AI dominance is shifting from models to 'agent infrastructure'... Google's Cloud's ace move

Some analyses point out that Google’s cloud focus in the AI race has shifted from “model performance” to the “control layer” and data infrastructure. As enterprise AI accelerates its adoption, industry assessments believe that the true determinant of victory or defeat will likely depend on “intelligent AI infrastructure” that enables agents to read, connect, and execute data.

Chief Analyst John Friel of The Cube Research diagnosed during the Google Cloud Next 2026 conference that Google is aiming to become an intelligent agent-based enterprise operating system. He believes that mastering the “control plane” that connects data and various systems will determine market advantage. This layer functions like a neural network, responsible for linking multiple internal applications and data streams within enterprises.

On-site, the speed of AI adoption has significantly increased. Some companies even evaluate that in coding tasks, machines now account for more than humans. However, not all organizations are progressing at the same pace. Many have yet to determine which business areas will benefit most from prioritized AI application. This situation further highlights the importance of co-designing intelligent agent-based AI infrastructure for data, security, governance, and execution environments.

The core is “contextual data”… More important than accurate information is “the most appropriate information.”

Google Cloud regards “contextual data” as key to overcoming AI model limitations. Google Cloud Database Engineering Lead Sailesh Krishnamurthi explained that while models are very powerful, the actual context needed for enterprise operations resides in the data. This means that answering questions is not simply about inputting large amounts of information, but about precisely extracting the “information needed at the moment.”

Therefore, some believe that the next-generation data cloud must differ from existing databases. It needs to go beyond simple storage and query functions, handling graph search, vector embeddings, full-text retrieval, and relational operations within a single system. Only by providing optimal results while minimizing data movement can intelligent agent-based AI infrastructure operate well in large-scale enterprise environments.

OpenText also agrees with this view. OpenText is collaborating with Google Cloud to build an intelligent agent stack based on contextual engineering, data sovereignty, and open interoperability. OpenText emphasizes that enterprise information is not just simple file storage but a system that includes classification, tagging, governance, and connection to business processes. To supply this information to large language models, unnecessary data flooding must be prevented, and only the required information should be transmitted at appropriate times.

Google is expanding its industry solutions by deeply integrating the Gemini enterprise-grade intelligent agent platform, enabling secure utilization of decades of accumulated enterprise documents. This again shows that the success of AI adoption depends more on data quality and access architecture than on the flashy aspects of models.

Hybrid environments and cost reduction… Partnerships determine AI infrastructure efficiency.

The competition for AI infrastructure cannot be decided by a single company alone. Google is partnering with NVIDIA, Dell Technologies, AMD, and others to build “AI-ready infrastructure” spanning cloud and on-premises deployments. Especially considering many enterprises find it difficult to entrust all data to external clouds due to security and regulatory concerns, Google’s distributed cloud supports enterprises in leveraging Gemini’s capabilities within local environments.

In this process, Kubernetes plays an increasingly important role. Google views Kubernetes as an almost operating system for AI, covering learning, reasoning, and reinforcement learning. This means Kubernetes is the core tool for coordinating AI agents dispersed across various environments. As enterprises expand AI across multi-cloud or hybrid architectures, this orchestration layer is indispensable.

Cost is also a significant variable. AMD explained that for many enterprise customers using both their own data centers and cloud environments, x86-based infrastructure remains the most practical alternative. Container workloads can be easily migrated between the two environments, and performance and cost-efficiency can be maintained without code modifications.

U.S. travel tech company Sabre reportedly migrated over 50k virtual CPUs to Google Cloud based on AMD instances, achieving both cost savings and performance improvements. The company states that without code changes, they obtained faster processing with a smaller infrastructure footprint, and the savings were reinvested into intelligent agent AI investments.

Google is also heavily investing in ecosystem expansion. To strengthen its Google Cloud partner network, it plans to invest $750 million (about 1.107375 trillion KRW) to boost the performance of over 120k member companies. Google is also conceptualizing a structure that enables partner agents to interact with Google-side systems, automating onboarding, training, and recommendation tasks.

AI applied to “difficult problems” can be effective… Going beyond simple automation to improve customer experience.

McKinsey suggests that the more companies feel their AI investments are underperforming, the more they should target bigger issues. Senior Partner Ashutosh Padi states that to create significant value, one must start with topics that can shake the entire enterprise. Because even successful small-scale experimental projects may not attract full organizational attention, solving core business problems can lead to change management and capability building.

California’s largest state health insurance marketplace, Covered California, is cited as an example. The agency partnered with Deloitte and Google to introduce Google Document AI, greatly automating eligibility determination and registration verification. It is estimated to save about 24k hours of operational time annually. Previously, document verification took up to 72 hours; now it can be completed in seconds.

This not only reduces labor costs. Evaluations indicate that employees are freed from repetitive paperwork, allowing them to focus on more valuable customer support, thereby improving service quality and customer experience. Deloitte interprets that AI is not meant to replace humans but to serve as a tool that enables people to focus on more human-centric work.

The message from Google Cloud Next 2026 is clear. The core of AI competition is no longer just “who built a smarter model.” In the enterprise market, the real winning factor is intelligent agent-based AI infrastructure that combines data context, control layers, hybrid infrastructure, cost efficiency, and partner ecosystems. In the future, the success or failure of AI deployment will likely depend more on whether models can reliably and flexibly connect with enterprise environments than on the models themselves.

TP AI Notice: This article is summarized based on TokenPost.ai’s language model. The main content may be incomplete or not fully aligned with facts.