China’s biggest tech firms turned the 2026 Lunar New Year into a marketing battleground, spending roughly ¥90–100 billion on cashback, vouchers and gift campaigns to seed standalone AI apps. On the peak night of the Spring Festival — February 16 — ByteDance’s Doubao reached 145 million daily active users (DAU), Alibaba’s Qianwen 73.5 million at its peak and 55.1 million on the night, and Tencent’s Yuanbao surged to 40.5 million. The fireworks were dramatic, but the firework‑effect was short: by February 23 Doubao settled at 103 million DAU, Qianwen at 32.5 million, while Yuanbao collapsed back to about 7.7 million, roughly its pre‑campaign level.
The three apps adopted distinct go‑to‑market tactics that explain much of the divergence. Yuanbao leaned on social virality and a cash‑splitting model aimed at leveraging WeChat/QQ social graphs; Qianwen subsidised transactions with a headline “30 billion free orders” and attached AI to a commerce loop; Doubao partnered with the CCTV Spring Festival Gala and combined cash red envelopes with existing user activation to revive dormant accounts. The result was that the same stimulus — money — produced sharply different retention outcomes, reflecting product positioning and the depth of behavior change each app could trigger.
Unit economics underscore the challenge. DataEye and QuestMobile calculations put the marginal cost to acquire each peak‑period daily active user at between roughly ¥69 and ¥144 across the three apps: Yuanbao about ¥69, Doubao ¥84–113, and Qianwen roughly ¥144. By comparison, established news or social apps typically acquire new users for single‑digit to low‑double‑digit yuan amounts, and tools and services on iOS average around ¥15 per new user. Those discrepancies matter because the post‑campaign retention a platform can achieve will determine whether the upfront burn converts into a sustainable business.
Beyond headline DAU, engagement metrics tell a nuanced story. Qianwen and Yuanbao saw per‑user session counts spike to 14.4 and 12.1 times per day early in the promotion, but both quickly retraced to at or below pre‑event levels. Doubao’s usage remained relatively stable, helped by its larger incumbent base and the Spring Gala tie‑in that converted one‑off curiosity into recurring interactions, particularly among lower‑tier cities where the Gala’s reach is strongest.
Analysts trace the divergent outcomes to the affordances of AI as a product. Traditional red‑envelope mechanics harness social rituals — gifting and group interaction — which are a natural fit for something like WeChat’s payments and messaging. But AI’s primary value proposition today is a one‑to‑one information or task interface: efficiency, answers and automation. That makes purely social viral mechanics less effective unless the app embeds AI into transactional or routine scenarios that users care about every day.
There are early signs of behavioural reallocation rather than wholesale new demand. QuestMobile data show month‑on‑month increases in time spent with general AI apps (up 22.3% between December 2025 and January 2026) while time in other categories such as search, camera and utility apps declined. That suggests AI is beginning to absorb attention previously scattered across categories, but it has not yet locked users into the kind of high‑frequency, sticky scenarios that define a SuperApp.
The strategic takeaway for China’s tech giants is clear: splashy cash promotions can accelerate awareness and reveal promising user segments, but they do not by themselves create the structural lock‑in of a SuperApp. To progress from episodic usage to indispensability, platforms must bind AI to everyday, high‑frequency activities — commerce, payments, communication and device integrations — and design incentives that channel paid activation into value‑creating habits.
For global observers, the New Year frenzy offers an instructive case study in the transition from model‑led hype to product‑led adoption. The costliest phase of the rollout has been performed in public; the harder work now is product and ecosystem engineering that turns subsidised trial into recurring value. How these three giants fold standalone AI tools into their broader ecosystems will determine whether they spawn new SuperApps or simply reallocate marketing budgets into temporary spikes of attention.