AWS and Pinterest Sign $4B Deal to Scale AI Visual Search

Pinterest and AWS deepen a 15-year partnership with a $4 billion infrastructure commitment, signaling how hyperscale AI workloads now depend on custom silicon and cloud-native architecture.

AI-powered visual discovery has become a defining workload for consumer platforms, but training and serving multimodal models at the scale of hundreds of millions of users requires both compute flexibility and predictable economics. Pinterest's expanded agreement with Amazon Web Services illustrates how leading platforms are restructuring their infrastructure around purpose-built silicon, Kubernetes, and large-scale data lakes to keep pace with generative AI demand.

What was announced

Pinterest announced a major expansion of its collaboration with AWS, its Preferred Cloud Services Provider, including a planned $4 billion commitment for cloud services through 2031. It is the largest infrastructure commitment in Pinterest's history and extends a partnership that began in 2010.

The deal is structured to support Pinterest's AI roadmap across model training, inference and platform infrastructure for more than 600 million monthly active users. Pinterest will expand its use of AWS Trainium for large language models and vision-language models, grow Graviton usage beyond the roughly one third of compute it already powers, and migrate from traditional EC2 environments to a Kubernetes-based architecture on Amazon EKS.

This expanded commitment with AWS gives us the compute flexibility, hardware optionality, and infrastructure efficiency to accelerate our AI vision for the next generation of visual discovery on Pinterest

Why this matters

For CIOs, CTOs and procurement leaders, the Pinterest-AWS agreement is a reference case for AI infrastructure strategy. It shows how long-term cloud commitments are being used to lock in price-performance on custom silicon, while Kubernetes adoption on managed services like EKS becomes the standard pattern for modernizing legacy compute estates.

The combination of Trainium for training and inference, Graviton for general-purpose workloads, and EKS for orchestration provides a blueprint that mid-market and enterprise organizations can adapt to their own AI initiatives, even at a far smaller scale.

Technical details

• Commitment: Planned $4 billion in cloud services through 2031.
• Training and inference: AWS Trainium hosting LLMs and vision-language models powering personalized visual search and AI-assisted discovery.
• General compute: Expanded use of AWS Graviton, already running about one third of Pinterest's compute infrastructure.
• Orchestration: Migration from EC2 to Amazon Elastic Kubernetes Service (EKS) for improved developer velocity and reliability.
• Data platform: Continued optimization of one of the largest-scale data lakes on AWS.
• Scale: AI workloads serving more than 600 million monthly active users globally.

Softprom and Amazon Web Services

Softprom is the official partner of Amazon Web Services. Our team helps enterprises design, migrate and optimize AI and cloud-native workloads on AWS, including compute strategy with Graviton and Trainium, container modernization on Amazon EKS, and data platform architecture.

This content was prepared as part of the Softprom DistriFlow project — an automated system for monitoring and adapting vendor news. Original source: original article.