The Spring Festival AI Subsidy War Cost RMB 45 Billion. Did It Buy Brands Anything?

Between late January and February 2026, Alibaba, Tencent, and ByteDance collectively spent an estimated RMB 45-90 billion (roughly USD 6.3-12.6 billion) subsidizing AI app downloads and usage during Chinese New Year. Qwen's DAU jumped from 10 million to 75 million in a single day. Doubao peaked at 145 million daily active users. And by February 23, most of that traffic had already halved. The real question for brands tracking their AI citation share isn't who won the download race — it's whether any of that spending changed who gets recommended.\n\n## What actually happened\n\nOn February 2, 2026, Alibaba's Qwen launched a "RMB 3 billion Spring Festival treat" campaign, bundling free rides through Taobao Flash Sale, Fliggy, and Damai to cover users' holiday spending on food, travel, and entertainment. The pitch was "one sentence gets things done" — order takeout, book a flight, buy movie tickets, all through natural-language commands to the AI, with the bill waived.\n\nThe results were immediate. On February 6, the first full day of the promotion, Qwen's DAU spiked to 58.48 million, up 727.7% from 7.07 million the day before. On February 7, cross-platform DAU (app plus PC) hit roughly 75 million. Qwen briefly topped the Apple App Store free chart in China, overtaking both Tencent's Yuanbao and Doubao.\n\nTencent countered with roughly RMB 1 billion in Yuanbao red envelopes. Baidu took a different approach, leaning on Ernie Bot's existing search integration rather than matching the cash subsidies directly. ByteDance's Doubao, already the largest player, used the period to extend its lead — peak DAU during the campaign window reportedly reached 145 million, versus Qwen's 73.52 million and Yuanbao's 40.54 million.\n\nBut the acquisition math was brutal. Analysts calculated the effective cost per daily active user during the campaign at roughly RMB 144 — far above normal app acquisition costs in China. By February 23, Qwen's DAU had settled around 32 million: still 3.6x its pre-campaign baseline, but roughly half its subsidy-period peak.\n\n## Why this matters for brand GEO, not just app metrics\n\nMost of the coverage of this event has focused on the user-acquisition war between Alibaba, Tencent, and ByteDance. For brands tracking AI visibility in China, the more useful question is what kind of usage that subsidized traffic represents — and whether it changes the calculus for where to invest GEO effort.\n\nThe campaigns were built almost entirely around transactional, task-completion prompts: "order me bubble tea," "book this flight," "buy two movie tickets." These are narrow-intent queries where the AI executes a single action inside a partner app (Taobao, Fliggy, Damai) rather than exploratory or comparison queries where a brand name would surface as a recommendation. A user who redeemed a free bubble tea order through Qwen during Spring Festival was not asking Qwen to compare bubble tea brands — the destination was often pre-selected by the promotion itself.\n\nThat distinction matters because it means the DAU surge does not translate cleanly into more brand-citation opportunities. A model can add tens of millions of daily users overnight and still show no meaningful change in how often it surfaces comparative brand recommendations, because the new usage pattern is procedural rather than discovery-oriented.\n\n## What the data suggests brands should actually track\n\n| Signal | What it tells you | Spring Festival 2026 read |\n|---|---|---|\n| DAU / MAU growth | Reach potential | Real, but partially subsidy-inflated and already reverting |\n| Query intent mix | Whether growth is discovery or task-execution | Skewed heavily toward task-completion during the campaign |\n| Post-campaign retention | Durability of the new user base | Roughly 50% of peak DAU retained by day 17 (Qwen) |\n| Citation frequency in comparison queries | Actual brand-visibility impact | Not reported to have moved meaningfully during the campaign |\n\nThe practical implication: a spike in a model's user base is a necessary but not sufficient signal for reallocating GEO budget. Before shifting content and citation-optimization effort toward a model because its MAU jumped, brands should check whether the growth shows up in comparison-style, brand-relevant query volume — not just raw usage.\n\nThis is a variant of a pattern GEO Hub has tracked before with Kimi's steep 2026 decline: user-base size and brand-citation share do not move in lockstep, and both directions of the disconnect — falling MAU with retained citation relevance, or surging MAU without new citation relevance — are possible.\n\n## Where the subsidy war does matter for brands\n\nThere is one place the campaign should shift brand strategy: the deepening tie between Qwen and Alibaba's commerce stack (Taobao, Fliggy, Damai). Every subsidized transaction ran through an Alibaba-owned checkout, reinforcing Qwen as an execution layer sitting directly on top of Alibaba's retail graph. For brands that already sell through Tmall or Taobao, this strengthens the case for treating Qwen citation-readiness as an extension of Alibaba commerce optimization rather than a separate GEO workstream — product data, reviews, and structured listings on Taobao/Tmall are now more directly upstream of what Qwen can recommend or execute.\n\nFor brands without an Alibaba commerce presence, the subsidy war is close to irrelevant: the surge was concentrated in transactional use cases tied to Alibaba's own ecosystem, not general-purpose brand discovery that would benefit a brand sold primarily through JD, Douyin Shop, or offline retail.\n\n## Takeaway for Brand Marketers\n\nDon't reallocate GEO budget purely on the basis of a subsidy-driven DAU spike. Check three things before treating a usage surge as a signal: (1) whether the growth is retained 30 days out or reverts toward baseline, (2) whether the new usage is discovery/comparison intent or narrow task-execution, and (3) whether the platform's growth is structurally tied to a commerce ecosystem your brand already participates in. For Qwen specifically, brands already selling on Tmall/Taobao have a genuine reason to prioritize citation-readiness there; brands outside Alibaba's commerce ecosystem should treat the February surge as noise, not a durable shift in where Chinese consumers discover brands through AI.\n\nRelated: See how this connects to the broader China AI MAU rankings and brand priority framework, or track live brand scores across all six models on the hubGEO brands page.