Pinterest Boosts AI-Powered Search Speed Sevenfold With Nvidia Tech

Pinterest has developed a new multimodal artificial intelligence infrastructure, in collaboration with Nvidia, that will enhance visual and conversational experiences on its platform. This new infrastructure will support visual search, content understanding, safety systems, and other AI-driven products, according to a press release issued on September 14.

The infrastructure utilizes precomputed visual representations to achieve 85 times faster response startup and 7.3 times faster overall latency, reducing the need for repeated image processing. Pinterest's Chief Architect, Kartik Paramasivam, emphasized the importance of investing in scalable and complex AI infrastructure to deliver faster, smarter, and more personalized experiences for users.

The multimodal AI infrastructure combines Pinterest's visual embeddings with Nvidia's accelerated computing platform, which includes Blackwell GPUs and Dynamo. This combination allows Pinterest to transform over 80 billion monthly searches into high-value signals for AI-powered discovery and shopping, enabling Pinterest Assistant to process 25 times more visual context per request.

This project is the result of nearly five years of collaboration between Pinterest and Nvidia. Ujval Kapasi, Vice President of AI & HPC Frameworks and Libraries at Nvidia, highlighted the transformative impact of AI on visual discovery, enabling Pinterest to provide intelligent and personalized experiences for hundreds of millions of users.

Pinterest's recent second-quarter results indicated the success of its AI-driven shopping strategy, with revenue surpassing $1 billion for the fourth consecutive quarter and monthly active users hitting a record high. Each pin displayed on the platform now passes through a recommendation model trained on the user's save, search, and board history.