At the age of 29, he founded a company to develop mind-reading "baseball caps", which is valued at 600 million US dollars.

Rahul Chhabra, CEO of Sabi, says that the way humans interact with artificial intelligence in the future will be completely transformed and far more effortless. We won't have to type commands, click buttons on a computer, or even speak to AI companions on our phones.

Simply think about what you want your personal AI agent to do, and it will carry out the task accordingly.

Chhabra admits that this sounds like magic or a sci-fi plot, but his two-year-old startup wants to turn it into reality. Sabi is developing a baseball cap that can convert neural signals into text prompts recognizable by AI systems. The inner side of the prototype cap is equipped with as many as 100,000 sensors, each of which is a chip ranging from 1 to 5 millimeters in size that can capture brain activity. Then a "brain foundation model" trained with 100,000 hours of labeled neural data interprets the meaning of a thought or idea in the human brain and transcribes it into prompts that can be input into AI systems.

This may sound whimsical, after all, the concept is still in its infancy and largely unproven. But that's how frenzied AI financing is right now: Sabi has secured support from a number of well-known venture capital institutions. On October 8 local time, Sabi announced that billionaire investor Vinod Khosla will lead its $50 million seed round of financing. Khosla ranks first on Forbes' 2026 Midas List of the world's top venture capitalists and is also the first institutional investor of OpenAI.

"Simply put, the idea of inputting information directly into a computer with your mind is a very attractive concept," Khosla told Forbes.

He added: "Typing is not a natural way of communication. Before you start typing, you have to figure out what you want to input. If you can capture brain signals directly... that would be amazing."

Top venture capital firms including Accel, DST Global and Initialized Capital, as well as Kevin Weil, former Chief Product Officer of OpenAI, have also participated in the investment in this startup. According to a person familiar with the matter, Sabi was valued at $600 million in this round of financing.

01

For many years, companies such as Elon Musk's $42 billion valued Neuralink have been developing so-called brain-computer interfaces. This technology can establish communication between the brain and electronic devices such as mobile phones or computers. Neuralink's solution requires surgically implanting a device into the brain, which means shaving the user's hair, drilling a hole in their skull and implanting a chip, so that they can control a computer or robotic arm with their mind. Other companies are focusing on less invasive technologies.

Precision Neuroscience is developing an ultra-thin device shaped like a piece of tape that can be attached to the surface of the brain. Implanting this device only requires a tiny incision in the skull, without traditional craniotomy. The company has raised $430 million from investors including billionaire Bill Ackman, with its latest valuation exceeding $1 billion. Jeff Bezos and Bill Gates have invested in Synchron. The company has developed a device that can reach and be placed above the brain through blood vessels in the neck, and has raised $345 million so far. In 2025, Sam Altman, CEO of OpenAI, founded Merge Labs to study replacing brain implants with ultrasound. OpenAI led the $250 million seed round of financing for the company in January 2026.

Most of these startups are committed to helping people who cannot control their bodies due to paralysis or neurological diseases. Sabi, on the other hand, wants to enable people to communicate with personal AI agents more easily. To achieve this goal, Sabi needs to significantly lower the barriers to using this technology.

"We can't perform a billion brain surgeries, there simply aren't enough operating rooms," Khosla said. "If this solution can deliver the same functionality, it will be far better than undergoing brain surgery."

Whether this "if" can come true remains a big question. However, the startup has already seen some initial progress. Chhabra claims that in a laboratory environment, an early prototype device shaped like a helmet and connected to multiple wires achieved up to 77% accuracy in transcribing a subject's thoughts into text after the subject used it for two weeks. This is because Sabi's AI technology gradually learns how to interpret each person's thinking over time. Chhabra also said that with the help of the AI model to supplement the necessary context, convert thoughts into clear instructions, and predict the next words that will appear along the user's train of thought, the transcription accuracy can be further improved.

"AI models are becoming more and more powerful, but people worry that they may not be aligned with human goals," Chhabra told Forbes. "If AI can understand human logic and feelings, it is more likely to act in accordance with our wishes."

02

Sabi's idea originated from a study by co-founder and Chief Technology Officer Atmadeep Banerjee, published at the 2023 Conference on Neural Information Processing Systems (NeurIPS). The research results show that AI models can judge which image a person is looking at only with data collected by functional magnetic resonance imaging (fMRI) equipment. During the brain scan, participants viewed about 10,000 different pictures of objects and scenes, each displayed for 3 seconds. Banerjee, who was then an open-source researcher at MedArc, a medical AI research organization, collaborated with researchers from Princeton University to train an AI model using these pictures and their corresponding brain scan data. The model achieved 93.6% accuracy in identifying the picture the participant was looking at from 300 candidate pictures.

"This study proves that brain signals do travel outside the brain, and the signal strength is strong enough for us to use deep learning models to understand what is happening in the brain," said 29-year-old Chhabra.

Chhabra previously studied biological signals and AI memory technology at Stanford University, and completed his undergraduate studies at BITS Pilani in Pilani, Rajasthan, India. In 2023, he began discussing ideas with Banerjee, his college roommate. At that time, Banerjee had already worked at Indian satellite enterprise Pixxel, responsible for training models to identify roads and buildings in satellite images. "It turns out that building a GPS for the Earth and building a GPS for the brain are essentially the same thing," Chhabra said.

In May 2024, the two co-founded Sabi. They initially planned to make a 4-pound (about 1.8 kg) helmet, but it was bulky and inconvenient to use. "No one would want to wear something like that," Chhabra said. So they finally settled on another more lightweight wearable device that can still cover the entire scalp: a baseball cap.

Image source: SABI

However, they still need chips that can capture brain signals through hair. "Traditional EEG tests usually require shaving off hair and using gel-like sensors... the whole process is somewhat troublesome," Chhabra commented on this medical test that records the electrical activity of the brain. Chhabra said that Sabi has developed customized EEG chips for this purpose, which can detect brainwaves with medical-grade precision through hair and scalp.

Chhabra said that all conscious thinking activities occur in the outermost 2 centimeters (0.8 inches) of the brain. Every time we think of something, such as an apple, a specific area in the brain that may be as small as 1 square millimeter will be activated. By observing the correlation between neural activity and specific concepts, "you can tell what a person is thinking about," he said. As for more complex ideas such as letting an AI agent book a flight ticket, Chhabra said that the underlying AI model of the device will try to fill in the missing information and infer the implicit but unstated words in people's thoughts.

03

In order to accurately capture a person's thoughts and infer the rest of the content that cannot be directly read, Sabi's AI model needs to be trained with massive amounts of data.

Therefore, the startup has hired more than 100 contract workers to wear head-mounted devices and complete different tasks such as reading, speaking, thinking and listening. Sabi claims that the cap can predict a person's next action, such as which keys will be pressed next, or whether the person is about to pick up a pen or a glass of water. This cap will not be launched until 2027. Sabi plans to showcase this product at the Consumer Electronics Show (CES) in January 2027.

Experts are skeptical that Sabi's cap can detect signals with sufficient precision to reliably convey tasks to AI models. Nicholas Hatsopoulos, a professor of neurology at the University of Chicago, said that the chips in Sabi's cap collect signals through the skull and scalp, so the signal quality is not high. "Even if you just want to input the right prompts to get the AI to give the ideal answer, it will take a lot of effort," he said, "Not to mention now you have to get information directly from the brain. This sounds really far-fetched."

Khosla said that the early version of this cap will not be perfect, just like the first generation of iPhone. "At the beginning, the error rate may not drop to zero, but it will gradually decrease later," he said. "A technology must first be practical, and then continue to improve on that basis."

Chhabra said that if they can really make this technology a success, its application prospects will be limitless. "Computers are no longer just a set of equipment consisting of a keyboard, a mouse and a display," he said. "It more refers to AI models that can do things for you, and these models can accompany you at any time even when you are walking on the road."

It is not just billionaire venture capitalists that are bullish on Sabi. Currently, more than 10,000 people have registered on the waiting list for this product.

This article is from the WeChat Official Account "Forbes" (ID: forbes_china), author: Forbes, published with authorization from 36Kr.