China's internet majors poured RMB 4.5 billion (US$620 million) into an AI marketing offensive during the 2026 Lunar New Year holiday, deploying red-packet giveaways, subsidized consumer purchases and live-event sponsorships to drive user adoption at scale. The campaign produced record daily active user figures across competing platforms — but also exposed a stark divergence in strategic execution, raising fundamental questions about whether subsidy-driven growth can translate into durable behavioral change.

The holiday blitz generated headline numbers that masked uneven outcomes. ByteDance's Doubao peaked at 145 million daily active users on New Year's Eve, while Alibaba's Qwen surged 940% from pre-campaign levels to reach 73.52 million DAU. Tencent's Yuanbao, despite a RMB 1 billion outlay, slipped to 18th place on the App Store free chart before recovering post-holiday. The divergence underscores a core tension now defining China's AI competitive landscape: the difference between capturing a moment and capturing a habit.

The campaign's broader significance extends well beyond market-share arithmetic. Coming months after DeepSeek's global breakthrough — achieved without a single yuan in consumer marketing — the Spring Festival spending spree crystallizes a distinctly Chinese AI commercialization path: one that prioritizes ecosystem penetration and behavioral conditioning over pure technical differentiation. Whether that path leads to lasting competitive advantage, or merely expensive user churn, is the question the industry must now answer.

Doubao Wins the Rankings; Qwen Wins the Argument

On the conventional scorecard, ByteDance's Doubao emerges as the campaign's clearest winner. An estimated RMB 1.1 billion in combined Spring Festival Gala sponsorship and red-packet distribution secured App Store top ranking and nearly 1.9 billion visits on New Year's Eve alone. The simultaneous showcase of its Seedance 2.0 video generation model added a technical dimension to what was otherwise a mass-market activation, reinforcing the "Doubao equals AI assistant" positioning in consumer consciousness.

Qwen's campaign, however, may prove more strategically consequential. Alibaba deployed approximately RMB 3 billion — the largest single allocation in the field — not to demonstrate model capability but to embed AI into transactional behavior. Users placed nearly 200 million orders through voice commands, redeemed 55.2 million free milk tea cups, and purchased more than 3,000 metric tons of eggs via AI-assisted purchasing. The 940% DAU surge is a function of genuine behavioral engagement, not passive app downloads.

Wu Jia, president of Alibaba's Qwen consumer business unit, framed the milk-tea moment in explicitly historical terms: "This is the first time AI has genuinely participated in a consumer's real purchasing decision and execution chain." The claim is commercially self-serving, but the underlying data lends it credibility.

Yuanbao's Stumble Illustrates the Limits of Replication

Tencent's Yuanbao offers the campaign's most instructive cautionary tale. Armed with RMB 1 billion — equivalent to Doubao's total spend — and the implicit advantage of WeChat's 800-million-plus user base, Yuanbao entered the holiday as the presumptive dark horse. It exited having slid to 18th on the App Store chart, recovering only after workers returned from holiday.

The failure was not primarily financial. Yuanbao's design logic attempted to replicate the viral mechanics of WeChat's landmark 2014 red-packet "shake" feature — a social gesture that succeeded because it mapped naturally onto existing group behavior. Asking users to involve an AI chatbot in group-chat New Year greetings inverted that logic: it introduced friction into a social ritual that users were already seeking to simplify. The post-campaign chart recovery, driven by returning office workers rather than organic enthusiasm, confirmed that Yuanbao had captured attention without converting it into habit.

The episode illustrates a principle that the entire campaign reinforces: AI adoption at scale requires precise scenario fit, not merely generous subsidy.

The Real Prize Is Behavioral Real Estate, Not Download Counts

Industry participants and platform executives have been explicit that the Spring Festival was never primarily a competitive battleground — it was an infrastructure investment. The holiday represents one of the few moments in the Chinese calendar when social, commercial and entertainment activity converge at national scale simultaneously, creating what amounts to a natural laboratory for mass behavioral experimentation.

A feature that might require six months to reach 100 million users in ordinary conditions can achieve that threshold overnight during the Spring Festival window. For AI platforms still in the market-education phase, that compression of adoption timelines justifies significant upfront expenditure. Wu Jia's formulation — "AI is not matured overnight; it is refined through real-world use" — reflects a genuine operational philosophy, not just post-hoc rationalization.

The campaign also served as a stress test that exposed real technical vulnerabilities: Doubao experienced failures in holiday voice-message delivery; Qwen's systems were overloaded to the point of requesting users return the following day; Yuanbao's guardrails were breached under adversarial prompting. Each failure is, in the longer arc, a data point that accelerates system hardening.

The deeper strategic logic is about redefining what a red packet means. In 2026, the red packet became not just a cash transfer mechanism but an onboarding event — the moment a user first instructed an AI to perform a task on their behalf. Platforms are not competing for session time; they are competing for the first position in a user's problem-solving hierarchy: "When I need something done, which AI do I reach for first?"

Marketing Spend and Capex Are Not in Conflict

A surface reading of the Spring Festival campaign invites a familiar critique: that Chinese AI companies are substituting marketing expenditure for genuine technical investment, pursuing growth metrics that will evaporate once subsidies end. The critique misreads the data.

Alibaba's AI-related capital expenditure reached approximately RMB 120 billion in 2025, with the company reportedly considering raising its three-year AI infrastructure and cloud computing budget from RMB 380 billion to RMB 480 billion. ByteDance's AI capex ran to approximately RMB 160 billion over the same period; Tencent's to approximately RMB 100 billion. These are not the investment profiles of companies treating technology as a secondary priority.

The more accurate framing is that marketing spend and infrastructure investment serve different functions within the same strategic architecture. China's three dominant internet platforms — Alibaba, ByteDance and Tencent — each operate as what might be called "super-portals": vertically integrated ecosystems that encompass commerce, social interaction, entertainment and payments within a single user relationship. In this environment, AI's most efficient adoption pathway is not the creation of a new standalone entry point — the OpenAI model — but rather the embedding of AI capability within existing super-portals. That embedding requires both the technical capacity to deliver and the behavioral conditioning to make users reach for AI instinctively.

The Spring Festival campaign was the behavioral-conditioning half of that equation.

A Different Path, Not a Lesser One

The contrast with the American AI development model is real but should not be read as a hierarchy. OpenAI's rise was predicated on deep technical differentiation, a globally addressable market and a mature consumer willingness to pay for software subscriptions. China's AI incumbents operate in a market where technical parity among major players is high, where consumer monetization through direct subscription is less established, and where the competitive advantage lies in ecosystem integration rather than model supremacy.

These are different constraints, not inferior ones. The Chinese internet industry's track record — Alibaba, ByteDance and Pinduoduo（拼多多）each emerged from intensely competitive domestic markets before expanding internationally — suggests that domestic competitive pressure, however commercially brutal, can be a forcing function for operational capability. The same dynamic that produced DeepSeek's cost-efficient model architecture, and Seedance as a Sora-competitive video generation system, is operating in the consumer AI market.

The RMB 4.5 billion spent across the 2026 Spring Festival was not a substitute for technical ambition. It was an attempt to solve a different, equally difficult problem: how to make AI indispensable to 1.4 billion people before the window for defining default behaviors closes. The answer to whether it worked will not be visible in the App Store rankings from February. It will be visible in whether, six months from now, users who first asked an AI to order milk tea during the Spring Festival are still asking AI to help them with something else entirely.