36Kr Exclusive | Volcengine Raises MaaS Revenue Target to 15 Billion Yuan for the Whole Year, with Seedance 2.0's Monthly Revenue Exceeding 1 Billion Yuan

Text by | Deng Yongyi

Edited by | Zhang Yuxin, Yang Xuan

36Kr exclusively learned that ByteDance's Volcengine raised its revenue target for the MaaS business to 15 billion yuan in April this year, and this revenue target has been increasing almost every month. At the end of 2025, this target was still 10 billion yuan.

 36Kr learned that in 2025, Volcengine's annual MaaS revenue was about 1.5 billion yuan. That means the revenue target for 2026 so far is 10 times the actual revenue of last year. It can be said that almost all of Volcengine's MaaS revenue growth this time is attributed to the explosion of Seedance 2.0 - the video model Seedance 2.0 can now bring in more than 1 billion yuan in monthly revenue.

A person close to Volcengine revealed that as of now, the average daily consumption of Volcengine Tokens still maintains a monthly growth rate of nearly 40%.

In 2025, the industry was still arguing about which cloud provider was selling the most and best - priced MaaS. In 2026, the industry consensus has become clearer than ever - creating a top - notch model is the best way to rapidly increase Token consumption and make customers willing to pay for it.

The Token war has now clearly divided into two fronts: one is the Coding (code) model, and the other is the video model. Compared with the crowded LLM model track where companies have driven the model prices to rock - bottom, the leading Coding and video models still have some pricing power, especially the video model.

ByteDance has reaped great benefits from the video model Seedance 2.0.

After ByteDance released Seedance 2.0 on February 10th, it immediately became a global sensation, topping the Artificial Analysis Video Arena with an Elo of 1269. It has surpassed overseas top - notch video models such as Veo 3 and Sora 2 in many indicators.

This popularity has directly led to a shortage of the model. Especially during the Spring Festival, there was a long - term user queuing situation - it even took ten hours to queue up to make a video of just a dozen seconds. After that, ByteDance implemented strategies such as a whitelist - providing a full - fledged version of Seedance 2.0 with stable supply to users who signed annual contracts, thus starting the commercialization of Seedance 2.0.

Multiple executives from large companies told 36Kr that just the Seedance 2.0 model alone can bring in more than 1 billion yuan in monthly revenue for Volcengine, and this figure is still on the rise. Moreover, since the API of Seedance 2.0 has not been fully launched overseas, this figure was achieved in a limited market. Volcengine is likely to continue to raise its revenue target for the MaaS business.

Seedance 2.0 has verified one thing: if the model's capabilities are far ahead, there will naturally be people willing to pay. It is understood that the penetration rate of Seedance in the short - drama industry has reached about 95%. "If you are an industry practitioner in the short - drama field, it's hard not to be a user of Seedance," a video Agent practitioner told 36Kr.

Currently, Seedance 2.0 ranks second in the global market share, only after Google's Veo. This year, the goal of Seedance 2.0 is to expand the overseas market and become the video model with the largest global market share. In this process of catching up, Volcengine's MaaS revenue will surely rise steadily.

However, Volcengine doesn't sell all models well.

The Coding front has not brought much revenue growth to Volcengine. The reason is simple: Seed's Coding ability is not good enough.

Who has reaped the benefits of Coding? More than one insider told 36Kr that Zhipu may be the biggest winner.

In 2024, Zhipu bet on a training route similar to Anthropic: focusing on pursuing model intelligence and improving Coding ability. Two years of accumulation have given Zhipu an opportunity to take the lead in this year's Token war.

Zhipu CEO Zhang Peng disclosed in a recent earnings conference call that in the first quarter of 2026, the cumulative price of GLM interface calls increased by about 83%. Despite multiple price adjustments, the API call volume still increased by 400%. Currently, the cached hit Token price of Zhipu GLM - 5.1 has approached the level of Claude Sonnet 4.6. This is the first time that a domestic large - scale model has achieved price alignment with overseas leading manufacturers in the core scenario. Behind this is that GLM - 5.1 has reached the leading domestic level in Coding ability.

Compared with the video model, Coding is currently a more crowded money - making track.

At the end of 2025, API service provider OpenRouter proposed a theory called the "Cinderella's Glass Slipper Effect": Users who enter in the first month after an AI model is released often have a higher retention rate than those who come later. Because when users use a model with good results, they will be as loyal as Cinderella wearing the glass slipper and deeply bind their business to this model.

Therefore, startups such as Yuezhianmian and MiniMax, cloud services of large companies such as ByteDance, Alibaba, and Tencent, and even operators are successively launching their own Coding Plans and increasing sales efforts. They hope to attract more users into their business scope with relatively low prices first, consume a large number of Tokens, and then find ways to increase profit margins.

36Kr learned that since last year, Alibaba Cloud has successively established several new MaaS sales teams, which can cooperate with the original direct sales team to win orders. The main task of the MaaS team is to explore usage scenarios and actively promote enterprise customers to use Agent capabilities to increase Token consumption. 

New stories are happening almost every day on the battlefield of large - scale models. Although Seedance 2.0 has brought rapid growth to Volcengine's MaaS revenue at this stage, if ByteDance wants to maintain the growth momentum in the long run, making up for the shortcoming in Coding is an issue it can't avoid.

Cover source | AI - generated

Scan the code to join the "Intelligent AI Exchange Group"

Welcome to communicate

This article is from the WeChat official account "Intelligent Emergence", author: Deng Yongyi. It is published by 36Kr with authorization.