causal discovery goal is met will be the milestone to watch.