The startup’s Moras platform helps overseas content creators drive sales through AI agents and targets $10 million in annualized revenue.
By Chen Mei
K2 Lab, the startup founded by Wang Ming, formerly the youngest vice president of Alibaba’s DingTalk workplace super-app, has completed three funding rounds in six months as it builds an AI-powered business platform for overseas content creators.
The latest financing was a tens-of-millions-of-yuan angel-plus round, according to exclusive reporting by Chinaventure. Another round is already in the works.
The company’s personal business agent Moras provides operational services to overseas content creators. Since beginning internal testing in March, K2 Lab has reported that annualized revenue exceeded $1 million in its first public-test month, that its AI helped influencers generate more than $10 million in sales last quarter, and that nearly 100,000 creators have registered.
Crucially, K2 Lab charges neither subscription fees nor fees for AI token usage. Instead, it takes a commission on completed sales. After paying creators their share, the company’s remaining commissions are now sufficient to cover computing costs — a notable milestone in an AI agent market where many companies are still spending heavily to acquire users.
Prove the business model before scaling
K2 Lab’s proposition is straightforward: help overseas content creators make money through product sales.
The process, however, involves a complex chain of tasks, from product selection and content creation to scripting, editing, publishing and performance analysis. K2 Lab automates these activities through a coordinated network of AI agents.
Wang spearheaded DingTalk’s AI native ecosystem and enterprise products before departing in July 2025 to set up K2 Lab. His new venture’s approach differs from most AI agent startups. Rather than burning cash to acquire users and then searching for a business model, K2 Lab first validated whether it could make money, then scaled.
“We first refine the results and establish whether we can actually make money,” he said. “Then we assess whether the system can support large-scale user interactions once we expand. If it can, we will open it up completely.”
Wang argues that AI companies must treat computing power as a production input whose cost needs to be recovered through tangible results. Without that discipline, rising usage can simply magnify losses.
“If you spend computing resources but cannot deliver results, your cash flow will suffer and your gross margins will be poor,” he said. This explains why many AI products, unable to recover results, calculate not gross margin but “gross loss rate” — “because they have no ability to make money, they can only help users consume expensive wholesale tokens and compete over who loses less.”
K2 Lab, by contrast, says its commissions have covered computing costs since June, when it opened the platform to public users and eliminated fixed base payments to its AI agents, moving to a purely commission-based model.
“We deliver a capability, not a tool,” Wang said.
Black Friday targets
K2 Lab is now preparing for the Black Friday shopping season, targeting monthly merchandise sales well above $10 million during the event and annualized revenue exceeding $10 million.
Its strategy combines a multi-agent architecture with a portfolio of specialized AI models. More than a dozen agents work together across lengthy task sequences, while the company is trying to reduce costs both by optimizing its agent architecture and by developing its own models to reduce reliance on expensive general-purpose AI APIs.
But lower costs are only part of the equation. The bigger goal is to improve commercial returns by making AI-generated content more effective.
K2 Lab has developed capabilities in multimodal understanding, business-outcome prediction and model orchestration, alongside externally developed content-generation models.
Its specialized systems analyze successful e-commerce content to identify the elements behind its performance. Smaller predictive models then assess which combinations of scripts, creators and products are most likely to generate sales.
This addresses a common problem in influencer commerce: a creator’s preferred products may not match their audience.
By analyzing large volumes of creator profiles and content data, K2 Lab aims to predict which products, formats and trending topics are most likely to convert.
Wang likens this to a world model — a system that learns underlying patterns and uses them to predict outcomes. While he believes a comprehensive, end-to-end commercial world model will take time to develop, he argues that combining specialized models with long-running agent workflows can deliver useful predictions today.
The accumulated data and industry expertise, he says, also provide a potential defense against larger technology companies. General-purpose models face challenges when handling long sequences of interconnected tasks, while K2 Lab’s sector-specific knowledge and growing body of commercial feedback could give it an advantage.
Funding amid a cooling AI market
Despite completing three funding rounds in six months, Wang says the fundraising environment has been challenging. AI applications were already losing investor momentum when the company launched.
He attributes investor interest to three factors: the team’s technical understanding of AI models and their limitations; its combination of startup experience and big-tech backgrounds; and a business model designed to generate measurable value and reinforce its own growth.
The company is expanding its product offering ahead of Black Friday, with around 100,000 high-quality products available through its platform and a growing network of creators.
It is also developing its Personal Agent service, which allows creators to delegate ongoing account management to AI. An agent can handle product selection, content creation, publishing and performance analysis for an extended period, even while its owner is away on vacation.
After Black Friday, K2 Lab plans to focus on improving its unit economics. Although commissions retained after creator payouts cover computing costs, the company says its overall unit economics and gross margins still need to improve.
In the longer term, Wang wants K2 Lab to become an operating system for creators’ AI agents. He believes tens of millions to hundreds of millions of people could eventually use agents to run their own businesses, creating commercial opportunities and broader social value.
Source:
chinaventure.com.cn
