Sabi announced on October 9, 2026, that it has raised $50 million, with investors including Khosla Ventures, Accel, Initialized, DST Global, and former OpenAI product lead Kevin Weil. The Palo Alto startup only emerged from stealth this April, and what it is building is specific: a cap that reads brain signals, so words you form silently in your head appear as text on a screen.
Inside the cap: chips and sensors
Sabi takes the non-implant route — no surgery, and no electrodes glued to the scalp. It designs its own chips and sensors, centered on what it calls the first non-contact EEG chip, fabricated by TSMC, which reads brain electrical signals without touching the scalp and is low-power enough to fit inside a cap. Sensor counts disclosed at its April debut ran between 70,000 and 100,000, while conventional EEG rigs typically carry dozens to a few hundred channels — only at that density might the signal be clean enough for a model to decode.
On the software side sits its own Brain Foundation Model, trained on neural recordings from about 100 volunteers totaling roughly 100,000 hours. The company hopes the model learns patterns shared across people, so a new wearer does not need a long calibration session every time. The first generation targets about 30 words per minute; according to Sabi, the model can already predict the next three to four keystrokes a person is about to make from brain signals alone.
Why investors will pay for a hat
Typing is only the entry point. Sabi's roadmap runs from silent speech to text, to silent speech delivered directly as a prompt to an AI, and finally to driving agents by intent. If that path works, input moves one step beyond keyboard and voice — and that is its fundamental difference from speech recognition: speech recognition processes sounds you say out loud, while Sabi processes inner speech you never voice. Neural data is far more sensitive than ordinary input; the company says it is encrypted in transit and during training, with outside neurosecurity experts involved.
Keep expectations in check for now
So far, Sabi has published no peer-reviewed performance data, and the cap has no announced price or shipping date, though a waitlist is open. Signals read without contact are inherently weak and noisy — a physics-level obstacle that stacking sensors does not by itself declare solved. The $50 million buys time to keep validating; the next thing worth watching is whether it dares to put its decoding accuracy into independent tests that others can reproduce.
