How Pinterest pushes boundaries of AI-powered discovery using AWS

AI has always been at the foundation of Pinterest, first generating related ideas based on users’ saved Pins. Over time, it moved to deep learning and multimodal models that deeply understand the relationship between content uploaded and user interests and actions to recommend the most relevant ideas out of the billions of Pins that exist.

However, building a fundamentally visual platform with a mission based on user trust and responsibility presented unique challenges that differ from its competitors.

“Our goal is to make Pinterest a positive corner of the internet,” says Paramasivam, “Others tune AI to focus on viral trends and maximizing time spent. We made the choice to prioritize positivity and enable our users to easily find genuine moments of inspiration.”

The technical challenges could have been daunting. As Pinterest grew from thousands to hundreds of millions of users, the company faced the complex task of maintaining its inspirational mission while building the technical infrastructure to support such a massive scale.

Pinterest's decision to build on AWS from day one in 2010 proved pivotal, giving it the freedom to innovate with generative AI and apply it rapidly at scale. Using a microservices-based architecture on AWS, Pinterest can process billions of images, understand user preferences across diverse categories from home décor to fashion, and deliver personalized recommendations in real-time—all while maintaining the positive, safe environment that differentiated the platform from other social networks.

At the company’s core is Pinterest’s recommendation system models. Built entirely on AWS, Pinterest uses Amazon EKS for rapid AI deployment across the business, 10,000+ Amazon EC2 G5 instances for inference and 600+ P4/P4de instances for training, analyzing over 18 terabytes of data daily. Pinterest processes over 10 million AI-powered recommendations per second to support personalized discovery, advanced content moderation, and responsible features that demonstrate how social platforms can be both wildly successful and genuinely beneficial to user wellbeing.

One of its recent innovations, Pinterest Canvas, is a homegrown latent diffusion vision foundation model that supports high-resolution image generation and manipulation. Using Amazon EC2 GPU instances to provide the speed and cost-efficiency required for demanding machine learning processes, Pinterest Canvas enhances images to paint more inspiring backgrounds for shopping products, allowing advertisers to create more engaging ads while removing time-consuming tasks like image cropping and design.

Pinterest also introduced new visual search updates  – a set of AI-powered search features that uses visual language models built using Amazon SageMaker to quickly recognize over 2.5 billion objects, enabling users to personalize discovery and refine searches for particular objects within an image.

Pinterest’s latest invention is Pinterest Assistant, a generative AI voice-enabled conversational experience. As the name suggests, Pinterest assistant acts as a visual-first collaborator that helps users find what they are looking for by enabling them to ask open-ended questions.

Using Amazon Bedrock, Pinterest is also deploying AI agents that don’t just personalize but actively moderate and filter out negative and inappropriate content. “For us, trust isn’t just a feature, it’s the foundation of Pinterest,” explains Paramasivam. “We build, train and leverage AI to filter out content and behavior that is not appropriate for our platform and instead tune our AI to allow our users to easily find what they are looking to explore by personalizing their search and feed.”

Through its strategic AWS implementation, Pinterest has built a scalable, efficient infrastructure that not only meets current business demands but positions the company at the forefront of responsible AI innovation, demonstrating that technology can be both profitable and positive for its users. Using AWS, Pinterest has realised remarkable business outcomes including:

• 17% year-over-year revenue growth
• Significant cost optimization through strategic resource management
• 11% increase in monthly active users
• 230 basis point improvement in search fulfillment
• 70% of user discovery now AI-driven

“In essence, AI algorithms are critical for us to ensure that users trust and enjoy our platform – and AWS has been instrumental in implementing this at scale,” shares Paramasivam.

Pinterest's long-standing relationship with AWS shows how having a strong technical foundation can support any company, from growing startup to global enterprise to deliver personalized experiences at massive scale. As Pinterest works towards its mission to counter traditional social media's addictive models, AWS provides the technological backbone to prove that responsible AI can thrive at any scale, setting a new blueprint for social media's future.