Scrambling for user traffic entrances and exploring monetization channels, why are large model companies collectively "deviating from their core business"?

On October 13, two large model companies launched AI hardware on the same day. In Shanghai, Yin Qi, Chairman of Stepverse, unveiled an agent smartphone named STEPX Neo. In Hangzhou, Alibaba Qwen launched two AI glasses and one AI headphone.

This is not a sporadic move by large model companies. On September 16, the nubia NaviX Ultra jointly developed by ByteDance and ZTE was launched on the market, more commonly known as the "Doubao Phone", with over 360,000 reservations on the JD platform. Earlier on September 2, Tencent WorkBuddy Open Platform was released in Shenzhen, with the first batch of 9 co-branded hardware products covering categories such as glasses, headphones, voice recording cards, speakers, desktop pets, and live streaming equipment.

Several mainstream large model companies in China have all stepped into the hardware sector.

An executive who once worked at a leading large model company in Beijing told the Economic Observer that large model companies develop hardware mainly to seize entry points — users have no loyalty to models, and through hardware, they can cultivate users' usage habits and retain them. This person once led a team to explore the AI smartphone project in 2024.

Aim to Seize Entry Points

Although hardware is being sold, what Alibaba Qwen and Stepverse focused on demonstrating at the press conferences, as well as what the outside world is paying attention to, are all AI capabilities.

The wake word for Qwen's headphones and glasses is "Hello Qwen". The hardware needs to be used in conjunction with a mobile app, and the core services recommended on the mobile app interface include AI task assistant, voice recording minutes, portable translation and other functions.

This pair of glasses can also order takeout and call a ride, which are all core capabilities promoted by the Qwen app. Song Gang, Vice President of the Qwen Business Unit of Alibaba Group, believes that AI wearables are the extension of Qwen in the physical world and also an important part of Personal Agent. He said AI hardware can enable the Qwen app to better enter the real world and gain active service capabilities.

Stepverse's smartphone also focuses on demonstrating AI capabilities. There is a separate screen on the back of the Stepverse smartphone, which displays whether the Agent has completed the task with different emojis. In addition, at the Stepverse smartphone launch event, Yin Qi announced a welfare policy: users who purchase the Stepverse smartphone can get a monthly usage quota of tens of billions of Tokens for 12 consecutive months, and get priority access to new model experiences. Calculated based on the official model package price of Stepverse, this welfare is equivalent to subsidizing users with nearly 3000 yuan worth of Tokens.

New Monetization Channels

In July this year, Yin Qi mentioned in an interview with media including the Economic Observer that Chinese large model companies need to find new entry points instead of only making profits by selling Tokens. His judgment is that any model company will hardly survive if it does not own an application that can support its operations. Therefore, he believes in the story of AI + terminal and firmly promotes this strategy.

Currently, the input of large model companies is not proportional to their returns. According to the financial reports disclosed by listed companies Zhipu AI and MiniMax, in the first half of this year, Zhipu AI's R&D expenditure reached 2.131 billion yuan, 2.2 times its revenue, while MiniMax's R&D cost was 297 million US dollars, about 2.5 times its revenue in the same period. Both companies are currently in a loss status: Zhipu AI recorded a net loss of 2.072 billion yuan in the first half of the year, and MiniMax posted a net loss of 358 million US dollars for the period (after excluding share-based payment and other items, the adjusted net loss was 293 million US dollars).

The aforementioned former executive of the leading Beijing-based large model company mentioned that currently all large model companies are living a "tough" life, because they spend too much but earn very little. Under such circumstances, selling hardware is another monetization channel that can generate revenue.

For large model companies, selling one device means locking in model usage, system entry point and subsequent Agent services at the same time. Especially for startups, hardware is one of the ways to bypass the disadvantage of insufficient user scale.

Compared with Doubao, Qwen, Kimi, DeepSeek and other similar products, Stepverse has no advantages neither in the comprehensive competitiveness of its model nor in the scale of C-end users. It is very difficult for it to squeeze into users' mobile phone desktops by only developing a standalone AI App. A device that can directly deliver the intelligent agent to users' hands gives Stepverse a chance to be selected by users.

However, up to now, no large model company has announced the exact sales volume of its AI hardware. Compared with the shipment volume of traditional hardware which often reaches tens of millions or hundreds of millions of units, the absolute figure of this new track is still very small.

The reporter experienced the products on site, and found that both Qwen and Stepverse's AI hardware still have shortcomings. When using the AI headphone for real-time translation on site, the accuracy rate was not high, and when using the AI smartphone to order takeout, the task was not completed even after several minutes.

The Doubao phone also encountered the same problem when it was launched. Some media tests asked the Doubao phone to "help me order a cup of coffee", but the external e-commerce platforms such as Taobao Flash Purchase and JD could not be directly operated by the phone's Agent, and finally the AI could only redirect the order to Douyin's services under ByteDance.

This article is from the WeChat Official Account "Economic Observer", author: Ren Xiaoning, published with authorization from 36Kr.