Seedance 2.0 Officially Opens with 120 Trillion Tokens as a Base: What Is Volcano Engine After?

On April 2, Volcano Engine announced two major developments at its Wuhan roadshow.

First, the Seedance 2.0 API is officially open for public beta to enterprise users. The official website states it is "oriented toward enterprise users," with a pay-as-you-go model that no longer requires a pre-paid minimum commitment of 10 million RMB.

Second is a figure. As of March this year, the daily average token usage of the Doubao large model has surpassed 120 trillion, doubling in the past three months and increasing 1,000-fold since it first began providing enterprise services in May 2024.

The Volcano Ark Experience Center simultaneously launched a new demo for Seedance 2.0, featuring a video of a piano performance that demonstrated exquisite audio-visual synchronization. The force of the key presses, the fluidity of finger movements, and the complete piano sound effects nearly reached the standards of professional production.

This is not a routine iteration. The opening of the Seedance 2.0 API, combined with the 120 trillion token figure and the enterprise data knowledge management platform released the same day, provides a clear view of what Volcano Engine intends to achieve.

The Door to "Openness" Is Still Ajar

In February, Seedance 2.0 was launched for internal testing. With its multimodal creation methods and built-in camera movement effects, it quickly garnered high attention globally. On overseas social media platforms, several demo videos surpassed one million views. The sentiment on social media at that time was typical: on one hand, "the strongest on earth," and on the other, "Hollywood is finished."

Feng Ji, CEO of Game Science, the developer of Black Myth: Wukong, evaluated Seedance 2.0 as "currently the strongest video generation model on earth" after a trial run.

However, very few people could use it at that time. During the internal testing phase, the barrier to entry for Seedance 2.0 was set at a minimum commitment of 10 million RMB. Access was held by a few top-tier teams, leading to gray-market operations such as group buying and private sub-licensing of interfaces. This situation lasted for nearly two months.

On April 2, Seedance 2.0 was finally announced as officially "open," but upon reaching the entrance, one finds that the threshold remains high.

While users pay based on actual calls and are no longer required to provide a 10-million-level commitment, newly signed users are defaulted to only 10 concurrent calls with no option for an increase. Furthermore, advanced features such as real human faces and custom virtual avatars are not supported; users can only perform secondary creation using the platform's public virtual avatar library. Additionally, newly signed teams must pay a security deposit of approximately 1 million RMB, which is gradually released and made available for use only after completing a set framework within one year.

For high-quality short drama teams, this remains a significant barrier. The restrictions on realistic human styles and the 10-concurrency limit cannot support mass production, and the 1 million RMB deposit locks out a considerable number of small and medium-sized teams.

The platform's explanation is copyright and portrait security. Seedance 2.0 has established copyright and portrait security protections covering various modalities and the entire creative process, with systems to detect and defend against infringement and deepfakes.

This consideration has a practical basis: during the February internal test, Seedance 2.0's ability to recreate an original voice from just a single facial photo sparked extensive discussions on ethics and data security. A video by Tim, a well-known blogger from Video Hurricane who participated in the initial internal test, caused a viral sensation, prompting the official team to urgently suspend the real human material reference function.

However, the business logic is equally clear. Stratifying by capability and grading by risk is essentially a way to filter users. Institutions with substantial capital and compliance needs are the enterprise clients Volcano Engine truly wants to serve. Small and medium-sized teams are not being expelled but are being pushed toward platform partners that have already integrated Seedance 2.0. For instance, Chanmama, a service provider in the ByteDance ecosystem, has become a primary "sub-distributor." The profit distribution of the entire ecosystem has thus been reorganized.

Compared to Kuaishou's Keling, another major competitor in the AI video track, Kuaishou's strategy is relatively more open, with a simpler pricing system for individual users and a relatively more relaxed content safety threshold. Volcano Engine has chosen a different path: establishing a traceable and accountable access framework on the enterprise side, ensuring that high-sensitivity capabilities flow only to high-value, high-compliance clients.

It is too early to tell which approach is correct. However, these two strategies correspond to two completely different business judgments. Kuaishou is betting on a larger creator market, while Volcano Engine is targeting the mass production needs of industry clients. This is determined by their fundamental positioning.

What 120 Trillion Tokens Means

The figure of 120 trillion only makes sense when placed within a coordinate system.

According to a report by IDC, the call volume of large models on public clouds in China reached 114.2 trillion tokens in 2024, with Volcano Engine ranking first in the Chinese market with a 46.4% market share. By the first half of 2025, this share is expected to rise further to 49.2%, meaning that one out of every two tokens produced on China's public clouds will come from Volcano Engine.

However, a long-standing unresolved question remains: how much of this 120 trillion tokens is consumed internally by the ByteDance ecosystem?

Douyin, Toutiao, CapCut, and Lark are themselves massive consumers of tokens. One of the most criticized points about Volcano Engine in recent years is its high proportion of internal revenue, leaving doubts about its external commercialization success. Public data has never broken down the ratio of internal to external traffic. Reports at the end of last year mentioned that internal revenue accounted for about 70% of Volcano Engine's total. If this proportion hasn't significantly decreased, the actual scale of external commercialization behind the 120 trillion figure might be much smaller than the number suggests.

In comparison, a more valuable data point might be that the number of enterprises with a cumulative token usage exceeding one trillion on Volcano Engine has grown from 100 at the end of last year to 140. This represents deeply integrated major clients rather than trial-based, occasional users. Adding 40 such clients in three months is not a slow growth rate. However, by comparison, Alibaba Cloud has a much larger base of enterprise clients, and Tencent Cloud has its own moats in the financial and government sectors. Volcano Engine's commercialization narrative needs to show how many of these 140 companies are running core businesses with large models rather than just circling the PoC (Proof of Concept) stage.

Alibaba Cloud and Tencent Cloud's positions in the MaaS (Model-as-a-Service) market are not ideal. Alibaba's Tongyi series has far less influence on the consumer side than on the enterprise side, and Tencent's Hunyuan is progressing even slower. Baidu's Wenxin started early, but in the massive wave of cost reduction and efficiency improvement, price wars have weakened its advantage. Volcano Engine was the first to drive large model API prices into the "li era" (referring to the price unit of 0.001 RMB), dropping prices per 1,000 tokens to the industry floor in May 2024, which almost all cloud vendors subsequently followed.

But as the price war reaches today's level, it's no longer just about price.

Enterprise Knowledge Management: Another Tough Nut to Crack

On the day of the Wuhan roadshow, another release was more low-key. Xiao Ran, General Manager of Solutions for Volcano Engine's Digital Intelligence Platform, proposed at the ArkClaw Data Intelligence Session that the positioning of AI applications in enterprise-level scenarios needs to be upgraded from a "personal assistant" to an "enterprise digital partner," and launched the Volcano Engine Enterprise Data Knowledge Management Platform.

The core logic of this product is to solve the problem of large models being "usable but not effective" in enterprises. General large models do not understand an enterprise's internal regulations, product knowledge, or historical decisions; each call requires a massive amount of contextual supplement, leading to low efficiency and frequent hallucinations. The Enterprise Data Knowledge Management Platform builds a company-specific "knowledge base" for the model: integrating personal knowledge bases, public knowledge bases, and scenario-based workspaces, allowing employees in different positions to trigger corresponding authorized knowledge when calling the model, rather than starting from zero every time.

This line of development is an extension of RAG (Retrieval-Augmented Generation) technology engineering and a deep-water zone for enterprise AI implementation. For large model cloud vendors, the challenge lies in the high sensitivity of enterprise data, diverse data formats, and complex permission systems. If any link in the experience is poor, enterprises will not renew.

Compared to Seedance 2.0, which can explode on social media as a consumer-facing product, enterprise knowledge management is a "slow burn." However, it corresponds to higher average contract values, stronger customer stickiness, and lower churn rates. This is a piece that Volcano Engine needs to fill.

Flags Planted at Every Layer, but Vertical Depth Is the Real Challenge

Deconstructing Volcano Engine's moves reveals a clear strategic map.

Model Layer: The Doubao large model family, covering text, vision, video (Seedance), voice, code, and vectors—multimodality is largely in place. Inference Service Layer: Volcano Ark, ranking first in national market share.

Application Layer: ArkClaw (AI programming assistant/enterprise version of OpenClaw) and Coze (low-code application building).

Data Layer: The Enterprise Data Knowledge Management Platform launched at the Wuhan event, which acts as the knowledge base for ArkClaw, allowing large models to "read and understand the enterprise's own knowledge assets."

The entire architecture, from underlying computing power to top-level applications, is theoretically integrated.

However, Alibaba Cloud is also on this path. The Tongyi family + Alibaba Cloud Baixian + DingTalk AI + Enterprise Knowledge Base follows an almost identical logic. Tencent Cloud, relying on WeCom and Tencent Meeting, has a natural entry point in collaboration tool scenarios. Huawei Cloud has strong relationships with government and enterprise clients.

Enterprise software's ultimate success has never been just about technology. It's about service capability, the ISV (Independent Software Vendor) ecosystem, the stability of private deployments, and the sales team's familiarity with government and enterprise decision-making processes. These areas have historically been ByteDance's shortest boards.

Volcano Engine's penetration in consumer internet and content industries is already deep: 90% of mainstream car brands and 80% of leading securities firms are real achievements. But expanding from media, entertainment, and automotive to manufacturing, healthcare, and government requires organizational capabilities that differ significantly from ByteDance's current DNA.

The three developments today show Volcano Engine continuing to plant flags at every layer. Penetrating each point deeply will take more time.

Risk Warning and Disclaimer

Market risk exists; investment requires caution. This article does not constitute personal investment advice, nor does it take into account individual users' special investment goals, financial situations, or needs. Users should consider whether any opinions, views, or conclusions in this article are appropriate for their specific circumstances. Investing based on this is at your own risk.