The best-selling AI hardware is not AI at all.
23-year-old Ye Bowen stood at the hackathon site and asked the audience below the stage:
"Why do we have two pockets?"
Because that empty pocket is meant to hold music.
Right after saying that, he pulled a "guitar" out of his pocket (it is said that he wanted to recreate the moment when Steve Jobs pulled out the product from his pocket while launching the iPod, making the whole audience cheer).
Ye Bowen took out a square box, where chords can be played as long as fingers touch the surface. He played a short piece, and someone from the audience shouted: "When will it be available on the market? I want to buy one."
He built all this within 48 hours. The joystick was removed from a Switch controller and connected to a development board, and the code was written by AI. He once tried to learn guitar but gave up due to the difficulty of finger techniques and music theory.
What he wants is very simple: to allow people who cannot play instruments to play beautiful melodies easily.
This "guitar" was later brought to the site of WAIC, and he played the opening section of Beethoven's Piano Sonata No. 30 (Op.109) on stage.
01 The newer the concept and the stronger the AI, the more miserable the failure
Ye Bowen's "guitar" came from a hackathon competition, and he was just one of 200 contestants. This 48-hour competition produced a total of 59 works, and more than 60% of the developers were born after 2000.
If you only look at this section, you will feel that AI products are full of future and hope.
But the real business world is more pragmatic. The AI Pin made by Humane is a small screenless device clipped on clothes. When you speak to it or make a gesture, it can make calls, send text messages, answer your questions, and even project information onto your palm.
Both founders are former Apple employees, who raised $230 million in financing, and the launched AI Pin was priced at $699 for the first batch.
In February 2025, HP acquired Humane's CosmOS operating system, engineering team and more than 300 patents for $116 million. However, it abandoned the AI Pin product itself. On February 28, all AI Pins could no longer connect to the server and became electronic waste.
Lv Cheng founded Transya Technology in 2014. After the company was acquired by Baidu, he went to the United States and launched a $199 orange small box called R1, aiming to allow people to order takeout and call a taxi with one sentence without opening different apps one by one.
R1 became famous overnight at CES, selling 100,000 units in four days. Eventually, the hardware business was cut down, and the product was reduced to an app in the mobile phone.
Avi Schiffmann dropped out of Harvard. At the age of 17, he built a COVID-19 tracking website used by millions of people. He made a $129 AI pendant that listens to you and replies to your messages.
After selling 3,000 units, it also entered the Museum of Failure.
With top-tier technology and resources and very innovative concepts, these products all lack a core logic of consumer goods: in what scenarios exactly users need to use the product, and why it is irreplaceable.
On the contrary, Ye Bowen did not first figure out what AI should do. He made the pocket guitar precisely because he was discouraged by the difficulty of guitar fingering and music theory when learning the instrument.
How powerful AI is and how innovative the concept is are not the criteria for evaluation. The core is whether the product can meet real demands.
The most typical products that failed due to no real demand are the wave of products focusing on companionship.
BubblePal from Bright Innovation is a pendant that can talk and remember children's names. Fuzozo from Robo Intelligence is an AI trendy toy for adult women. ByteDance's "Visible Buddy" is a plush toy embedded with the Doubao large model.
What they sell is the feeling that "someone understands you and accompanies you".
As a result, the return rate generally exceeds 30%, and some single products even reach 40%. Up to now, none of them has become a real hit product.
Moxie in the United States is a $799 children's companion robot, backed by Intel, Toyota, Amazon and Sony. In December 2024, a key financing fell through, and all robots turned into bricks within a few days.
Conversation does not equal companionship.
Humans need companionship, but companionship is never just back-and-forth conversations.
It is not just "understanding you" and "empathizing with you".
A melody without lyrics can also accompany you through a whole night. The problem is not whether AI is powerful enough. No matter how strong the conversation ability is, it cannot fill those unspeakable parts in the concept of "companionship".
These products did not fail because their concepts were not innovative enough. They failed because they assumed that "understanding you" is a new concept that can stand on its own, but never landed in any real companionship scenarios of users.
02 Products that sell well all hit a certain real scenario
"Instead of making AI a brand new thing, we should embed AI into an existing product with established demands, get the product category right first, and then let technology play a supporting role."
Ray-Ban glasses sell well, first of all, because they are Ray-Ban products.
Behind Ray-Ban is EssilorLuxottica, the world's largest eyewear group. Its business covers everything from Ray-Ban and Oakley brands to lenses, optometry services and retail channels, with more than 18,000 physical stores spread across 150 countries.
What Meta has been trying to make is never "a pair of AI glasses", but "a pair of Ray-Ban glasses that you can never go wrong with". This pair of Ray-Ban glasses weighs only 38 grams, can be worn all day long, and has a stylish appearance. AI is just an additional function that supports shooting, calling and Q&A.
The real moat of this product line is brand and channels, not large models.
In 2025, it sold 7.4 million pairs in a single year, and more than 60% of buyers bought a second pair for others within half a year. People buy Ray-Ban, not "AI".
Plaud and Timekettle defined a small category that was not noticed by large manufacturers as a new track in a window period with almost no strong competitors. What they relied on was not how smart AI was, but the well-established channel system of "crowdfunding for verification, Amazon for sales, and independent sites for user precipitation" that they laid out overseas very early.
Plaud makes a magnetic card with a thickness of more than 2mm, which is attached to the back of the mobile phone to solve a common trouble that everyone faces: keeping records of conversations and sorting out meeting minutes.
Whether it is a call or a face-to-face meeting, recording the spoken content, transcribing it, and sorting it into key points is originally the most energy-consuming thing in the workplace, and its essence is "sorting out meeting minutes".
With almost no financing, it entered 170 countries in less than four years, with an annual revenue of $250 million. In addition to hardware sales, it also generates continuous revenue through subscriptions priced from $99 to $240 per year.
Timekettle makes translation earphones, allowing two people who do not speak each other's language to chat as fluently as native speakers after wearing the earphones.
The founder Tian Li is a former gold medal salesperson of Huawei. His mother once had altitude sickness and was sent to hospital during a trip to Europe in Switzerland, and the translation software error almost delayed the treatment, which made him determined to develop this product. In 2024, its revenue exceeded 200 million yuan, 70% of which came from overseas markets, making it the best-selling translation earphone brand in the world.
Looking at these products together, their common feature is very clear: none of them win by relying on AI. Instead, they defined a real small trouble as a product category in the gap where others had not reacted, and then consolidated the market through channels.
Oura and Radio Master Miaoplay are samples of "old demands plus long-termism".
Oura has been making sleep rings for 13 years, with total sales of 5.5 million units and more than 100 patents. It relies on the hard fact that sleep monitoring through fingers is more accurate than that through wrists, as well as a membership fee of $5.99 per month.
Hardware is only the entry point, and data and services are the core to retain users.
Radio Master has been making radios for 15 years. In March 2025, it integrated large models into its products, and the average daily listening duration rose from 40 minutes to 250 minutes.
The founder Zeng Dejun is 69 years old. He says the radio is just a carrier, and the demands behind it have always existed.
Putting all these cases together, a more macro judgment emerges: those who use AI most smoothly are often the companies that have long been selling hardware and already have a large user base.
Educational hardware follows the same path. Youdao Dictionary Pen has sold more than 10 million units in total, and iFlytek AI Learning Machine has ranked first in the high-end segment for many consecutive years. They win by digging deep into a vertical scenario, not by innovative concepts.
They sell well not because their AI is particularly excellent, but because the demands behind are real. A specific trouble can stand on its own even without the idea that "there should be a new product in the AI era".
03 Ideas do not equal value
32-year-old Wang Ning found tumors in his neck and thoracolumbar segments successively, and he has to sit in a wheelchair for the rest of his life.
While others make wheelchairs as commodities, he is making the wheelchair that will accompany him for the rest of his life. He taught himself brain-computer interface technology, used a headband to collect EEG signals, so that the wheelchair can move according to his mind. He also 3D printed the patches and circuit boards into the shape of "Golden Hoop". When others asked him why he did this,
He said: "Since I have to sit in a wheelchair for my whole life, why can't I make it cooler?"
Zhang Zhenyao, a post-2000s generation, started his business at the age of 18 and won all robot championships. He developed an all-in-one bathroom robotic arm that can wash, blow and scrub. The founder of Insta360 immediately handed him a cooperation invitation on site.
Li Pengcheng, who has hearing and speech impairments, developed a "Conan Voice Changer" that uses AI to repair his vague pronunciation in real time, so that people with speech impairments can speak smoothly. There is also a mahjong robotic arm with cost controlled within 3000 yuan, which is specially designed to solve the problem of "lacking one player for mahjong".
But such stories are not only happening in competitions.
Huang Guan works in online education in his daily job. In his spare time, he made a small chat box with a development board worth dozens of yuan, open-sourced it on GitHub, and it topped the global open source trend list within a month. Nearly 100,000 developers followed to modify it, and more than 1 million devices were connected to the project.
A stall owner in Huaqiang North followed the open source tutorial and turned a plush toy into an AI partner that can talk and recognize emotions in half an hour. A primary school student from Hubei became the youngest developer in this project.
Xiao Qiaowei, a post-90s generation, quit his job at a large Internet company to develop desktop robots. He open-sourced all the codes, circuit boards and shells, and nearly 10,000 people in the community followed to carry out secondary development.
In the past, making hardware had technical barriers: writing codes, designing circuits, and making shells. Now AI can help you write codes, write firmware, draw circuits and build models according to your demands. 3D printing can help you make the shells. A good idea can be easily realized and landed.
However, let's calm down first.
Technology inclusiveness only makes ideas easier to be developed. There is still a long way to go before commercialization. Just like in the mobile Internet era, everyone can post short videos on Douyin and write articles on official accounts, with the threshold almost zero. But the ones that can really become popular are still the most professional people.
Xue Zhiqian once said a passage on Roast which is very appropriate here:
"In this era, there is no such thing as unrecognized talent. The self-media platforms in this world are enough to carry any form of talent, but the key is that you really have the talent."
There are more than 160,000 developers on Xiaohongshu, and 90% of them have released their own products. But most of the works either face pseudo-demands, or cannot be realized technically, or cannot be mass-produced.
Similar stories have happened for several rounds. A few years ago, there was a popular saying: "All consumer goods are worth remaking."
Zhong Xue Gao remade ice cream, selling one ice cream at 66 yuan, with its valuation approaching 4 billion yuan. But in the end, the founder had to sell sweet potatoes through live streaming to pay off debts.
Tiger Head Bureau remade Chinese pastries, opening nearly 100 stores, but eventually closed all stores due to wage arrears. MissFresh remade the grocery shopping experience, burning tens of billions of yuan, and finally disbanded on the spot.
"Products with only innovative concepts and AI capabilities are fleeting. The correct logic is not AI plus hardware, but hardware plus AI."
05 The future is promising
From the steam engine era to the mobile Internet era, every technology wave will eventually return to the same question:
Whether it can land on a real person and solve a real trouble, or create a brand new business ecosystem like the 4G network did.
A young man who is afraid of the stage takes out a guitar from his pocket, a paralyzed man lets his wheelchair move according to his mind, a person with speech impairment is finally understood. The reason why these moments are touching is not because AI is so powerful or the concept is so innovative, but because they all grow out of a person's real life, rather than coming from a random idea that "there should be a new product in the AI era".
This article is from the WeChat official account "Consumer Chronicle", written by Xiaoyang, and authorized by 36Kr for release.