Sabi has raised $50 million in a seed round led by Khosla Ventures to build a baseball-style brain cap the start-up says can turn brain signals into text and commands for A.I. assistants. The round included participation from Accel, Initialized Capital, DST Global, Collaborative Fund, Ascend and Kevin Weil, OpenAI’s former chief product officer.
How the brain cap works
Sabi’s device relies on electroencephalography (EEG) but departs from conventional caps: its sensors are designed to measure electrical activity without direct contact, through hair, across a gap of up to five millimeters. Each sensor is one to five millimeters in size and one report said up to 100,000 of them could fit into a single cap. The company has also developed its own EEG chip, which it says was more than 50 times as power-efficient as a widely used Texas Instruments component in internal tests. The cap connects wirelessly to a phone or laptop, much like earbuds.
The heavy lifting is performed by an A.I. model Sabi calls the Brain Foundation Model. The start-up says it is training that model on more than 100,000 hours of EEG recordings from participants imagining words or performing mental tasks, which Sabi describes as the largest disclosed dataset of its kind. The cap is intended to do three things: display imagined words as text, send trained mental commands to A.I. assistants and predict which keys a user is about to press. “We are building a faster interface that does not require surgery or a laboratory,” said Rahul Chhabra, Sabi’s chief executive and co-founder.
Sabi was founded by Mr. Chhabra and Atmadeep Banerjee, the company’s chief technology officer, who co-authored MindEye, a study that reconstructed images people had viewed from non-invasive brain scans. The team includes alumni of Kernel, Apple, Microsoft, Meta and Nike. Sabi plans to show a first prototype at the CES electronics show in Las Vegas in January and is preparing devices for people on its waitlist; it has not announced a price. Brain data is supposed to be encrypted at the sensor and processed while still encrypted so that servers never see the raw signals.
The competitive and technical challenges
Sabi enters a crowded and varied field. Elon Musk’s Neuralink, which relies on implants, was valued at $9 billion (about €8 billion) last year, and Sam Altman has backed Merge Labs as a rival to implant approaches. Meta already sells a wristband that reads muscle signals and is researching brain-to-text systems.
Existing research underscores the technical hurdles. In a Meta study with 35 participants, the error rate for decoding typed characters from conventional EEG placed on the scalp averaged 67 percent, while magnetoencephalography (MEG) performed better at 32 percent but requires a room-size scanner. That study used conventional EEG directly on the scalp; Sabi aims to measure through hair. There are no independent accuracy measurements for Sabi’s cap so far.
