AWS re:Invent 2025 - AI at the speed of news: Bloomberg Media’s vision for the future (IND3331)

Transforming Media with AI: Bloomberg's Vision for the Future

Unlocking the Power of Media Archives

• Bloomberg Media is a global news and business platform that produces a vast amount of content across text, audio, and video formats
• They currently have over 60 million unique viewers consuming their video content each month, distributed across 48 major streaming platforms
• Bloomberg's existing media ecosystem utilizes a hybrid cloud architecture to handle content ingestion, production, management, transformation, and distribution workflows
• However, the exponential growth of their media library (3,000 hours added per day) has made it increasingly challenging to effectively analyze, understand, and leverage this unstructured data

Evolving from Traditional Media AI Approaches

• Traditional AI approaches for media analysis, such as task-specific models and rule-based systems, have limitations in handling the complexity and scale of Bloomberg's content
• Challenges include:
  • Lack of context and temporal understanding when analyzing individual assets
  • Difficulty in scaling and adapting workflows as new AI models and requirements emerge
  • Challenges in seamlessly integrating and transitioning between different vendor solutions

Introducing a Disposable AI Strategy

• Bloomberg recognized the need for a more flexible, adaptable, and future-proof approach to media AI
• Key principles of their "Disposable AI" strategy:
  1. Versioning: Ability to version and target specific model, embedding, and workflow versions
  2. Multi-level Data Quantification: Differentiating between production-ready and non-production data
  3. Federated Search and Metadata Management: Handling multiple databases of metadata and embeddings

Unified Media Analysis and Understanding

• Leveraging a combination of task-specific models, vision-language models, and vector embeddings to extract rich, multimodal insights from media assets
• Task-specific models: Generating labels, transcripts, and other predefined outputs
• Vision-language models: Enabling free-form, natural language understanding and description of media content
• Vector embeddings: Transforming media into numerical representations to enable semantic search and similarity-based retrieval

Hybrid Search and Knowledge Graphs

• Implementing a hybrid search approach that combines keyword-based and vector-based search capabilities
• Dynamically weighting search results based on the user's intent, as determined by a language model
• Constructing knowledge graphs to capture relationships between entities, events, and content, enabling deeper contextual understanding

Automated Content Creation and Distribution

• Leveraging AI agents to orchestrate end-to-end content creation workflows, including:
  • Automated content selection and assembly from media archives
  • Optimizing content for different platform formats and aspect ratios
  • Integrating human review and approval processes
• Enabling rapid, consistent, and traceable content distribution across multiple platforms

Business Impact and Future Outlook

• Bloomberg's "Disposable AI" strategy aims to:
  • Drastically reduce time-to-market for content distribution
  • Unlock new distribution opportunities by creating platform-optimized content
  • Unlock insights from their vast 13 petabyte media library, which is growing by 3,000 hours per day
• Collaboration with AWS has been crucial in shaping and implementing this vision, leveraging the latest cloud-based AI and media services