New Way Now: PayPal paving the way for the future of agentic commerce with Google Cloud

Summary

Srinivasan Manoharan, Director of Machine Learning Platform at PayPal, discusses how the company approaches the era of agentic commerce. PayPal serves millions of consumers and processes approximately $1.8 trillion annually. At that massive scale, the stakes of getting artificial intelligence right are enormous. PayPal faced two primary challenges when adopting AI, as their on-premise infrastructure could not keep up and every domain within the company was adopting AI independently. The goal became building a unified and governed AI platform that makes the entire lifecycle accessible to every engineer. To achieve this, PayPal built its entire AI stack on top of Google Cloud to secure the compute horsepower and scaling flexibility needed to handle high demand, while also preparing to transition into B200 GPUs. On the inferencing side, Google Kubernetes Engine proved transformational. Before GKE, every software release required weeks of regression testing prior to launch. After implementing GKE, the release lifecycle went from weeks down to minutes. With Google Cloud infrastructure, PayPal now serves over 500 models in production, handling nearly one billion requests every single day while the platform operates without interruption. To let developers focus on applying AI rather than managing infrastructure, PayPal built a unified large language model stack. They run open-source models on their inferencing stack on GKE and close-sourced models on the Gemini Enterprise Agent Platform. Developers get a choice of Claude, Gemini, and open-source models through one consistent and governed access layer, bringing the total cost of ownership for developers to zero. Model deployment went from taking weeks down to minutes, allowing 10,000 engineers across PayPal to focus entirely on solving business problems rather than managing infrastructure. Scaling production-grade AI with confidence requires strict visibility and control over how agents behave. PayPal uses the Model Context Protocol to let agents communicate with internal payment systems in a structured and auditable way, ensuring every action gets logged without a black box. Furthermore, AP2 on the other end defines transaction rules such as maximum amounts, approved merchants, and time windows. Today, PayPal supports hundreds of internal tools through their MCP layer, enabling agents to take real actions across internal platforms. Through these advancements, PayPal evolves from a traditional payment processor into a trust infrastructure, creating a new way to power the future of agentic commerce with Google Cloud.