The evolution of infrastructures for agentic artificial intelligence in enterprises

Upgrading storage and accelerated computing systems now allows data to be kept close to computing resources, improving the efficiency of intensive workloads. New options include support for NVIDIA Blackwell architectures, both air-cooled and liquid-cooled, offering flexibility in model development and fine-tuning.

Octavian Tanase, chief product officer of Hitachi Vantara, noted: “AI is entering the production phase faster than many organizations’ data infrastructures are ready to supportThe goal of the new software solutions is to provide greater control in the transition from pilot projects to industrial production.

A central element of this evolution is theintroduction of AI blueprints that define structured roles for virtual agents, divided into “supervisor” e “worker“. While the latter perform operational tasks, the former coordinate multi-agent workflows, automating complex processes under constant supervision. The integration of open standards such as the Model Context Protocol (MCP) also allows agents to securely connect to external and distributed data sourceseliminating the need for massive information transfers and preserving data sovereignty within corporate environments.

Strategic collaboration between storage and networking vendors aims to simplify access to historical datasetsimproving the explainability of AI systems. Jason Hardy, vice president of NVIDIA, pointed out: “A full-stack AI infrastructure optimized for enterprise needs enables organizations to support a broader range of AI outcomes while maintaining performance and operational consistencyThe new reference architectures promise constant data availability and flexibility for hybrid cloud environments, essential elements for enterprises aiming for secure and scalable digital transformation.