Authors:
Shin Wei Lim, AI Software Engineer, Intel
Jocelyn Teh, Machine Learning Engineer, Intel
Ehssan Khan, AI Software Engineer, Intel
Gary Koay, Software Engineering Manager, Intel
Richard Chuang, Principal AI Engineer, Intel
Introduction
One of the most pressing challenges for media and entertainment today is navigating between the explosion of data and the need to deliver personalized, monetizable content experiences at scale and in real time. AI makes this possible, but conventional GPU-based approaches often require heavy capital investment and complicated deployment. In this article, we will cover how Intel® Xeon® processors with built-in Intel® Advanced Matrix Extensions (Intel® AMX) and Intel® Advanced Vector Extensions (Intel® AVX-512) make it possible to deploy high-performance AI inferencing for media data curation and retrieval-augmented generation (RAG) without requiring discrete GPUs. At IBC2025, Broadpeak—partnering with Intel—will demonstrate how Intel Xeon processor-powered AI data curation integrates seamlessly into Broadpeak’s Agentic L-Banner Advertising solution, a Dynamic Ad Insertion solution named BannersIn2, showing how Intel Xeon CPUs already deployed across data centers can drive new monetization opportunities for content distributors.
AI Data Curation on Intel Xeon CPUs: Built for Media Workloads
Figure 1 - Media Data Curation Framework
The Media Data Curation Framework developed on Intel Xeon processors is a comprehensive data curation framework designed to derive actionable insights from various forms of unprocessed raw data, including text, images, videos, audio, and more. This versatile platform allows users to seamlessly register new data sources and define custom data curation pipelines. These pipelines guide the framework in acquiring, processing, and analyzing raw data, automatically extracting valuable insights. The processed insights are then stored in connected backend databases for future applications such as model refinement, machine learning training, and Video Retrieval-Augmented Generation (Video RAG).
The core architecture of this framework is composed of four major components:
- Unified Data Access Layer: This module supports the integration of diverse data sources, enabling the continuous flow of raw data into the framework. It handles various formats (text, image, video, audio) and protocols (static files, live streaming), ensuring robust and scalable data acquisition.
- Data Pre-Processing: As the 1st stage of the data transform layer, raw data undergoes transformation and cleaning. Advanced algorithms are employed to normalize, filter, and enrich the data, preparing it for subsequent analysis. The system is designed to handle large volumes of data efficiently, utilizing parallel processing and distributed computing techniques.
- AI-powered Data Transform Pipelines: This component uses sophisticated analytical models and machine learning techniques to uncover patterns, trends, and correlations within the processed data. This framework can extract meaningful insights from complex datasets by leveraging natural language processing, computer vision, and audio analysis.
- Database Management and Plugins: The framework offers an integrated, scalable storage solution, where extracted insights in JSON and vector formats are securely stored and easily accessible for future retrieval. This service supports various database technologies and ensures high availability, data integrity, and efficient query handling, enabling seamless integration with downstream applications.
The Data Transform layer supports user-defined multimodal pipelines, including:
- Video and audio understanding: Segmenting and transcribing streams, extracting entities, and vectorizing metadata using Intel AMX-accelerated models.
- Ad content analysis: Ingesting diverse assets—commercial clips, posters, text, and web content—and summarizing them with LLMs (e.g., DeepSeek, Qwen, Florence-2, Llama).
- Vector search & RAG: Using AI agents to retrieve and rank the most relevant ads based on context, ensuring non-intrusive, personalized ad placement.
- Intel® QuickAssist Technology (Intel® QAT): Securing and accelerating ad media ingestion and streaming with high-throughput encryption/decryption.
This framework is optimized for Intel Xeon platforms. It uses the OpenVino™ toolkit for optimized model execution and techniques such as INT8 quantization, KV caching, and mixture-of-experts (MoE) to run LLMs efficiently on CPUs. The BOM of the models used in the media data curation framework is summarized below.
Early POC results demonstrate that a 36-core Xeon system with AMX can curate 15 minutes of video in minutes, with ad vector searches completing in under 10 seconds—all without a discrete GPU. This means that the data curation and search time can be further shortened if we deploy the POC to Intel Xeon 6700 (SP) or 6900 (AP) platforms with higher core counts.
If we zoom into data transform and query pipelines in this framework, the data flows can be depicted as below.
Figure 2 - Data flows in the data transform and query pipeline layers.
Below is an agentic AI view of this data curation framework that connects with the advertisement content library and dynamic ad insertion agent.
Figure 3 - Agentic AI View of Data Curation Framework for Dynamic Ad Insertion
Dynamic Ad Insertion with Broadpeak
Broadpeak’s Dynamic Ad Insertion software, trusted by major broadcasters and streaming providers worldwide, is the pilot solution that integrates Intel’s AI-driven media data curation framework via APIs. The results are:
- Contextual ad placement: Ads are matched precisely to video scenes, ensuring relevance and maximizing viewer engagement.
- Real-time performance: Xeon-powered pipelines can also enable insertion into live streams, not just VoD.
- Scalable monetization: Content providers gain a flexible, AI-driven ad engine without re-architecting workflows.
- Lower cost: Intel AI-based Data Curation Reference Framework can be implemented entirely with Intel Xeon processors, without the need for discrete GPUs.
Learn more about Broadpeak’s approach to Dynamic Ad Insertion and BannerIn2 solution.
Demonstration at IBC
At IBC2025, Broadpeak will demonstrate agentic Dynamic Ad Insertion powered by Intel’s AI-driven Data Curation Reference Framework.
The demo highlights:
- Video segmentation, transcription, and metadata generation using Intel AMX.
- AI-based ad summarization and ranking via RAG pipelines.
- Seamless integration into Broadpeak’s BannersIn2 platform for ad insertion.
Intel’s Data Curation Reference Framework can be implemented entirely with Intel Xeon processors, without the need for discrete GPUs.
The integrated solution can be prompted to capture web information to understand the nature of the “to-be-promoted” product, provide a product summary, and vectorize the information. After all advertisement content is added to the library, the platform starts searching for the best advertising time stamps using the curated data of a movie. The solution automatically proposes the best-suited time stamps to insert the targeted advertisement, where the ad content is closely related to the scene and actions in the movie. The Ad Ops team can decide on the placement of the advertisement and program it to the content distribution service for broadcasting. Learn more about Broadpeak’s BannerIn2 solution.
This marks the first step in extending the media data curation and Video RAG frameworks to a wide range of use cases, from video production and social media reels to metadata automation and content personalization.
Why it Matters
- For ISVs: Simplified integration of AI into media pipelines.
- For content providers: A cost-effective monetization path leveraging existing Xeon-based infrastructure.
- For viewers: More relevant, less disruptive ad experiences.
The Media & Entertainment AI market is projected to reach $99.5B by 2030(1). With data volumes exceeding 4.6 TB per second globally, curating media data in real time is essential. Intel Xeon with Intel AMX ensures that the CPUs already powering today’s content networks can also fuel tomorrow’s AI innovation.
The Intel-Broadpeak collaboration proves that AI innovation in media doesn’t require GPUs. Intel Xeon processors with Intel AMX provide inferencing performance, scalability, and ecosystem support to make AI-driven data curation and monetization a reality today.
At IBC, the joint demo will show how Dynamic Ad Insertion powered by AI on Xeon CPUs delivers real-world value: unlocking new revenue for broadcasters and distributors, optimizing infrastructure investments, and enabling personalized viewing experiences at scale.
Media AI meets monetization with Intel Xeon processors.
Learn more about Intel Xeon processors and Intel AMX capabilities, and discover Broadpeak’s dynamic ad insertion solutions. Meet with the Broadpeak team at IBC 2025 | Hall 1 Booth #F.83
Notices and Disclaimers
Performance varies by use, configuration, and other factors. Learn more on the Performance Index site.
Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates. See backup for configuration details. No product or component can be absolutely secure.
Your costs and results may vary.
Intel technologies may require enabled hardware, software, or service activation.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.