China’s 2026 Lunar New Year AI War: Billions Spent to Win the Next AI Gateway

China’s 2026 Lunar New Year AI War was not a conventional product launch cycle. It was a coordinated-looking but fiercely competitive push to make AI apps part of everyday life during the country’s most important holiday season.

The major platforms used cash red envelopes, shopping vouchers, luxury cars, gadgets, televised promotions and new model releases to turn curiosity into downloads and repeated use. The immediate payoff was attention and scale; the longer-term goal was to secure a durable consumer and developer position before the market settles.

What the “Lunar New Year AI War” means

The phrase describes the February 2026 contest among China’s leading technology companies to acquire users for their consumer AI assistants and strengthen their model ecosystems. Lunar New Year was a particularly powerful acquisition window because digital red envelopes are already associated with holiday sharing, payments and good fortune.

The campaigns encouraged people to download apps, invite friends, complete AI tasks, place orders and share content. That made the promotions more than traditional advertising: each reward mechanic was also a way to generate activity, distribution and product trials. Alibaba, Tencent, Baidu and ByteDance all used the holiday to put their AI products in front of a mass audience. 238

The main consumer-AI campaigns

ByteDance’s Doubao campaign

ByteDance’s Doubao offered more than 100,000 prizes, including robots, drones, 3D printers and other technology products. Its red-envelope prizes reportedly reached CNY8,888, and the campaign was tied to China’s nationally televised Spring Festival Gala. 345

The strategy produced a measurable short-term result. A private survey reported by Reuters found that Doubao surpassed 100 million daily active users on February 16, around four times its early-February level. That makes Doubao the clearest reported holiday acquisition winner, although daily activity during a heavily subsidized promotion is not the same as durable engagement or profitable usage. 1

Alibaba’s Qwen campaign

Alibaba put roughly CNY3 billion behind its Qwen holiday campaign, using incentives connected to shopping and digital consumption. The company said the promotion generated more than 10 million orders within nine hours, while reports described unusually high demand and service disruption on the Qwen app. 2

That combination is strategically important: Alibaba was not presenting Qwen only as a chatbot. Its campaign connected AI with the company’s existing commerce ecosystem, giving users a reason to interact with the assistant while they were already shopping.

Tencent’s Yuanbao and Baidu’s Ernie

Tencent committed about CNY1 billion to Yuanbao’s campaign, while Baidu allocated about CNY500 million for Ernie-related promotions. Their incentives included digital red envelopes, vouchers and task-based rewards designed to bring users into each company’s AI assistant. 28

The spending gap mattered symbolically as well as financially. Alibaba’s announced commitment was about three times Baidu’s and Tencent’s combined? No: it was three times Tencent’s budget and six times Baidu’s, based on the headline figures. That comparison shows how aggressively the largest platforms were willing to spend, but it does not establish that spending levels translated directly into product quality or long-term market share.

The second front: a rush of new AI models

The holiday battle also played out among model developers. Zhipu AI released its open-source GLM-5 on February 11, highlighting coding and long-running agent tasks. Reuters reported that the company positioned the model competitively against leading international systems on selected coding and reasoning benchmarks. Those are company-reported or benchmark-specific comparisons, not a universal ranking of model quality. 18

MiniMax launched M2.5 in mid-February and emphasized strong performance at a lower cost. Alibaba released Qwen3.5 just before the holiday began. Meanwhile, the industry was watching DeepSeek after the earlier impact of its R1 and V3 models, with a next-generation V4 release widely anticipated at the time. 1720

This release timing reveals the developer side of the competition. Model companies were trying to win attention from engineers, application builders and API customers at the same moment consumer apps were competing for mass adoption. Open-source releases can help models spread through developer communities, while consumer distribution can provide a path to feedback, usage and commercial services.

Why viral video mattered

ByteDance also benefited from attention around Seedance 2.0, its AI video-generation system. A widely shared 15-second clip depicted realistic versions of Tom Cruise and Brad Pitt fighting on a rooftop. The video became a showcase for the quality and speed of Chinese generative-video tools, while also provoking copyright complaints from Hollywood groups over the use of recognizable actors and copyrighted material.

Some reports attributed a short reaction to Elon Musk, but the evidence supplied here is not strong enough to treat that specific reaction as independently verified. The broader point is better supported: Seedance’s viral output created global awareness while simultaneously demonstrating the legal and safety risks that can accompany rapid product exposure.

What companies are really trying to win

The visible prizes were aimed at consumers, but the strategic prize is control of the AI gateway.

An AI gateway is the product people open first when they want to search, write, generate images or video, shop, communicate or delegate a multi-step task to an agent. The company that owns that entry point can potentially direct attention toward its own commerce, advertising, cloud, payment and developer services.

That is why the campaigns combined several forms of competition:

• User acquisition: cash, vouchers and prizes made first-time trials easy to justify.
• Distribution: holiday television and social sharing exposed products to huge audiences.
• Habit formation: repeated tasks and red-envelope mechanics encouraged users to return.
• Developer adoption: new and open-source models competed for attention in coding and agent workflows. 1820
• Ecosystem control: companies with commerce, social, entertainment or cloud businesses could connect AI to existing services.

The risk is that promotional activity can inflate downloads without creating lasting demand. Doubao’s reported 100 million daily active users demonstrate acquisition success during the holiday; they do not, by themselves, show that users stayed after rewards ended or that the app generated sustainable margins. 1

The government’s role: AI as infrastructure and industrial policy

The competition unfolded alongside an official push to commercialize AI at national scale. On February 11, Premier Li Qiang called for better coordination of power and computing resources and faster large-scale, commercial use of artificial intelligence.

China’s 2026 government work report also described faster industrial AI adoption, continued digital transformation of manufacturing and the emergence of new intelligent terminals.

That policy context matters because frontier AI requires more than a popular app. Scaling models and agents depends on access to compute, electricity, data, talent and distribution infrastructure. Government support can therefore influence the competitive environment even when individual consumer campaigns are run by private technology companies.

What happens after the giveaways?

The next test is retention. Companies must show that users continue to use their assistants when prizes, vouchers and holiday attention disappear. They also need to convert usage into revenue through subscriptions, advertising, commerce, cloud services, enterprise software or model APIs.

The same question applies to model launches. A strong benchmark result can attract developers, but durable platform power depends on reliability, cost, tools, documentation and the ability to support real workloads over time.

The Lunar New Year campaigns therefore offer a clear but limited verdict. China’s AI companies demonstrated that they can mobilize enormous promotional budgets, launch competitive models quickly and turn cultural rituals into digital distribution. They have not yet demonstrated which company can retain the users, developers and economics created by that spending.

The “war” was ultimately a race to establish default behavior before the market consolidates. The winners will be determined less by the biggest red-envelope headline than by which products remain useful after the holiday is over.