对话火山引擎CEO谭待：日均消耗180万亿Token后 火山的下一步是什么？

This article is an exclusive interview by Yibang Power with Tan Dai, CEO of Volcano Engine. It outlines the current development stage of China's large model industry and the core business progress of Volcano Engine, and extracts key takeaways for general users.

1. Core industry data: Doubao large model's daily average token usage has exceeded 180 trillion, growing more than 10-fold over the past year. Volcano Engine ranks first in China's public cloud MaaS market with a 49.5% market share. The industry has crossed the inflection point of productivity transformation, and large models have officially entered real-world production scenarios.

2. Key takeaways for general users: Doubao Pro has been officially launched, with a standard plan priced at 68 RMB per month, and premium plans up to 500 RMB per month. The free basic version of Doubao will remain free and maintain its high quality, with no degradation to make way for the paid tier.

3. Shifting usage pattern: Large model usage on weekdays now exceeds that on weekends, proving that large models have evolved from early-stage casual chat tools to productivity tools for work. General users can leverage large models to improve their daily work efficiency.

This interview reveals the latest development trends of the large model industry, and provides multi-dimensional actionable insights for brands planning their AI strategy, AI-powered marketing and product upgrades.

1. Consumer trends: China's AI user base is now fully established. Doubao, the largest AI application in China, has more than 200 million daily active users. Consumers' acceptance of AI-enabled products has increased significantly, creating room for brands to leverage AI to optimize user experience. Already, automakers including Mercedes-Benz and Tesla have improved customer satisfaction by upgrading their in-car infotainment systems with AI, while hardware manufacturers have driven sales growth through AI product enhancements.

2. AI marketing applications: Volcano Engine's Seedance video generation model enables brands to quickly produce multilingual product introductions, after-sales training videos and marketing content. It is also equipped with industry-leading IP copyright protection mechanisms to safeguard the interests of brands and IP holders, making it suitable for commercial use.

3. Favorable ROI: Today, the growth rate of value created per token of large models far outpaces the growth rate of token pricing. This means the ROI of brands' AI布局 is steadily improving, and the timing for commercial deployment is already mature.

This interview sorts out the latest opportunities, risks and cooperation directions in the large model industry, providing clear guidance for sellers looking to develop AI-related businesses.

1. Opportunity outlook: While the industry has crossed the productivity inflection point, it remains in an early stage of development, with room for 10x to 100x overall market growth. Beyond short dramas, the Seedance video generation model has unlocked demand across industries including manufacturing, education, autonomous driving and retail, creating huge incremental growth in new use cases.

2. Partnership-driven growth: Volcano Engine opens up its MaaS capabilities to support third-party sellers in developing products based on its large models. Multiple SaaS providers including WPS and KuaiZi Technology have grown revenue by leveraging Doubao's new generation models, forming a positive value-added cycle.

3. Risk warning: As the industry is still in its early days, short-term competitive outcomes will not determine the long-term landscape. Sellers should avoid blindly engaging in price wars. When choosing partners, they should make pricing decisions based on the actual value the model can deliver, and prioritize tangible deployment outcomes above all else.

This interview reveals the transformation opportunities and development insights that large models bring to manufacturing factories, offering valuable guidance for factories advancing digital and intelligent upgrades.

1. R&D and production applications: AI Coding is now ready for deployment in hardware R&D. Current large models can generate production-ready, verified core code. One chip manufacturing client has iterated 1,300 lines of usable RTL design code via large models, greatly boosting R&D and production efficiency.

2. Commercial application opportunities: The Seedance video generation model can help factories produce multilingual product manuals and after-sales training videos. It can also generate synthetic training data for industrial intelligence and embodied intelligence, solving the pain point of high costs for acquiring real-world training data.

3. Deployment guidance: To roll out AI applications successfully, factories need to combine top-down management promotion with bottom-up employee capability building. Factories can cultivate internal AI pioneers to replicate successful deployment experiences, and use dedicated digital employee management systems to standardize AI application governance.

This interview clarifies the development trends, core customer pain points and growth directions of the large model service industry, providing actionable insights for AI service providers.

1. Industry trends: Large models have now crossed the productivity inflection point and officially entered the production deployment phase. Enterprise clients have become far more pragmatic, and their core demand has shifted from simply accessing large model APIs to solving actual business problems. The overall market remains in its early stage, with enormous room for growth.

2. Core customer pain points: When enterprises adopt large models, they require more than just strong model capabilities. They also need to integrate with existing internal systems and data, as well as meet multi-dimensional requirements for security compliance, copyright protection and organizational management. Offering model access alone cannot satisfy client needs.

3. Growth guidance: Service providers can develop industry-specific skill modules based on mature leading large models, drawing on their own domain experience. If the target market is smaller than 1 billion RMB in size, there is no need to develop independent Agent products, which reduces development and operating costs.

This interview shares the development experience of Volcano Engine, a leading MaaS platform, offering valuable reference for the strategy and operation of large model platforms and cloud platforms.

1. Core customer demands: Today, platform clients require end-to-end large model services. They need not only high-quality foundation models that have crossed the productivity inflection point, but also inference optimization capabilities adapted to multiple domestic chips, as well as end-to-end deployment support for security compliance, copyright protection and integration with enterprise internal systems.

2. Strategic direction: Platforms need to make strategic layout in advance. Volcano Engine seized the growth opportunity by committing to MaaS three years ago. Platforms must balance both model capability and end-to-end deployment service capability. After models cross the productivity inflection point, the importance of Harness (deployment service capability) will continue to rise.

3. Risk mitigation: The industry is still in an early stage. Platforms should not overemphasize short-term stock price volatility and short-term competition outcomes. They should focus on creating real value for clients, price services based on actual model value, avoid pure price wars, and build long-term advantages through comprehensive cost-effectiveness.

This interview discloses the latest development trends and characteristics of China's large model industry, offering high reference value for both industrial and academic research.

1. New industry dynamics: Large models have now crossed the productivity inflection point, moving from laboratory demos to formal adoption in enterprise production scenarios. The fact that weekday usage now exceeds weekend usage is clear evidence of this shift. China's public cloud MaaS market has grown more concentrated, with Volcano Engine holding the top spot at 49.5% market share, and the head effect is already emerging.

2. New progress in business models: C-end AI commercialization has officially launched. Doubao, China's largest C-end AI application with over 200 million DAU, has launched a paid professional tier. The B-end MaaS business model has been proven viable: third-party developers can achieve positive revenue growth building applications on top of leading large models, and a complete commercial cycle has been established.

3. New industry characteristics: Large-scale deployment of large models is no longer a technical challenge, but rather a challenge of commercial implementation and organizational adaptation, which requires end-to-end capability support. Competing purely on model parameters can no longer meet market demand. The industry as a whole remains in an early stage of development, with still room for 10x to 100x growth.