Pinterest has integrated Nvidia Blackwell GPUs and the Nvidia Dynamo inference framework into its AI infrastructure. This new layer combines the hardware with the company's proprietary visual embeddings to support search, recommendations, and the Pinterest Assistant.
The technical upgrade resulted in an 85x improvement in response startup times and a 7.3x reduction in overall latency. Additionally, the Pinterest Assistant can now process 25 times more visual context per request, allowing for simultaneous analysis of multiple image elements.
The platform processes over 80 billion searches every month. Chief Architect Kartik Paramasivam stated that this shared framework allows engineering teams to deploy AI features across multiple products without building custom infrastructure for each one.
Pinterest maintains a multi-vendor strategy for its infrastructure. The company has committed $4 billion to AWS for AI-related services through 2031, while utilizing Nvidia silicon for compute-intensive tasks.
