Stanford launches a five-year project to build a digital twin of the human brain

By Inside BCI · Edited by Velco Dar · November 4, 2026

If scanning starts on schedule in early 2027, about ten paid participants at Stanford will spend roughly four hours a week in MRI and MEG sessions for a year, watching videos, making memory judgments and describing what they saw. Their recordings are the first training data for an AI model meant to predict how a human brain responds across a wide range of situations.

What Stanford announced

Stanford’s Wu Tsai Neurosciences Institute launched the Stanford Digital Brain Project on 2 November 2026. Researchers across the university will spend the next five years building what Stanford calls large brain models: foundation models trained on brain imaging data that predict how the brain responds across perception, memory, language, movement, decision-making, emotion and social interaction.

The plan runs as about ten research campaigns. The first will scan roughly ten people about 100 times on each of two systems: a 7-tesla MRI scanner, which maps blood flow at roughly millimetre resolution, and optically pumped magnetometer magnetoencephalography (OPM-MEG), a wearable helmet of magnetic sensors that tracks brain activity millisecond by millisecond. The team aims to collect about 14,000 hours of data from roughly 100 people. Later campaigns may add video games and virtual reality tasks. Models will be trained on Stanford’s Marlowe GPU cluster, and Stanford says the datasets and models will be made freely available to researchers worldwide, with data that could identify a person shared under tighter access.

Philanthropists Mark and Mary Stevens provide lead funding, with additional support from Clara Wu Tsai. Mark Stevens, a former managing partner of Sequoia Capital, runs the family office S-Cubed Capital and sits on Nvidia’s board. Stanford declined to disclose the amount. Psychologist Russell Poldrack and Dan Yamins, a psychologist and computer scientist, are among the leaders of a Stanford team of almost a dozen researchers.

The stated payoff

Stanford says the models could help find early markers of conditions such as depression and Parkinson’s disease, match patients to treatments and guide precisely targeted brain stimulation. Poldrack said the ultimate goal is to predict a person’s response to treatments such as transcranial magnetic stimulation or focused ultrasound. Yamins said the project is looking for partners over time.

A Meta-funded project with the same name

A separate Digital Brain Project, funded by Meta and coordinated by Paris’s Rothschild Hospital Foundation with the Université de Montréal, launched in May 2026 with $5 million and expanded to $15 million in July. It funds nine research teams, including groups at Columbia, Geneva, Harvard-affiliated Massachusetts General Hospital and Western University, at up to $500,000 each, with a target of 15,000 hours of recordings of people playing interactive games inside scanners. Meta’s Jean-Rémi King, senior author of Meta’s TRIBE v2 brain-prediction model, sits on its executive committee, and the anonymised dataset and models are due to be released publicly. Poldrack sits on that project’s scientific committee as well.

Why it matters

Stanford’s plan is to collect hundreds of hours of scans per person, build one detailed model, then adapt it to new people with far less data. Implant makers are building on their own recordings. Neuralink said on 1 October that its trial participants had streamed more than 50,000 hours of brain activity during everyday use, and that it had pretrained a model for each participant on that person’s own recordings. If the licence terms allow it, openly released brain models built on non-invasive scans could give device companies, drug developers and neuromodulation firms a shared starting point that none of them has to pay to collect.

The bet is unproven. Bing Wen Brunton, a University of Washington biologist not involved in the Stanford project, said no one knows yet how brain models scale with data, and asked what evidence there is that it will work “besides that we’re hoping it does.”

Not disclosed

Stanford has not disclosed the size of the Stevens and Wu Tsai gifts, which partners it is seeking, or the licence terms for commercial use of its data and models. The Meta-backed project says its de-identified dataset will be released under open-source licensing but has not named the licence.

What to watch

Whether the Stanford project starts scanning in early 2027 as planned.

Whether either project names industry partners or allows commercial use of its models.

Whether the first models predict brain activity better than existing ones such as Meta’s TRIBE v2.

Sources

Primary:

Coverage:

Cross-reference to prior InsideBCI coverage:

news Stanford Wu-Tsai-Neurosciences-Institute Digital-Brain-Project foundation-models brain-data Meta Rothschild-Hospital-Foundation MEG fMRI open-data Mark-Stevens