Google, Gemini local deployment extension… Can the contradiction between "control vs. performance" in enterprise AI infrastructure be resolved

html

Beyond generative AI, the era of “agent AI” capable of executing tasks independently has already arrived, and the standards for enterprise infrastructure design are rapidly changing. Especially in regulated industries, government agencies, and companies with strong data sovereignty requirements, there is a demand for an “AI-ready” infrastructure that allows the latest AI models to run within internal environments without needing to send sensitive data externally.

To meet these needs, Google is expanding its infrastructure strategy to enable enterprises to utilize their own AI models and cloud technologies in local environments. Muninder Sambi, Vice President and General Manager of Google Network and Security, stated at the recent Google Cloud Next event: “Businesses have always faced a choice—either comply with sovereignty and regulations or abandon these principles and move to the cloud. Google Distributed Cloud is precisely the solution that brings Gemini and Google’s AI capabilities into on-premises environments.”

Gemini Collaborates with NVIDIA and Dell, Expanding to Internal Networks

Google has partnered with NVIDIA ($NVDA) and Dell Technologies ($DELL) to support running Gemini-based models in air-gapped environments isolated from public networks and in connected on-premises environments. In particular, the Gemini Flash model now supports deployment on NVIDIA Blackwell B200 and B300 GPUs locally. For enterprises, this means they can run “sovereign AI” workloads without data leakage.

Sambi emphasized that this is not just about providing an “AI factory,” but about offering an “AI engine” that enables companies to build their own AI production systems. For financial, healthcare, defense, and public sector organizations that face difficulties using cloud services, this represents a significant and noteworthy transformation.

Kubernetes Rising as the Operating System for the AI Era

As local AI infrastructure expands, the role of Kubernetes becomes increasingly important. Drew Bradstock, Senior Product Director of Google Kubernetes and Google Compute Engine, commented that Kubernetes now acts as an “operating system” for AI tasks, covering training, inference, and reinforcement learning.

He explained that during the early proliferation of large language models, it was unclear whether Kubernetes could become the core control layer for AI. However, as the open-source ecosystem rapidly evolves to be more AI-friendly, it has become the foundation for running agents in hybrid cloud environments. In the context of enterprises deploying AI across multiple environments, the standardization benefits of Kubernetes are once again highlighted.

Now, users are no longer limited to humans

The spread of agent AI is not only changing how infrastructure operates but also transforming product design philosophies. Bradstock pointed out that the focus of developer experience no longer necessarily has to be humans. As AI agents significantly replace coding and operational tasks, documents, interfaces, and even tools are being redesigned into structures that are “easy for AI agents to read and use.”

He stated, “New DevOps are using Claude and Gemini to handle work,” and noted that user environments, documentation, and various tools are being reorganized around “skills.” This suggests that future enterprise software may go beyond UI designed for humans, evolving into structures that allow AI agents to directly call and execute functions.

Core Goal: Achieving “Maintaining Control” and “Ensuring Performance” Simultaneously

The core of this transformation is that enterprises no longer want to choose between data control and AI performance. AI-ready infrastructure is no longer just about server expansion but has shifted into a comprehensive design challenge that integrates data sovereignty, GPU computing resources, model deployment, governance, and orchestration.

From a market perspective, Google’s strategy indicates that cloud providers are once again delving deeply into local and hybrid cloud domains. As enterprises begin to adopt AI, the key to success is no longer solely about model performance but about whether AI can be deployed in real operational environments while ensuring security and flexibility.

TP AI Notes This article is a summary generated using a language model based on TokenPost.ai. It may omit key details or differ from actual facts.