Why has the AI learning machine, which is selling like hotcakes, become one of the three major hardware sectors?

Besides new energy vehicles and smartphones, learning machines should be the most promising growth track in the hardware field.

Especially after being labeled with AI, the imaginative space of this story has been further expanded. Particularly during the current summer vacation, AI learning machines are selling like hotcakes.

At the industry macro - level, in the first quarter of 2025, the total channel sales volume of AI learning machines reached 1.265 million units, a year - on - year increase of 29.4%. The sales revenue was 4.02 billion yuan, a year - on - year increase of 15.8%.

Among them, Zuoyebang, Xueda.com, iFlytek, BBK, Xiaoyuan, and Xiaodu are the top six brands in terms of sales volume, accounting for 74.4% of the market share in the industry.

Specifically for each company, AI learning machines are increasingly becoming the "growth driver" in financial reports.

In the fiscal year 2025, the revenue of TAL's content solutions (including smart hardware and AI learning devices) increased by 50% year - on - year, and the revenue share increased to 29.4%, becoming the fastest - growing segment of the group. Among them, Xueda.com learning machines are an important part of this segment, "playing a key role in promoting overall revenue growth."

It's not only the education and training company TAL that is trying to increase revenue potential through AI learning machines; iFlytek is also among them.

In 2024, the revenue of iFlytek's intelligent education business segment was 7.229 billion yuan, a year - on - year increase of 29.94%, accounting for 30.97% of the total revenue. Among them, the growth of learning machines was particularly strong, with the annual revenue doubling.

Learning machines are not a new thing. Why have AI learning machines become the target pursued by companies in different fields and of different sizes? In 2025, which companies are growing and which are falling behind in the AI learning machine market?

Where does the growth momentum come from?

An obvious turning point is that after the "double reduction" policy in education, education and training companies collectively turned to hardware to survive.

As early as the beginning of 2023, Caixin quoted an industry insider in the education advertising business of an Internet company, pointing out the mystery behind it. In addition to increasing revenue, the more crucial intention of each company to develop the education hardware business is to acquire new customers and reach users. In the K12 field, current policies do not allow the placement of commercial advertisements. By using hardware and teaching - aid peripheral products, companies can "achieve their goals indirectly" and accurately target the desired audience.

Among many hardware products such as learning tablets, dictionary pens, and electronic word cards, learning machines are the largest product in the education hardware market.

Most fundamentally, the signature content of famous teachers' lectures included in learning machines is the best way to relieve parents' anxiety about the lack of after - school tutoring after the "double reduction" policy.

After the addition of the capabilities of large AI models, the iteration of learning machines has been further promoted, and individualized teaching has become the biggest selling point at present. Personalized training, a human - like interaction mode, and even the ability to recognize emotions. The full - process AI tutoring makes learning machines seem to have transformed into AI teachers.

AI precise learning, AI homework/essay correction, AI fingertip word - checking and translation, AI oral English graded practice, etc. are the mainstream AI functions in the market.

AI learning machines have obvious seasonal sales fluctuations. The prospectus of Readboy shows that during the winter and summer vacations in February and from June to August each year, the shipment volume of personal student tablets will reach a small sales peak.

Data from technology companies shows that online promotions also play an important role in promoting the short - term sales volume of learning machines. iFlytek's 618 battle report this year mentioned that compared with 2024, AI hardware increased by 42% year - on - year, and among them, AI learning machines won the sales championship in their category.

In addition to the above factors, major hardware manufacturers are also making market positioning layouts. For example, they are penetrating from high - end models to mid - range models.

Currently, the supply of AI learning machine products shows an "olive - shaped" structural distribution. In the first quarter of 2025, the price range of 2,000 - 3,999 yuan has become the mainstream choice in the market (accounting for more than 50% online), while the share of high - end models above 6,000 yuan decreased by 10.1 percentage points year - on - year to 7.6%, and the low - end models below 1,000 yuan also shrank simultaneously.

Xueda.com adopts such a typical expansion path:

In February 2023, Xueda.com launched its first learning machine, and then iterated at a rate of 2 - 3 new products per year. Starting from high - end models priced at 4,500 - 9,000 yuan, it gradually moved down to the mass - market price range of 2,000 - 3,000 yuan and expanded to the category of learning and practicing machines. Currently, there are 8 SKUs on sale.

This change in the product pattern may be related to factors such as "the reduction of development costs by open - source large models and the precise matching of national subsidy policies with family consumption expectations".

In 2025, learning machines were officially included in the "national subsidy" catalog. Consumers can enjoy a 15% subsidy (capped at 500 yuan) when purchasing products below 6,000 yuan, which directly stimulates the consumption growth of mid - range models. Coupled with the increasing maturity of large - model technology, it promotes the upgrade of products from "electronic question banks" to "AI tutors", allowing ordinary people to enjoy the functional benefits brought by AI technology without paying nearly 10,000 yuan.

The cards in hand of companies with different genes

Now, the player landscape of AI learning machines is very clear: the education and training camp represented by Xueda.com, Zuoyebang, and Xiaoyuan; technology companies led by iFlytek, Huawei, and Xiaodu; and traditional hardware manufacturers such as BBK and Readboy.

These three branches have different genes, but they all have strong cards in hand:

Technology companies have the strongest large - model capabilities, and the key lies in how to combine them with education scenarios. The learning machines of such companies have industry - leading capabilities in AI precise learning, essay correction, and oral English evaluation.

The advantage of education and training companies lies in their massive teaching resources of question banks covering various subjects. Therefore, the built - in resources are the core reason for the hot sales of the learning machines under their brands.

For example, the Xiaoyuan learning and practicing machine launched by Yuanfudao has sold more than one million units. It mainly targets the question - doing scenario, with the number of question banks counted in billions. The massive learning data combined with AI functions is very attractive. Most importantly, the data is constantly updated according to the teaching and research situation.

The barriers of traditional manufacturers lie in their long - term accumulated offline channels and mature supply chain systems.

BBK, which has been deeply involved in the education field for 28 years, once ranked first in the learning machine shipment volume for four consecutive years and has more than 18,000 offline sales points and after - sales service outlets. In 2022, Readboy also had 4,600 stores, and the offline channel accounted for more than 80%.

This poses real pressure on other companies. Therefore, companies including Yuanfudao and iFlytek have stepped up their offline store layouts. Yuanfudao plans to open experience stores in third - and fourth - tier cities in 2025, improve the conversion rate through VR trial learning, and transform the original hardware - selling ability into a full - cycle service system.

Although they have different genes, the target of this competition in the AI learning machine market is "the depth of educational resources and the scarcity of content".

A survey by iiMedia Research shows that 66.92% of consumers list "learning resources" as the primary factor for purchase, far exceeding the factors of hardware configuration (22.1%) and price (18.3%).

It can also be seen from the sales volume share that in Q1 2025, among the top 10 brands in the learning machine market, the total shipment volume of the education and training camp accounted for more than 60%.

That is to say, education institutions have accumulated a large amount of educational data resources, student analysis models, and teaching evaluation systems in their past business, which have become the core assets for the R & D of intelligent learning machines.

In addition, the decision - makers and users of learning machines are not the same. Parents buy them, and children use them. The younger the children are, the greater the parents' right to speak. Therefore, education and training companies have always preferred to convert students in the lower - grade school age group because this group can bring them longer - term and more revenue.

This almost points out the direction for other companies to focus on.

For example, iFlytek's "Xinghuo" large education model has been piloted in more than 50,000 schools, and its intelligent classroom covers 14 million teachers and students. The procurement of high - end AI coach products by education institutions accounts for 35%.

Although the figures show growth, in reality, there have been many voices feedbacking that AI teachers are not performing well: AI correction is not satisfactory, question - judging is incorrect, descriptions are inaccurate, etc.

Therefore, it will take a longer time to verify whether AI learning machines can really help students improve their learning effects and have practical significance or whether they are more in a gimmicky stage.

After all, the continuation of human civilization itself is not an easy task.

This article is from the WeChat official account "Newberry daybreak" (ID: new - daybreak), author: Qiao Yuqing, editor: Zhai Wenting. It is published by 36Kr with authorization.