China's AI fashion machine: agents that source, design and sell at scale

In March 2026 Alibaba turned Accio, its AI search engine for wholesale buyers, into Accio Work, an agent platform that can draft requests for quotation and haggle with factories over several rounds. Alibaba says it now has more than 10 million monthly active users. In July it added a sourcing toolkit that contacts several suppliers at once, chases unanswered messages across time zones and negotiates not just price but sample costs, lead times, minimum order quantities and shipping terms.

That is the clearest sign of a broader shift. In China, generative AI is no longer a pilot sitting in an innovation lab. It runs on the marketplaces where brands sell, inside the factories that make their clothes and in the livestreams that move their stock. During the Double 11 festival in autumn 2025, JD.com's AI avatar hosts streamed for five million hours and generated 2.3 billion yuan in merchandise value, according to Caixin.

The direct answer to what China's AI fashion machine looks like: platforms supply the models and agents, often free or bundled into merchant tools; suppliers use them to cut sampling and listing time; brands such as Anta and Bosideng are building their own design and styling models; and the state sets adoption targets while regulating labels. For Western brands, which still buy heavily from Chinese suppliers and compete with Shein and Temu, the question is no longer whether China uses AI in fashion, but how far its speed and cost advantages will travel.

What is China's AI+ policy and why does it matter for fashion?

In late August 2025 China's State Council issued its guideline on the "AI+" action. It sets a target for the penetration of new-generation intelligent terminals and AI agents to surpass 70% by 2027 and exceed 90% by 2030, with AI to become a significant driver of what Beijing calls the intelligent economy. The document does not single out textiles or apparel; it talks of industrial development and consumption upgrades. But the signal to platforms, local governments and manufacturers is unambiguous: AI adoption is a policy priority, not a private experiment.

The state has paired that push with control. Rules issued jointly by the Cyberspace Administration of China and three other authorities took effect on 1 September 2025. They require both explicit labels, such as watermarks and notices, and hidden labels in metadata on AI-generated content. Douyin asks creators to declare AI use and tags undeclared material as suspected AI content; SenseTime labels its digital humans as AI-generated. For brands producing AI imagery, virtual models or avatar hosts for Chinese channels, labelling is now a compliance task, not a choice. The pace of change since the summer of 2025 has been striking.

How is Alibaba using AI agents to change B2B sourcing?

Alibaba.com's argument is that wholesale buying suits agents better than consumer shopping. Zhang Kuo, the platform's president, told Caixin in November 2025 that B2B buyers weigh more than 20 factors when choosing a supplier, against about five in consumer purchases. AI Mode, announced at a London event that month and rolled out from December, was designed to cover the workflow from interpreting a buyer's requirements to comparing factories and generating trade documents, potentially cutting sourcing cycles from several weeks to under an hour.

The commercial context matters for European buyers. Alibaba.com reported that European order volume rose 57% year on year between April and October 2025, and that Europe now trails only the United States on the platform. Zhang attributed that to matching and localisation (local payment options, regional logistics, more local suppliers) rather than macroeconomics.

The supplier side is moving too. Caixin cited Guangzhou Healy Apparel, run by Lin Zhilie, which earns more than 60% of its revenue in Europe, about USD 8 to 10 million a year, and uses AI for design, marketing and customer service; the company reports annual growth above 30%. A slipper maker on Alibaba.com, Quanzhou Divitanci, reportedly lifted its conversion rate from 9.96% to 21.67% and cut product image work from a month to two days. These are company and platform claims, not audited results, but they show where the gains are claimed: content, speed of response and matching.

Alibaba is careful about one point. The Accio sourcing agent cannot close deals on its own: users must approve pricing, commercial terms and purchase orders, and payments need explicit confirmation. "The founder still makes the decision," Zhang wrote. The design keeps legal responsibility with humans, but it also means a Western buyer may soon face a supplier whose agent answers instantly, remembers every prior concession and cites rival offers.

What AI tools do Taobao and Tmall merchants actually use?

On the domestic marketplaces the scale is larger still. Alibaba says its generative content suite for Taobao and Tmall merchants produces about 200 million images and 5 million videos a month, including videos built from a few static product photos, and that together these lift click-through rates by more than 10%. Its AI Business Advisor has produced 10 million analytical reports with suggestions on dynamic pricing and inventory, and AI enriched and indexed more than two billion product listings.

Caixin reported that Alibaba's AI business manager reached about 13 million merchants by mid-November 2025, reportedly cutting manual work by 30%, and that its AI customer service handled about 300 million queries between 15 October and 2 November, resolving about 100 million without human support. Tmall says an AI engine distributed 50 billion yuan in coupons and lifted conversion by 15%. For apparel shoppers, Taobao also offers virtual try-on: a single ordinary photo creates an avatar dressed with fabric texture and drape, though Alibaba published no performance figures for it.

The next step is autonomy. Alibaba reported that more than 230,000 businesses had deployed its AI agent teams by April 2026, and its Business Advisor already suggests pricing and inventory moves to merchants. Tmall says AI-managed promotions improved return on investment by more than 30% year on year for over one million products during Double 11. For fashion brands that sell through Tmall, the more autonomy these tools gain, the more pressing a practical governance question becomes: who sets the floor price, the markdown depth and the voucher budget when software adjusts them overnight, and who is accountable when a discount erodes a brand's price positioning?

How do Shein and its Guangzhou suppliers use data and AI?

Shein's model is the reference point for what data-driven supply looks like at scale. In disclosures around its Hong Kong listing, summarised by KrASIA in July 2026, the company described a large-scale automated test and reorder system: thousands of new products launched daily, initial orders of 100 to 200 pieces, and replenishment triggered by sales data without manual approval. Shein reports inventory turnover of 36 days, against more than 120 days at many traditional apparel retailers, and slow-moving stock in the low single digits as a share of products, against 20% to 30% in the traditional industry.

Much of this is automation and analytics rather than generative AI, and Shein itself frames it as digital infrastructure. By the end of 2025 it had developed more than 180 tools and delivered 7,500 units of smart equipment to suppliers, with average efficiency gains of 35% in related processes, according to the same report. A Panyu factory owner, Wang Kai, said he can now test a design's popularity in about ten days; before, "customers usually placed orders five or six months in advance."

The network is dense. Founder Xu Yangtian said in February 2026 that Shein works with nearly 10,000 suppliers in Guangdong, supports more than 600,000 jobs in the province and turns designs into deliveries in two to three weeks; he pledged more than 10 billion yuan over three years to build a "smart supply chain system" there. Tariffs complicate the picture. In April 2025 Reuters reported that factory owners in Guangzhou's so-called Shein villages saw orders fall as work moved to Vietnam; one said orders were down 50%. Shein called reports of capacity leaving China untrue and said its Chinese supplier base had grown to 7,000 from 5,800.

Are Chinese brands such as Anta and Bosideng using generative AI in design?

Yes, and they are building their own models rather than only buying platform tools. Anta Sports, the Chinese sportswear group whose portfolio includes Fila, Descente and Kolon Sport, is pursuing what it calls an "AI 365" strategy. According to KrASIA's account of its first half 2026 results, its Linglong design model turns a sketch into a design rendering in about 15 seconds, while a styling model called Lingxi recommends outfits and offers virtual try-on in stores. Anta also uses AI to analyse social media and consumer trends for forecasting and inventory planning. Revenue rose 12.9% to 43.5 billion yuan in the half. Co-CEO Lai Shixian nonetheless described the initiative as early stage, saying its value will show in better products, faster response, healthier inventory and a better consumer experience.

Bosideng, the down jacket maker marking its 50th anniversary, says it has built a vertical large model, the "Bosideng AI Aesthetic Brain", covering design through fitting and delivery. Chairman Gao Dekang told a Beijing forum in July 2026 that its Wangfujing flagship would gain AI fitting rooms, shopping assistants and personalised recommendations this year. "AI is profoundly restructuring retail scenarios," he said. Neither company has published hard figures on design cycle time or sell-through gained from these models, so the effect on margins remains unproven.

Do AI livestream hosts really sell fashion in China?

They sell, though the strongest evidence so far comes from electronics and groceries rather than clothing. In June 2025 an avatar of Luo Yonghao, the Smartisan founder, co-hosted a six-hour stream on Baidu that drew more than 13 million views and over 55 million yuan in gross merchandise value. JD showed a digital version of its founder Liu Qiangdong, whose stream drew about 50 million yuan in sales. At the platform level, Caixin reported that JD's avatar livestreams ran for five million hours during Double 11 2025 and generated 2.3 billion yuan at about one tenth of the cost of human hosts.

The economics are the point: an avatar can stream 24 hours a day, does not get tired and carries no risk of a celebrity scandal. For mid-market apparel sellers with thousands of SKUs, that turns livestreaming from an expensive event into a permanent sales channel. Analysts quoted by China Daily were more cautious, however. Mo Daiqing of the Internet Economy Institute said virtual hosts handle repetitive work but "will not totally replace real human hosts", because emotional connection with fans is hard to build. Fashion, where fit and styling advice matter, may be one of the harder categories to automate.

What are the risks and limits of China's AI fashion push?

First, demand. Caixin cited an Accenture survey from November 2025 in which only 37% of Chinese consumers had used AI for shopping, mostly for comparisons, and most worried about privacy, reliability and a lack of personalisation. Supply has run ahead of consumer behaviour. Second, returns on investment: Morgan Stanley downgraded JD in November 2025, arguing that its efficiency-focused AI gave no clear near-term edge in user growth or profitability. Third, measurement: most figures in this article are platform or company claims, not audited outcomes, and few disclose results for apparel specifically.

There are structural risks too. Labelling rules raise the cost of mistakes with synthetic models and avatars. US tariffs are pushing parts of Shein's supply base towards Vietnam, which may dilute the cluster advantages that make Guangzhou fast. And as Accio-style agents spread, negotiations may become more transparent and more aggressive at once, compressing margins for suppliers that cannot compete on speed. For Western brands the main risk is complacency: assuming that AI capability in China is about cheap content when it increasingly sits in sourcing, pricing and replenishment.

What should Western fashion companies do now?

The practical response is not to copy China's platforms, which Western brands do not control, but to understand where Chinese AI capability touches their own business: in the suppliers they buy from, the marketplaces they sell on in China, and the competitors that set price and speed expectations for their customers. The evidence so far suggests the biggest gains sit in sourcing coordination, content production and replenishment, not in fully automated design. That is where buyers, merchandisers and supply chain teams should focus first, with clear rules on what software may decide and what stays with people.

1. Map your supplier AI exposure Ask key Chinese suppliers which platform agents and tools they use for quoting, sampling and content, and how outputs are checked.
2. Prepare for agent negotiations Define floor prices, MOQ limits and approval rules in advance so your team is not outpaced by an agent that answers instantly.
3. Benchmark against small batches Compare your own initial order sizes and replenishment speed with the 100 to 200 piece test model before assuming your lead times are competitive.
4. Govern marketplace repricing If you sell on Tmall or JD, set written rules on what platform agents may change on price, vouchers and service.
5. Label synthetic content by default Treat China's labelling rules as the baseline for AI imagery, virtual models and avatar hosts in Chinese channels.
6. Ask for evidence, not demos Require sell-through, return rate and margin data before scaling any AI design or livestream pilot.