Q&A: Google's AI and computing chief talks about its shapeshifting data centers

Google is reshaping its data center infrastructure to meet the demands of the agentic era. Mark Lohmeyer, vice president and general manager of AI and computing, notes that the rise of AI agents has triggered a significant shift in processing requirements. As users move from simple chat queries to complex tasks that require multiple sub-agents working in parallel, the underlying hardware must handle significantly higher transaction volumes.

To accommodate this growth, Google is upgrading its data centers with a focus on speed, cost-effectiveness, and reliability. The company is actively working to reduce inference costs, with new platforms delivering nearly double the work efficiency of previous generations. This allows customers to support more users without a linear increase in expenditure.

Energy management remains a priority in these upgrades. Google continues to use liquid cooling and has introduced its Axion-based N4A CPUs, which prioritize efficiency for agent orchestration and reinforcement learning loops. The company is also rolling out its eighth-generation TPU platform, which features specialized chips for training and inference to maximize output across different workloads.

Orchestration is evolving alongside the hardware. Google is turning its Kubernetes Engine into an agent-native solution to manage the rapid scaling of sub-agents. By optimizing container startup times and improving network performance through the Virgo fabric, Google aims to provide the capacity needed for large-scale AI clusters. Storage solutions are also being scaled, with Managed Lustre 10T providing 10 terabytes per second of bandwidth to support massive data requirements. These technical changes signal a long-term commitment to maintaining an infrastructure that supports the specific needs of modern AI.