AI Short Dramas Dominate Token Consumption

Chinese Data Reveals AI Short Dramas Lead Token Usage

AI short dramas have emerged as the largest consumer of computational tokens globally, according to new data from a major Chinese tech player. This shift marks a significant pivot in how generative AI resources are allocated across different industries.

On June 3, China Online Literature Group (COL) disclosed that AI-generated short-form video content now accounts for the majority of token consumption in its domestic operations. The company reported that this sector holds a dominant 55% share of all token usage.

This revelation challenges previous assumptions that software development or enterprise automation would drive the highest demand for large language models (LLMs). Instead, entertainment and visual generation are leading the charge.

Key Facts at a Glance

• Dominant Sector: AI short dramas and video generation hold a 55% market share in token consumption.
• Secondary Use Case: E-commerce and marketing applications, including live streaming and advertising, account for 24% of usage.
• Coding Lag: Software development utilizes only 15% of tokens, despite hype around AI programming assistants.
• Volume Comparison: Video generation consumes more than double the tokens compared to the software development sector.
• Source Data: Figures were released by China Online Literature Group during institutional investor communications.
• Trend Indicator: Visual content creation is becoming the primary driver of AI infrastructure costs.

Entertainment Outpaces Enterprise Coding

The most striking aspect of these findings is the sheer volume of tokens consumed by video generation compared to traditional coding tasks. For months, Western tech narratives have focused heavily on Vibe Coding and the integration of AI into software engineering workflows.

Prominent models like Anthropic's Claude have been marketed specifically for their advanced coding capabilities. Investors and developers alike expected that automated code generation would be the killer app for LLMs.

However, the data suggests otherwise. While coding tools are valuable, they do not generate the same continuous, high-volume token traffic as video production. A single line of code might require a few hundred tokens to generate and verify.

In contrast, generating a coherent short video sequence involves complex prompt engineering, frame-by-frame synthesis, and audio synchronization. This process demands significantly more computational power and context window usage.

Why Video Drives Higher Costs

• Complexity: Video generation requires multi-modal processing (text, image, audio).
• Iteration: Creators often generate multiple versions before selecting the final output.
• Length: Even short clips require sustained attention mechanisms from the model.
• Resolution: High-definition outputs increase the data payload per token processed.

The Rise of the Micro-Drama Economy

To understand why short dramas dominate, one must look at the explosive growth of the micro-drama industry in Asia. These are fast-paced, vertical-format videos designed for mobile consumption, often released in episodes of one to two minutes.

Producing these traditionally was expensive and slow. Now, AI allows studios to generate scripts, voiceovers, and visuals rapidly. This has lowered the barrier to entry, leading to a surge in content volume.

The economic model relies on high turnover. Studios produce dozens of series simultaneously to test audience engagement. Each iteration consumes tokens. When scaled across hundreds of production houses, the aggregate demand becomes massive.

This trend is beginning to influence Western markets as well. Platforms like TikTok and YouTube Shorts are seeing an influx of AI-assisted content. While not yet at the scale seen in China, the trajectory is clear.

Entertainment companies are realizing that AI can reduce production costs by up to 80% for certain types of content. This efficiency drives adoption faster than B2B software tools, which often face longer sales cycles and stricter compliance hurdles.

Marketing and E-Commerce Follow Suit

While short dramas lead, e-commerce and marketing remain substantial consumers of AI resources. Holding a 24% share, this sector is the second-largest driver of token economics.

Live streaming hosts use AI avatars to maintain 24/7 broadcasts without human fatigue. These avatars interact with viewers in real-time, requiring constant low-latency token generation to respond to comments and questions.

Additionally, digital advertising campaigns rely on AI to generate thousands of unique ad variations. Personalization at this scale requires dynamic content creation. Text, images, and video snippets are tailored to individual user profiles.

This creates a steady, high-volume baseline of token usage. Unlike software development, where usage might spike during deployment phases, marketing needs are continuous. Brands run campaigns daily, ensuring consistent demand for generative AI services.

Marketing vs. Development Usage

What This Means for Developers and Businesses

For businesses, this data signals a need to reassess AI investment strategies. If your goal is to leverage AI for immediate customer engagement, video and interactive media offer higher ROI potential than internal coding tools.

Developers should note that the infrastructure requirements for video generation are distinct. They need robust GPU clusters capable of handling multi-modal workloads. Standard CPU-based servers used for text-based LLMs may not suffice.

Furthermore, the cost structure differs. Video generation is computationally intensive. Companies must optimize their prompts and workflows to minimize waste. Efficient pipeline design is crucial to manage the high token costs associated with video output.

Investors should watch for startups focusing on video-specific AI optimization. Tools that reduce the token count per minute of generated video will become highly valuable. Efficiency gains in this sector could unlock further growth.

Looking Ahead: The Future of Token Economics

As AI models become more efficient, the cost per token will likely decrease. However, the total volume of consumption is expected to rise. The democratization of video creation means more users will enter the market.

We anticipate a bifurcation in the AI landscape. One stream will focus on high-efficiency text and code for enterprise productivity. The other will prioritize high-fidelity multimedia for creative industries.

Regulatory bodies may also take notice. The ease of generating realistic video content raises concerns about deepfakes and misinformation. Expect tighter guidelines on labeling AI-generated video content in the near future.

The dominance of short dramas highlights a cultural shift. Consumers are increasingly comfortable with synthetic media. This acceptance paves the way for broader applications in education, training, and virtual reality.

Gogo's Take

• 🔥 Why This Matters: This data proves that creative entertainment is currently the primary economic engine for generative AI, not enterprise productivity. It validates the "creator economy" model over the "coder assistant" narrative, suggesting that mass-market adoption is driven by visual content consumption rather than backend optimization.
• ⚠️ Limitations & Risks: The high token consumption of video generation raises serious cost and sustainability concerns. Training and running these models requires immense energy. Additionally, the proliferation of AI-generated short dramas could lead to market saturation and raise ethical issues regarding copyright and deepfake misuse.
• 💡 Actionable Advice: Businesses should audit their AI spend to see if they are over-investing in coding tools while under-utilizing video marketing. Start experimenting with AI video pipelines now, but focus on prompt optimization to control costs. Watch for emerging tools that promise lower token-per-second ratios for video generation.