Google Cloud report finds 69% of firms scaling AI agents by 2028

The transition from batch processing to real-time data streaming is no longer optional for platforms aiming to deploy autonomous AI agents. For streaming providers, this shift means moving away from legacy silos toward unified architectures that can process unstructured data like video logs and emails in real time. As companies like Shopify and Deutsche Telekom integrate these agents, the competitive gap will widen between 'data leaders' with high accuracy and 'laggards' who only trust 22% of their AI outputs. The industry must now prioritize data governance and contextual modeling to ensure these systems can act autonomously without human intervention. Watch for a surge in infrastructure investment toward Systems of Action that prioritize low-latency data availability over static archival storage.

Google Cloud has been aggressively expanding its AI agent platform across enterprise verticals, with particular emphasis on real-time data streaming as a prerequisite for autonomous decision-making. In early 2026, Google Cloud announced that its Vertex AI Agent Builder had reached general availability with built-in grounding to enterprise data sources, enabling organizations to connect agents directly to live data pipelines rather than static repositories. Verizon scales Gemini Enterprise AI to automate customer service and network operations divisions, signaling that telecom operators view agent-based automation as a core operational strategy rather than an experimental layer.

The business case for real-time data infrastructure is being reinforced by major cloud providers competing for enterprise AI workloads. Meta Platforms and BlackRock announced plans to build a 1-gigawatt data center complex in Texas costing approximately $14 billion, underscoring the capital intensity of the compute and storage layer required to support always-on AI agents at scale. Meanwhile, Cerebras filed for an IPO with a reported $10 billion contract from OpenAI, suggesting that inference workloads from agentic systems are driving demand for specialized hardware beyond traditional GPU architectures. These infrastructure investments directly address the data-quality bottleneck identified in the Google Cloud report: agents that must act on fresh, contextual data require low-latency pipelines and purpose-built compute rather than batch-oriented legacy stacks.

On the technical side, the shift toward streaming-first architectures is already visible in production deployments. Deepgram launched its real-time speech-to-text and voice agent models as Amazon SageMaker endpoints, running inference inside customer VPCs with sub-300-millisecond latency for use cases including live captioning and contact-center transcription. This deployment pattern illustrates the architectural shift the Google Cloud report advocates: moving AI processing closer to the data source, eliminating round-trips to external clouds, and maintaining compliance through inherited IAM and encryption controls. For streaming platforms evaluating , recommendation, or operational automation, the Deepgram-SageMaker model demonstrates that real-time agent inference is achievable within existing cloud security boundaries without sacrificing performance. The industry is also seeing to further streamline these automated workflows, while are becoming essential as these agentic workloads drive a massive surge in token usage.