Q.ai

The limit of voice assistants: you have to speak out loud

All the effort put into voice assistants over the past decade has run into one wall that does not move: to talk to the device, you have to make a sound. In practice, this rules out a large share of the hours when we carry our devices with us: in a meeting, in a library, on a train, next to someone who is asleep.

There is also a side to this that is social, not technical. Saying a command out loud near other people feels strange to most people, and that barrier is not solved by better recognition. You can improve the model endlessly and still run into it.

At the far end there is a group for whom this is not an inconvenience but a complete block: people who have lost the ability to speak for medical reasons. For them, an interface that understands the intent to speak without sound is not a convenience but the only way.

This is exactly the gap Q.ai worked on, and it deserves attention because it sits on the side that is hardly discussed: not what the device answers or which model it uses, but how you get input into it at all.

Silent speech: what exactly they try to read

When a person speaks, even in a whisper or with almost silent pronunciation, the face is working. The muscles of the jaw, tongue and lips move in different patterns for different sounds, and the skin above them moves accordingly. These movements are too small for the eye to notice, but not too small for a sensor.

According to a description of patent applications linked to the company, it worked on systems that decode tiny movements of the facial skin to understand speech, on decoding "silent speech" from muscle movement alone, and on head-mounted systems that project light or a laser onto the face to detect minute muscle activity.

Apple described the company in broader terms: developing machine learning applications for audio and communication, including the ability to understand speech at a whisper level and to improve audio performance in difficult environments. In other words, not only "input without sound" but also "better input in noise." These are two different problems solved from the same direction, because in both the missing information is in the body, not in the air.

A company that said almost nothing

Q.ai was founded in 2022 in Ramat Gan, shortly after its founder left Apple. It was led by Aviad Maizels, together with Dr. Yonatan Wexler, who came from OrCam, and Dr. Avi Barliya, an artificial intelligence researcher.

For a reader, what sets it apart is mostly what is missing: the company published almost no technical information, showed no product, named no customers and announced no milestones. Almost everything known about it comes from patent applications, media reports around the acquisition, and a short description from the buyer.

This directly affects what can be written about it: you can describe the direction and who was behind it, but you cannot say how mature the technology is, because no one outside the company examined it. Any assessment on this point is speculation, and at least for now there is no way to verify it. The reported investors included Kleiner Perkins and GV, Google's investment arm. That is a sign that professional players did see up close what the public did not.

January 2026: Apple confirms the deal, not the price

On 29 January 2026, Apple confirmed that it had acquired Q.ai. It is worth noting what was not said: Apple did not publish the price, or the size of the deal in any other form.

The numbers going around in the media are outside reports, and they do not match each other. Calcalist estimated the deal at about $1.5 billion; The Jerusalem Post cited $1.6 billion; the Financial Times reported an amount close to $2 billion. Some of the reports described the deal as the second-largest acquisition in Apple's history, after the purchase of Beats in 2014 for $3 billion.

Johny Srouji, Apple's senior vice president of hardware technologies and the most senior Israeli at the company, said it was a remarkable company that is pioneering new ways to use imaging and machine learning. The fact that it was the head of hardware who responded, and not the head of software, hints at the direction. According to reports, about 100 of the company's employees moved with it to Apple.

A second meeting between the same founder and the same buyer

Aviad Maizels is not a new name for Apple. He earlier founded PrimeSense, the Israeli depth-sensor company that Apple acquired in 2013 in a deal widely estimated at about $350 million. PrimeSense's technology, which measures the distance between objects using a sensor, was built into the iPhone's Face ID system starting in 2017.

After the acquisition, Maizels moved to Apple, and he founded Q.ai shortly after leaving it. So this is a second meeting between the same entrepreneur and the same buyer, and in both cases what was bought was the ability to sense the human body precisely.

This return is not just an anecdote. It shows something about how giant corporations buy technology: they buy not only patents but a proven relationship with a team that has already delivered for them once. The risk in any acquisition of early technology is that it will not reach a product, and a team that has already completed this path successfully reduces that risk. For an Israeli entrepreneur, one successful sale opens a second door that money cannot buy.

What is still unknown

The picture ends early, and it is worth saying so plainly. Apple has not announced which product the technology will go into, when, or whether it will at all. The guesses in the media (earbuds, smart glasses, a new interface for Siri or the Vision Pro headset) are journalists' interpretations based on the company's areas of work, not an official announcement.

What can be said with confidence is the type of asset that was bought: a team and patents, not an operating business with customers and revenue. This is the kind of acquisition meant to speed up a roadmap, not to add a revenue line. It is a familiar pattern at Apple, which tends to buy small companies and quietly integrate their technology.

One more remark is worth adding, and it is not about technology: if the ability reaches a product, its most important contribution may turn out to be in accessibility. This is a recurring pattern in the history of technology: an ability developed for the convenience of the general user turns out to be critical for a small minority that has no other alternative.