---
title: "10x Revenue in Six Months, ARR Approaching RMB 1 Billion: How Kuaizi Grew on Token-Driven Demand"
author: "Unique Research"
sourcePublication: "Unique Research Substack"
originalPublishedAt: "2026-09-30T09:13:49+00:00"
canonical: "https://ffcap.cn/en/research/10x-revenue-in-six-months-arr-approaching"
source: "https://uniqueresearch.substack.com/p/10x-revenue-in-six-months-arr-approaching"
language: "en"
---

# 10x Revenue in Six Months, ARR Approaching RMB 1 Billion: How Kuaizi Grew on Token-Driven Demand

[![cover](https://substackcdn.com/image/fetch/$s_!aVK6!,w_1456,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0c80c03b-a8c3-4707-b778-6aac1b0c3efb_2730x1536.jpeg)](https://substackcdn.com/image/fetch/$s_!aVK6!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F0c80c03b-a8c3-4707-b778-6aac1b0c3efb_2730x1536.jpeg)

_AI Industry Watch_

“We’re no longer betting on the declining delivery costs of the software era. We’re betting on the exponential explosion of AI intelligence consumption in the token era — and so far, it feels like the bet was right.”

That was the opening line from Winder, founder of Kuaizi.ai (筷子科技), when I sat down with him after this year’s Alibaba Cloud Apsara Conference.

The talk he gave at the conference was titled “The Tokenization Leap of Video Commerce” (视频商业的Token化跃迁) — built around two words that have come up frequently lately: video and tokens.

As our conversation continued, it quickly moved to how enterprises are actually using AI, and how an AI application company’s revenue really grows.

During the exchange, Winder shared Kuaizi’s latest operational numbers: the company’s annualized revenue (ARR) has now approached RMB 1 billion — roughly 10x what it was six months ago.

He expects that figure to reach RMB 2 billion by the end of this year. In addition, comparing September against January, Kuaizi’s platform-wide token usage has surged roughly 200x.

A ~10x change in revenue run-rate over six months, plus a 200x jump in token consumption within the year, makes this enterprise AI application company — which already had some foundation — worth getting to know all over again.

Kuaizi enters from the video commerce scenario of enterprises, organizing models, compute, tools, commercial assets, and distribution capabilities to serve companies’ day-to-day video business.

In recent years, as generative AI entered content production, it has kept integrating these capabilities into its video commerce agent system.

This made me want to understand one thing: when enterprises start letting AI continuously execute real business work, what happens to an application company’s growth?

To understand Kuaizi, start with one very concrete scenario.

A chain brand launches a new product: headquarters needs standard assets, stores need local versions, e-commerce teams need to test different selling points, and overseas teams need to handle language and cultural differences.

The same product can end up corresponding to hundreds of content versions.

And what’s done this week has to be done again next week. New products, new campaigns, new channels — tasks keep being generated.

In the past, this work was scattered across different teams, vendors, and software.

After introducing AI, scripting, generation, editing, and distribution can gradually be integrated into the same workflow, and enterprises begin purchasing the ability to execute these tasks on an ongoing basis, according to business needs.

This adds a new dimension to an application company’s growth: the same customer can go from one team using it to many teams, from one scenario to multiple business lines, continuously increasing usage.

Kuaizi’s revenue space has accordingly expanded from software procurement fees into the business budgets corresponding to content production, marketing operations, and agent execution.

What enterprises hand over to the platform is a portion of work that needs to be done continuously — plus the budget for getting that work done.

Understanding this round of Kuaizi’s growth is impossible without mentioning this year’s overall advance in video foundation models.

Across the first three quarters of 2026, led by Seedance 2.5, Wan3.0, and MiniMax H3, video models continued to improve in multimodal input, audio-visual coordination, long shots, and subject consistency.

Seedance 2.5 can generate videos up to 30 seconds long with multimodal reference support; Wan3.0 puts text, images, video, and audio into a single generation model; MiniMax H3 supports native audio-visual generation and 2K output.

AI video is getting closer and closer to what enterprises need for controllable production.

In Unique Research’s August 2026 Overseas AI Web Rankings, the 18 video-generation products tracked posted combined traffic of 54.92 million visits, down 4.3% month-over-month, with 11 of them declining; video editing products fell 10.3% overall.

Seedance 2.5 launched on Dreamina at the end of July and rolled out globally in early August — yet Dreamina’s monthly web traffic still fell 30.3% month-over-month; likewise, in August, when Wan3.0 launched, Wan AI’s traffic fell 20.4% month-over-month.

Models are getting stronger, but video tools’ web traffic isn’t rising in sync. One important shift: the model and the application entry point are separating.

The same model can be used inside an official product, or called through APIs, aggregation platforms, or an enterprise’s own workflow.

For enterprises, the underlying model is becoming an interchangeable capability component that can be switched based on quality, speed, and cost.

This precisely explains the value of Kuaizi’s niche.

Upstream keeps releasing stronger models; what downstream enterprises actually need is to connect those models with brand assets, licensed resources, business processes, and distribution channels — then keep track of costs and measure results.

Through model routing, video commerce agents, and the KP wallet, Kuaizi organizes the capabilities of different models into services that enterprises can purchase, call, and manage on an ongoing basis.

Every step forward in model capability gives it a chance to take on more tasks and enter more scenarios, converting model progress into token consumption and business revenue.

The market is already pricing this layer.

In August this year, Stripe announced the acquisition of model gateway and routing platform OpenRouter, with media reporting a transaction value of over $7 billion.

And just this May, when OpenRouter completed a $113 million Series B round, its valuation was around $1.3 billion.

It doesn’t train its own foundation models, yet it connects more than 400 models and 80+ providers, handling over 10 trillion tokens a day.

Model routing, unified calling, and cost management are moving from developer tools to a new layer of AI infrastructure.

There’s another important point.

Unique Research’s rankings only count direct web access — they don’t include API calls, desktop usage, or private enterprise deployments, and a significant share of enterprise AI consumption happens in those places invisible to public traffic.

Kuaizi’s ~10x annualized revenue change in six months and ~200x token consumption within the year at least demonstrate one thing: web traffic alone can no longer explain the real growth of this kind of enterprise AI application company.

In September this year, Kuaizi launched its enterprise KP wallet.

KP is defined as a unit of video commerce AI value consumption built on top of tokens.

Model tokens, inference compute, agent execution, tool calls, and even traffic and IP resources can all be measured and managed uniformly within this system.

For enterprises, the immediate change is: through one account and one wallet, they can use different video commerce AI capabilities and allocate budgets across different teams and tasks.

What this solves is the management problem that appears when AI is used at scale.

A few people trying AI can each buy a couple of tools and get started.

Once headquarters, stores, e-commerce, and overseas teams are all using it, enterprises need to know: who can use what, what the quotas are, which tasks money is being spent on, and which scenarios deserve continued investment.

KP puts these questions into one management system. As the models and tools behind it keep updating, enterprises can keep using the same accounts, permissions, and budget arrangements.

The key is that this wallet connects to a system that actually executes work: the Lijing Engine (丽帧引擎) handles enterprise-grade compliant AI video generation, the Kuaizi main platform covers “script, shoot, edit, publish, manage,” the international version serves cross-market content production, and “Tuini” (推你) connects to a KOC content distribution network.

From planning and generation to distribution, then back to performance feedback and the next round of optimization, budget can be used along the same business pipeline.

KP’s significance is letting enterprises purchase and manage video commerce AI as an everyday capability.

Kuaizi had already been building toward a model of usage-based and AI-consumption-driven growth.

The wallet launched in September further unifies capability consumption across products, providing a clearer entry point for scaling this model.

Why is video suited to this model?

Because commercial video demand tends to have three characteristics at once: high frequency, many versions, and the need for continuous optimization.

An enterprise’s video needs depend on its number of products, stores, markets, and content update cadence.

Once business expands, the tasks to be done expand with it.

After AI lowers the production threshold for each piece of content, enterprises can cover more versions they previously couldn’t get to, test more selling points, and serve more stores.

Unit cost falling and overall usage expanding can happen at the same time.

According to customer cases provided by Kuaizi, a tea beverage brand’s daily video output rose from 20 to more than 80, asset reuse rates improved by 50%, and click costs fell 18%.

Another 3C team focused on Southeast Asia raised daily output from 50 to 500 pieces, with cart-attach conversion climbing from 1.2% to 6.8%.

These cases show a path enterprises follow in adopting AI: first raise content supply capacity, then find effective content through distribution and testing, and finally keep investing resources in the scenarios that work.

When this process repeats across more products, channels, and teams, the platform’s usage has sustained room to grow.

This is also the key to understanding token-driven growth.

Behind token consumption there must be real business tasks; whether customers are willing to expand budgets depends on the value those tasks create.

In this model, customer count determines coverage, while the depth of usage and business expansion of the same customer further determine the revenue ceiling.

A ~10x change in annualized revenue over six months makes Kuaizi a concrete sample for observing this growth mechanism.

“We’re no longer betting on the declining delivery costs of the software era. We’re betting on the exponential explosion of AI intelligence consumption in the token era — and so far, it feels like the bet was right,” Winder said, adding that this shift defines the leap in business model between the software era and the token era.

For large enterprises, using a general-purpose platform is often just the beginning.

How brand assets are integrated, how existing systems connect, how departments collaborate, how data and permissions are managed — all of this affects whether AI can truly enter daily operations.

Kuaizi’s other new business, FDE (enterprise video commerce agent deployment), is built around exactly these issues.

It connects models, Skills, tools, data resources, and agent engineering capabilities into enterprises’ specific processes, helping customers build their own video commerce agent environments.

You can think of it this way: an enterprise begins to own an AI production system that can continuously execute content tasks.

The team handles goals, rules, and judgment, while agents take on more and more of the work that can be standardized and executed in batch.

For Kuaizi, FDE’s value also extends beyond deployment.

Once the system enters daily business, it has the chance to drive continued capability consumption and service expansion as new scenarios are connected and new teams start using it.

Platform products make it easy for enterprises to begin using AI; FDE helps large customers integrate into deeper business processes; KP centrally manages the capability consumption within them.

Combined, the three create a path forward between “starting to use AI” and “using AI across the whole business.”

By this point in the conversation, my understanding of Kuaizi had moved forward another step.

Unified billing certainly has value. But the reason enterprises are willing to keep budgets here, ultimately, is that the system can get the work done — and, over subsequent use, become better suited to them.

Model choices and combinations need to adapt to the quality, speed, and cost requirements of different tasks.

Brand assets, copyright licenses, and industry experience need to be used correctly in the production process.

After content goes out, you need to know what worked and how to adjust the next round.

Only when these links connect do you get a workflow an enterprise can use over the long term.

As usage deepens, an enterprise’s assets and rules can gradually accumulate in the system, and performance data can be used for later optimization.

How much of this accumulation a company can generate determines the long-term value of an application company.

According to the company, Kuaizi has served 15,000 enterprises cumulatively, with over 130 million commercial videos generated and distributed, covering scenarios including chain restaurants, retail, e-commerce, and overseas expansion.

These business touchpoints provide the foundation for organizing general AI capabilities into industry workflows.

As model capabilities continue to improve, more tasks will meet the conditions for being handed to AI.

For a company like Kuaizi, the opportunity lies in quickly plugging new capabilities into customers’ real businesses and turning them into services customers are willing to keep purchasing.

Entering from a specific scenario also gives direction to future technical buildout.

Kuaizi already has a clear compute capacity plan, serving enterprises’ needs for token quality, cost, speed, and deployment models.

Enterprise scenarios and model/compute capabilities reinforce each other, and Kuaizi’s technical and business space can keep expanding along customers’ real needs.

Winder’s expectation that Kuaizi will enter the RMB 2 billion annualized revenue range by year-end leaves a clear time coordinate for this exchange.

What’s worth watching next: whether more enterprises move from trials to sustained use, whether the same customer expands into more scenarios, and whether FDE delivery results in business systems that run stably.

These developments will make the foundation of growth more solid.

In this year’s discussion of token commercialization, Kuaizi provides a sample worth studying seriously: enterprises can continuously purchase business capabilities — delivered jointly by models, tools, and agents — through unified budgets; and an application company’s revenue can grow with the expansion of real task volume.

Looking back at the title “The Tokenization Leap of Video Commerce,” it describes more than just a billing method.

It describes a new way of organizing video commerce: enterprises set goals and allocate budgets, agents execute tasks, and performance data flows back into the next round of production.

For investors focused on AI applications, this also broadens the coordinates for evaluating a company: beyond customer count, look at how many real tasks customers hand to the system, and whether those tasks keep generating new usage and budget.

Whoever can make AI continuously complete real business work has the chance to capture continuously expanding enterprise budgets. Kuaizi’s ~10x growth over six months is turning that opportunity into visible, sustained revenue at scale.

* * *

_This article was adapted from the original Chinese feature by Unique Research (非凡产研), published September 29, 2026. Original title: 半年10倍，ARR已近10亿元：这家视频商业智能体公司跑出Token驱动的增长. Company-reported figures are attributed to Kuaizi and its founder Winder as stated in the source._

---

Original publication: https://uniqueresearch.substack.com/p/10x-revenue-in-six-months-arr-approaching
On-site reading page: https://ffcap.cn/en/research/10x-revenue-in-six-months-arr-approaching