A whole-brain interface usually sounds like a hardware question. More electrodes. Better materials. A bigger implant. Precision Neuroscience's new preprint argues the binding constraint sits somewhere else: the dataset.
The New York company published a manuscript on bioRxiv on 1 October describing an atlas of the human central nervous system. It is built from 70 patients, twelve academic medical centers, and more than 100,000 electrode deployments.
The framing matters more than the headline number. "The BRAIN Initiative once set a moonshot target: a system capable of recording from one million neurons simultaneously," co-founder and chief science officer Benjamin Rapoport said in the release. "That framing imagined the whole-brain interface as a hardware milestone. We believe it is, more fundamentally, a data milestone."
The paper claims coverage that is functionally complete, not anatomically complete.
What 70 patients actually add up to
The Layer 7 array is a thin film that sits on the brain surface. It carries up to 1,024 channels at a 400-micron pitch, which the company says is more than 600 times denser than contemporary clinical standards.
Across the cohort, the manuscript reports 87,342 individual microelectrodes recording from more than 100,000 unique anatomical locations. The recordings total more than 180 hours of synchronised neural and behavioural data.
The atlas covers motor, somatosensory, language, auditory and associative cortex. It also reaches brainstem and spinal cord structures that the authors describe as historically difficult to access at this electrophysiological resolution.
The results are structural as well as descriptive. Recordings reproduce somatotopic organisation across patients, including digit representations that follow a lateral-to-medial order. Adjacent representations shift over roughly 0.8 to 2.0 millimetres.
The paper also reports canonical frequency signatures, such as movement-related beta-band suppression and focal high-gamma increases during motor and speech tasks.
The hardware record is real but older. In April 2024, four arrays placed together in one patient at Mount Sinai recorded from 4,096 electrodes, which Precision called a world record for electrodes placed on a human brain.
Why the atlas doubles as a training set
Two demonstrations carry the AI angle. The team decoded upper-extremity movement from motor cortex in real time, enough for neuroprosthetic control.
A second decoder read ventrolateral cortex to infer spoken language, facial movements, and facial expressions of emotion. The paper presents this as early feasibility for combined verbal and non-verbal communication interfaces.
That is where the infrastructure argument lands. A shared anatomical reference makes recordings comparable across patients, tasks and institutions. Comparable data is the raw material for models that generalise, instead of decoders retrained for every new participant.
What the preprint does not prove
The manuscript is a preprint. PubMed flags it as not yet peer reviewed.
Layer 7 holds FDA 510(k) clearance for temporary use of less than 30 days. The paper reports no outcomes from a permanent implant, and the company's own release describes the whole-brain framing as a direction rather than a finished system.
The 70-patient cohort is also a snapshot. Precision says it has now passed 100 patient procedures, and more than 110 across 18 institutions as of the Abu Dhabi announcement.
The dataset is also surgical by construction. Every recording came from a patient already in a neurosurgical programme, which means the sample reflects people with tumours, epilepsy, or other conditions that justify an operation. That shapes which regions and which patients the atlas represents.
The history of neurotech argues for patience here. Grand claims about reading the brain have outrun clinical evidence before, and an atlas is a reference tool, not a therapy.
Why capital follows the data
Precision closed a $250 million Series D on 24 September. The round was oversubscribed and co-led by Pershing Square, the Ackman Oxman Institute, and an undisclosed life-sciences fund. Total capital raised since 2021 is $430 million.
Participants included Duquesne Family Office, B Capital, ARK Invest, Invus, Mubadala Capital, Mirae Asset Capital, Korea Investment Partners and Hitachi Ventures.
On 29 October the company announced its first clinical research site outside the United States. A two-year memorandum with Cleveland Clinic Abu Dhabi, supported by the Department of Health – Abu Dhabi, will evaluate Layer 7 in collaborative research.
Mubadala Capital, an Abu Dhabi sovereign-wealth-linked investor, had already participated in the Series D. The clinical map and the cap table are moving together.
Precision runs its own MEMS fabrication facility in Texas. It also has a commercial partnership with Medtronic to integrate Layer 7 with the company's neurosurgical platform, which places the atlas inside an existing surgical workflow rather than a research silo.
The strategic bet is legible. If neural data compounds like other large datasets, the company with the widest clinical footprint gets the reference atlas, and the atlas becomes the asset. If it does not, Precision still sells a temporary cortical interface to surgeons.
Rapoport's framing is the one to watch. A data milestone has a different failure mode from a hardware milestone. It can be large, real, and still not answer the clinical question.