Subsidies are back at the center of attention in China’s technology sector, this time through artificial intelligence. What makes the latest push notable is that it comes on top of already heavy spending on AI infrastructure and model development. Alibaba said in February 2025 that it would invest at least RMB 380 billion (USD 54.9 billion) in cloud and AI infrastructure over three years. Earlier this month, Tencent said it spent RMB 79 billion (USD 11.4 billion) on capital expenditures in 2025 and plans to step up AI investment again in 2026.
In the run up to the 2026 Lunar New Year, China’s largest internet companies spent heavily to accelerate adoption of their consumer AI apps. In February 2026, Alibaba said it would spend up to RMB 3 billion (USD 433.6 million) on Qwen promotions during the holiday period. Tencent and Baidu pledged RMB 1 billion (USD 144.5 million) and RMB 500 million (USD 72.3 million), respectively, for similar campaigns for their AI offerings.
Alibaba led the pack, and it did so with a more targeted commercial experiment than a standard chatbot giveaway. Its campaign centered on Qwen’s AI-assisted shopping features, offering users free drinks and other coupons if they completed purchases through the app. The response was immediate. Qwen reportedly logged ten million milk tea and beverage orders on the first day, briefly overwhelming the app and prompting Alibaba to pause parts of the promotion. Alibaba later said the campaign generated more than 120 million consumer orders from February 6–11.
A shopping experiment, not just a chatbot push
What Alibaba is asking users to do also differs from a conventional e-commerce flow. Rather than browse a menu, open multiple tabs, and move through a standard checkout sequence, users can stay inside Qwen, ask for recommendations by text or voice, refine an order conversationally, and authorize payment without leaving the chat interface. This builds on services Alibaba had already integrated into Qwen, including Taobao, Alipay, Fliggy, and Amap. Alibaba said Qwen could complete food and beverage orders directly in conversation through Taobao Shangou, the instant commerce arm of Taobao.
Meanwhile, ByteDance used the year’s biggest broadcast stage to promote Doubao. The app’s holiday growth was helped by its partnership with CCTV’s televised Lunar New Year gala, during which it handled more than 1.9 billion AI-related queries, according to ByteDance.
Tencent, for its part, tied Yuanbao’s push to cash incentives, mainly through “red packets” for users who shared links to the app, while Baidu joined the holiday race with its own RMB 500 million campaign.
All of this echoes an old playbook: spend first, win users first, and work out the economic logic later. The obvious question is why companies would revive that strategy in AI, where monetization remains unresolved even for global leaders. Reuters reported in March 2025 that OpenAI did not expect to be cashflow positive until 2029. The media outlet also said in February that subscriptions alone appear difficult to scale in China, citing earlier Tencent management comments that the user-paid model popularized in the US is hard to replicate locally.
That is what makes Alibaba’s campaign more notable. It appears to be subsidizing not chat for its own sake, but user behavior.
Qwen’s new feature set is built around what is commonly called agentic commerce: using an AI assistant not just to discover products, but to help complete the transaction. Alibaba’s own description of Qwen increasingly frames the app as an interface sitting directly on top of shopping, payments, travel, and local services. Seen that way, one plausible longer-term endgame is not subscription revenue alone, but some form of transaction take rate, commission, or service fee if enough commerce activity can eventually be pulled into the AI layer.
There are at least some early signs that the idea is not purely theoretical. Alibaba said Qwen generated more than 120 million orders in six days, ranked first on China’s iOS app store free chart for six consecutive days, and had surpassed 100 million monthly active users after its public beta launch last November. It also said nearly half of the campaign’s orders came from county and rural users, and that about 1.56 million users ages 60 and older made their first online purchase through the app. Those details are not sufficient to prove durable adoption, but they do suggest Alibaba was testing more than a novelty use case for urban early adopters.
Still, the harder question is what happens after the subsidies end. Subsidies can drive trial, but they may be less effective at sustaining long-term user habits. Qwen, Yuanbao, and Doubao all saw their daily active user counts fall after the Lunar New Year peak.
A cautionary signal from abroad
Alibaba also does not have the field to itself, and the experience of foreign peers offers a cautionary signal. On October 14, 2025, Walmart announced a partnership with OpenAI that would let shoppers complete purchases directly inside ChatGPT through a feature called Instant Checkout. Wired later reported that the feature underperformed, with Walmart executive Daniel Danker saying conversion rates were three times lower for products sold directly inside the chatbot than for products that required users to click out.
Wired also reported that Walmart is now shifting toward embedding its own Sparky chatbot inside ChatGPT and Google Gemini instead. At the same time, Danker told the media outlet that ChatGPT was bringing in new customers at about twice the rate of search engines, suggesting there is real discovery value even if the native checkout layer still needs work.
That matters because it suggests agentic commerce is not a solved business model anywhere. The appeal is intuitive enough: let users move from intent to transaction in one conversational flow. The difficulty is that shopping is not always a one-step action, and consumers often want to compare options, build bundles, revisit decisions, or simply feel more in control than a chatbot-native purchase path allows. Alibaba may be right that agentic commerce can become a viable monetization path, but if so, it will probably take more product iteration, more trust-building, and more habit change than a single holiday blitz can buy.
Qwen also still has ground to make up. In early February, before the holiday promotions peaked, Qwen had fewer than ten million daily active users. During the holiday, it reached 30 million at its peak, still behind Yuanbao’s 50 million and Doubao’s 100 million. Earlier QuestMobile data from the second week of December had Qwen ranked fifth among mainland China’s consumer AI apps, though that snapshot was based on weekly active users rather than monthly active users. Qwen’s holiday surge therefore looks less like a definitive market reshuffle than a sign that Alibaba may have found a differentiated lane within a still unsettled contest.
Viewed from that angle, Alibaba’s RMB 3 billion push is less a subsidy campaign than a test of whether users will adopt a new behavior: not just asking AI for answers, but allowing it to handle transactions. That may yet emerge as a viable path to monetizing consumer AI. For now, however, the campaign says more about where Alibaba wants the market to go than about whether the market is already there.
Note: RMB figures are converted to USD at rates of RMB 6.92 = USD 1 based on estimates as of March 30, 2026, unless otherwise stated. USD conversions are presented for ease of reference and may not fully match prevailing exchange rates.