At Google Cloud Next, Google outlined an ambitious expansion of its AI stack, introducing new TPU chips, enterprise AI agent tooling, and security systems, which it touted as enabling enterprise AI at both lower cost and reduced operational complexity.
Central to the event’s announcements is the latest generation of tensor processing units, comprised of two chip architectures: one optimized for model development and another for inference, where applications deliver real-time responses.
As AI evolves, the balance of compute demand is tilting toward inference, particularly with the rise of AI agents that execute multi-step tasks. Google’s new TPUs are engineered to address both stages with greater efficiency, emphasizing lower latency and improved performance per watt.
Google’s TPUs are widely seen as an emerging competitor for NVIDIA’s dominance in the AI chip market, but they still face an uphill battle.
“Google TPUs already represent the main alternative for NVIDIA chips for AI compute,” Gil Luria, Head of Technology Research at D.A. Davidson, told Techstrong.it. “We expect future generations of TPUs to get better, but at the same time NVIDIA GPUs are still getting better as well and maintaining their lead.”
The training-focused TPU supports large-scale model development by enabling thousands of processors to operate as a unified system. Enhancements in networking and memory efficiency accelerate model training while managing power constraints in data centers. The inference TPU, by contrast, prioritizes speed and concurrency, using expanded on-chip memory and optimized data pathways to handle complex reasoning tasks more quickly.
Cost control is part of Google’s messaging. The company claims significant gains in performance relative to both energy consumption and dollar spend, acknowledging the growing pressure on cloud providers to make AI economically viable. The company’s approach mirrors an industry-wide trend toward cost-conscious custom silicon as companies seek alternatives to NVIDIA’s pricey GPUs.
Managing AI Agents
Beyond hardware, Google is repositioning its enterprise AI platform to address what it sees as the next enterprise challenge: managing large numbers of autonomous agents.
The newly introduced Gemini Enterprise Agent Platform consolidates model development and deployment into a single system. Along with supporting governance, it improves earlier tooling by adding orchestration capabilities and integrated security controls.
Also on the security front: Google is also introducing simulation tools to test agent behavior before deployment to address concerns about reliability and unintended actions in autonomous AI systems.
Google’s focus on agentic security extends from internal security to using agents to combat cyber-attacks. The company debuted a set of AI-driven security agents built to automate threat detection and response. These tools aim to reduce the gap between vulnerability discovery and remediation, a gap that is narrowing as attackers increasingly use AI to identify exploits. The new agents handle tasks like threat hunting and contextual analysis, building on earlier automation efforts.
To upgrade governance, each agent is assigned a unique identity, enabling audit trails and access controls similar to those used in financial or HR systems. The platform enables centralized oversight so IT teams can monitor usage and enforce policies as agent adoption expands across organizations.
Broadcom Partnership for Observability
To offer customers greater IT visibility across increasingly complex AI environments, Google also expanded its partnership with Broadcom to debut Cloud Network Insights, which provides end-to-end observability across multi-cloud and agent-driven systems.
The offering integrates Broadcom’s AppNeta technology into Google Cloud’s infrastructure, giving enterprises a unified view of network and application performance. The goal here is to reduce diagnostic time and improve reliability by distinguishing between network and application issues. This is a growing requirement as organizations deploy AI agents and data-heavy workloads across multiple platforms.
Bottom Line: A Cross-Environment Control Layer
Among the Big Three cloud vendors, Google Cloud perennially runs third, behind AWS and Azure. At this year’s cloud event, the company has promoted integrations with third-party clouds and development tools. In essence, Google appears to be making a competitive move to position its cloud platform as a cross-environment control layer. This makes sense in an environment where AI systems must operate across a multi-cloud landscape of fragmented data and compute resources.
To boost adoption of its platform, Google also announced a $750 million fund to support partners building AI solutions. The initiative includes collaboration with consulting firms and early access to new models, demonstrating an effort to expand the surrounding ecosystem alongside its core technology.
