What challenges did TwelveLabs face before AI implementation?

Scaling video AI from research prototypes to robust global production systems capable of handling the massive volume and complexity of unstructured video data. Traditional search methods failed at petabyte scale and at the granularity required to find specific video moments. Need for cloud infrastructure that supports intensive training of multimodal AI models, high-volume inference, and seamless global deployment with reliability and cost efficiency.

Source

Scaled productionEvidence: Medium55/100

TwelveLabs Unlocks the Full Potential of Video with Amazon Bedrock

Use case typeProduct discoveryUpdated Jun 13, 2026

TwelveLabs developed breakthrough multimodal video AI foundation models, Marengo and Pegasus, that understand video as unified stories across sight, sound, and time, enabling powerful video search and analysis. To deploy these models at production scale, TwelveLabs leveraged AWS infrastructure, including Amazon Bedrock, Amazon SageMaker HyperPod for model training resilience, Amazon EKS for scalable model serving, EC2 G6e and P5 instances for inference, and Amazon S3 with S3 Vectors for unified storage and vector-based semantic search. This unified AWS platform supports petabyte-scale video indexing and sub-second search across billions of vector embeddings, delivering reliable, cost-efficient video intelligence at unprecedented scale. TwelveLabs' solution transforms massive unstructured video archives into searchable, analyzable assets, powering applications across media, advertising, manufacturing, and government sectors.

Organization: TwelveLabs
Industry: Tech & Comms
Location: United States
Published: April 2026

Reported outcomes

Strategic outcomes

New product / capabilityEnabled semantic video search at petabyte scaleScale & capacitySupported production video AI at global scaleCost efficiencyImproved production reliability and cost efficiencyCustomer experience & trustDelivered fast, scalable video intelligence capabilities

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: TwelveLabs
Provider: AWS
Maturity: Scaled Production
Linked source: AWS Customer Stories

Improved production reliability and cost efficiency of video AI workflows at scale

Customer identity supportedSource describes one deploymentMaturity supported

Primary read

Use case focus

Showing 3 of 3

1Video Search
2Multimodal AI
3Semantic Search

Scaling video AI from research prototypes to robust global production systems capable of handling the massive volume and complexity of unstructured video data.
Traditional search methods failed at petabyte scale and at the granularity required to find specific video moments.
Need for cloud infrastructure that supports intensive training of multimodal AI models, high-volume inference, and seamless global deployment with reliability and cost efficiency.

Built on AWS to support the full AI lifecycle: Amazon SageMaker HyperPod for fault-tolerant multi-GPU training, Amazon EKS for scalable containerized model serving, EC2 GPU instances for inference, and Amazon S3 with S3 Vectors for unified data and embedding storage.
Developed models Marengo and Pegasus to encode video content into multi-dimensional vector embeddings enabling semantic search with approximate nearest neighbor search over billions of vectors.
Deployed models through Amazon Bedrock for easy developer access and integration while ensuring data control and security.

Technologies

Amazon Bedrock Amazon SageMaker HyperPod Amazon Elastic Kubernetes Service Amazon EC2 G6e Instances Amazon EC2 P5 Instances Amazon S3 Amazon S3 Vectors

Enabled precise, contextual video search and rapid insight discovery at petabyte scale, making video AI viable for enterprises with massive video archives.
Improved production reliability and cost efficiency of video AI workflows at scale.
Supported diverse users from large media companies to government entities with fast, scalable video intelligence capabilities.

Architecture

In-depth description of AI models Marengo and Pegasus, their vector embedding-based semantic search architecture, and the use of AWS managed services EKS, SageMaker HyperPod, EC2 GPU instances, and S3 Vectors for scalable storage and inference.

Sources & evidence1

Evidence: Medium55/100Evidence strength