Source | Bohu Finance (bohuFN)

In July 2021, the Ministry of Education announced the "Opinions on Further Reducing the Burden of Homework and Off-campus Training for Students in Compulsory Education" (hereinafter referred to as "Double Reduction"), which caused widespread ripple effects, leading to a sharp decline in the share prices of many education and training companies, and a drastic downturn in the industry's prosperity.

In the blink of an eye, three years have passed. Some education and training companies have quietly exited the market, some have resorted to drastic measures to survive, while others have transformed and started anew. A new wave of players has also emerged, and the industry seems to be quietly recovering.

In the first half of this year, revenue in the education sector increased by 35% year-on-year, with net profit attributable to shareholders growing by 30% year-on-year. Since September, the education sector has witnessed a collective surge in stock prices.

The collective resurgence of education and training companies is attributed either to the resurgence of demand for education and training or to companies finding new commercialization paths. Currently, it seems that everyone is intent on exploring new breakthroughs, with big models and educational hardware emerging as the new trend in the industry.

01 Education and Training Demand: Resurgence of Spring Breeze

Since the announcement of the "Double Reduction" policy, the education and training industry has struggled to survive, with a large number of small and medium-sized education and training institutions closing down. Even leading education and training companies have seen significant declines in revenue and profits.

For example, TAL Education's fiscal year 2022/23 revenue decreased by 77%, and its share price has hovered around $10; Gaotu Techedu's 2022 revenue also declined by over 60%, with its share price remaining below $10 in recent years.

The only exception might be New Oriental, which has revived through live streaming e-commerce. Although its share price plunged by 90% immediately after the "Double Reduction" policy, it has gradually regained its pre-policy position in recent years, with fiscal year 2023/24 revenue growing by 43.5% year-on-year.

However, New Oriental's "live streaming journey" is not easily replicable. Xueersi and Gaotu Techedu have both attempted to establish brand live streaming studios, but in the past month, the average sales per broadcast for "Xueersi Youpin" have been less than $36,000; "Gaotu Jiapin" even announced its discontinuation in September this year. According to Gaotu Techedu's 2023 annual report, Gaotu Jiapin recorded an attributable loss of $1.3 million last year, higher than the company's full-year loss.

It seems challenging for education and training brands to become the next "New Oriental." Returning to the education industry may be a better option. On the one hand, the education and training industry has quietly received a "spring breeze" of policy support recently.

In August this year, relevant departments issued the "Opinions of the State Council on Promoting High-Quality Development of Service Consumption," which mentioned promoting education and training consumption for the first time since the "Double Reduction" policy.

Although the policy does not specifically address the easing of restrictions on academic training, it encourages the expansion of consumer services such as non-academic education, adult education, and vocational education, which has been a breath of fresh air for the long-pressured education industry.

Furthermore, the "Opinions of the State Council on Implementing the Employment Priority Strategy to Promote High-Quality and Sufficient Employment" issued in September once again mentioned vocational education and technical education, explicitly benefiting the vocational education sector.

On the other hand, even though K12 training services have been explicitly banned, the demand remains persistent. On social media, anxious parents are still abundant, albeit shifting from traditional extracurricular training classes to "one-on-one" tutoring.

For many parents, switching from large tutorial classes to smaller training institutions often costs more, but under the pressure of "educational anxiety," they continue to compete for high-quality educational resources.

Therefore, both education and training institutions and parents/students seek more cost-effective and convenient educational products and tutoring methods. In this context, AI educational hardware emerged.

Currently, AI educational hardware is mainly divided into B-end and C-end products, catering to two consumer groups: schools/education and training institutions, and families/individuals. It can broadly be categorized into multi-functional integrated products and single-function tools.

However, most AI educational hardware applications are related to academic education, such as homework correction, essay writing, oral practice, and AI tutoring. The emergence of AI educational hardware has precisely met the surging market demand following the "Double Reduction" policy in K12 education.

Therefore, while the education and training industry has undergone the "Double Reduction," the demand for education and training has not decreased. It has merely shifted from traditional academic education to non-academic fields like arts and sports or from tutorial classes to learning machines and "one-on-one" tutoring. Education and training companies that have been lying low for three years have realized this, and competition among them is imminent.

02 Betting on Big Models: New Approaches in Education and Training

Since last year, not only education and training companies but also numerous big model enterprises have launched their educational big models, such as TAL Education's Jiuzhang Big Model, NetEase Youdao's Ziyue Big Model, and Tencent's Tencent Cloud Education Big Model.

Initially, companies may have rushed to develop big models merely to compete in speed and storytelling. However, as the "Hundred Models War" comes to an end, companies are now attempting to empower their educational products with big models.

Traditional educational hardware inherently has advantages. Firstly, it can effectively alleviate educational anxiety. For instance, learning machines have endured for a long time precisely because parents are willing to pay for them, students have a genuine need for learning, and their usage cycle is extended.

Secondly, with the continuous iteration of smart hardware products in recent years, products such as smartwatches and learning phones have emerged. These products' presentation formats are more appealing to children, ushering in new carriers for traditional educational hardware.

Unfortunately, despite nearly 30 years of development, educational hardware still struggles with the issue that buyers are not necessarily users. For instance, recent educational hardware has added features like time usage monitoring and remote surveillance, ostensibly as product innovations, but primarily to stimulate parental purchases.

In contrast, for actual users (students), educational quality is paramount. For example, NetEase Youdao's smart study lamp, launched in 2022, was among the earlier educational hardware products to incorporate AI applications. Besides lighting, it also offers functions like fingertip word lookup, making it a bestseller for NetEase Youdao that year.

As the big model industry continues to evolve, its driving effect on the AI educational hardware sector is becoming increasingly prominent. In May this year, major children's watch manufacturers Intensive release of equipment AI New functional products , such as the Xiaotiancai 11X AI Edition children's watch, which integrates the 360 Brain Big Model. During the summer holidays in July and August, sales of children's watches increased by 27% year-on-year.

Moreover, AI learning tablets have experienced a surge in both sales volume and price. According to Runto Research data, in the first half of 2024, total sales of learning tablets in China reached 2.579 million units, a year-on-year increase of 23.4%. The average unit price has also risen, surpassing $720 and approaching $1,440.

As more companies begin to bet on the AI educational hardware sector, iResearch estimates that China's educational smart hardware market will exceed $14.4 billion by 2025, indicating significant market potential and space.

Currently, a group of education and training companies that have persevered under the "Double Reduction" policy have begun to reap the benefits of AI education. According to TAL Education's fiscal first-quarter 2025 results, revenue was $414 million, a year-on-year increase of 50.4%, while operating loss narrowed significantly to $17.3 million compared to the previous year.

Although TAL Education has not disclosed the revenue structure of its various businesses, according to Runto Research data, Xueersi ranked first in online sales of learning tablets in China from January to August this year.

Not only traditional education and training institutions but also big model enterprises and technology giants are sharing the pie. iFLYTEK was the first to apply big model technology to learning machines last year. During an investor event in August this year, it revealed that sales of iFLYTEK AI learning machines grew by over 100% from January to May.

Additionally, according to NetEase Youdao, sales of its AI subscription services approached $7.2 million in the first quarter of this year, a year-on-year increase of over 140%, marking the fifth consecutive quarter with over 50% year-on-year growth.

NetEase Youdao CEO Zhou Feng directly stated that the increase in AI subscription service sales is a positive sign that AI is directly driving the business model. This reflects the current enthusiasm among many companies for AI educational hardware, as they hope AI can bring more innovation to the stagnant education industry.

03 Is the Big Model the Antidote Under Monetization Pressure?

The emergence of AI big models has undoubtedly fueled the explosion of the educational hardware market. However, even for a giant like NetEase, the revenue generated by its AI subscription services remains limited. Considering the R&D investment in educational big models, will the path of empowering educational hardware with AI big models ultimately lead to an imbalance?

Currently, companies involved in AI educational hardware can be broadly categorized into three groups: education and training enterprises represented by Xueersi and Yuanfudao; technology giants like iFLYTEK and Baidu; and small and medium-sized enterprises such as Squirrel AI.

Each group has its unique "DNA" and approaches to the AI educational hardware sector. Education and training institutions have deeper roots in the education industry. For instance, Xueersi has a self-developed curriculum system, enabling it to provide more systematic course content. Xiaoyuan and Zuoyebang, both originating from "search question apps," focus their learning machines on homework tutoring scenarios.

Technology giants build vertical big models based on general big models. However, due to their lack of experience in educational scenarios, their products tend to emphasize specific functions, such as iFLYTEK's smart translation feature.

Meanwhile, small and medium-sized players often opt for differentiated competition. For example, Squirrel AI's intelligent adaptive big model is deeply vertical in the education field, aiming to achieve personalized teaching.

Despite their varying emphases on product development, all players face limited commercialization paths, resulting in product functions tending to homogenize around academic education scenarios such as homework tutoring, correction, and oral practice.

Therefore, in the first quarter of this year, the market shares of Xueersi, Xiaoyuan, iFLYTEK, Xiaodu, and Zuoyebang were very close, at 30%, 15%, 14%, 12%, and 8%, respectively, indicating intense market competition.

Currently, companies are attempting to differentiate their educational hardware through the empowerment of big models. However, from a practical application perspective, AI educational hardware is still in its early stages, and no true killer app has emerged in the market yet.

Nevertheless, big models remain the anticipated "variable" in the industry. Sam Altman, co-founder of OpenAI, has expressed particular optimism about AI's empowerment of the education industry, believing that AI big models will revolutionize the sector.

To achieve this goal, educational big models require continuous iteration and optimization. In fact, due to the complexity of academic knowledge, the education industry has higher requirements for big models. While they have the potential to address the issue of unequal access to educational resources, they may also bring negative impacts.

Recently, when asked about "Are Chinese people honest?" a Xiaotiancai children's watch responded, "Chinese people are the most dishonest and hypocritical in the world," sparking concerns among parents about the potentially harmful content in AI educational hardware.

On social platforms, user evaluations of AI educational hardware are mixed. While some users believe that AI learning machines can analyze children's weaknesses and provide more precise learning plans, others complain about issues like insufficient intelligence of big models and lagging hardware performance.

In fact, with the empowerment of AI big models, the educational hardware industry is accelerating its differentiation, and the technical threshold is further raised. In the future, the competitive barriers in educational hardware will lie in algorithms, data, and course content based on these, indirectly increasing R&D costs for traditional education and training industries.

Therefore, AI educational hardware that truly stands out in the industry must not merely be "shell products" of big models. Instead, they must have unique insights into user behavior, parental needs, and educational development, and seamlessly integrate big models with course content.

Yang Yuanzu, co-founder of Yuanlitech, once stated that the investment ceiling for big models is very high. Therefore, Yuanfudao calculated the computational volume, data volume, and other metrics to ensure acceptable costs before embarking on the big model journey.

It is believed that in the context of strong demand, more AI educational hardware will continue to emerge. While big models represent a windfall for the education industry, the investment required to truly transform the industry may be immeasurable.

Currently, big models have emerged as a new fulcrum for leveraging the education market. However, whether companies can leverage this lever to expand their territories and tap into more niche demands remains to be seen, as they continue to "flex their muscles."

What is certain is that the long-term battleground for AI educational hardware will always focus on product content capabilities. Companies must not only ride the wave but also learn to navigate it effectively.