Google은 인프라 배포 역량, 클라우드 플랫폼 포괄성, 에코시스템 성숙도 측면에서 Gartner의 AI 인프라 Magic Quadrant를 선도합니다.

Google has been named a Leader in Gartner’s inaugural Magic Quadrant for AI Infrastructure, achieving the highest ranking for Ability to Execute and positioning furthest for Completeness of Vision. This recognition underscores Google Cloud’s commitment to advancing its AI infrastructure capabilities, particularly for organizations developing and deploying large-scale AI models and agent-based systems.

The assessment highlights Google’s extensive infrastructure development, built upon systems originally engineered for its own products, including Gemini, YouTube, and Search. According to the company, its AI stack is currently utilized by nine out of ten frontier AI labs, with prominent clients including Citadel Securities and Mercedes-Benz.

To further bolster its AI infrastructure, Google recently launched two new generations of tensor processing units (TPUs): the TPU 8t for training and the TPU 8i for inference. The TPU 8t supports connections of up to 9,600 chips within a single superpod, delivering nearly three times the compute performance per pod compared to its predecessor. Designed specifically for inference workloads, the TPU 8i features 288 GB of high-bandwidth memory and 384 MB of on-chip SRAM, effectively tripling the memory capacity of the prior generation.

Google maintains its collaboration with Nvidia to deliver GPU-based systems via Google Cloud and plans to introduce A5X instances powered by Nvidia’s Vera Rubin platform upon availability. The company is also expanding its investments in open-source orchestration and inference software, including llm-d and vLLM, while introducing TorchTPU to help PyTorch developers migrate workloads with minimal code modifications.

Gartner also acknowledged Google’s AI Hypercomputer, a unified architecture that consolidates hardware, networking, storage, and software into a cohesive environment for AI training and inference. As AI infrastructure expenditures continue to rise, Google notes that customers are increasingly prioritizing cost efficiency and higher resource utilization.

On the storage side, Google reports that Managed Lustre—leveraging C4NX instances and Hyperdisk Exapools—now delivers 10 TB/s of bandwidth. Meanwhile, Rapid Buckets can handle up to 20 million operations per second for object storage, significantly improving checkpointing and recovery processes for large-scale training runs.

Another critical component, the Virgo Network, can interconnect over one million TPUs across multiple data center locations within a single training cluster, or scale to 960,000 GPUs without performance degradation. For model serving, the Google Kubernetes Engine (GKE) Inference Gateway combines routing, caching, and disaggregated serving via llm-d, reportedly boosting throughput by up to 40% while cutting serving costs by up to 30%.

Google’s platform is engineered to support AI workloads across cloud, edge, and on-premises environments. Through Cluster Director and Google Kubernetes Engine, training workloads can scale to 130,000 nodes. The GKE Agent Sandbox can provision up to 300 sandboxes per second per cluster, dynamically adjusting capacity to match demand and reduce idle compute costs. Furthermore, the Cross-Cloud Network and Cloud WAN facilitate distributed enterprise and AI operations across diverse environments, leveraging Google’s private backbone network, which spans more than 10 million kilometers of fiber optic cable and extends to over 200 countries and territories.

This Gartner designation arrives as leading cloud providers and semiconductor manufacturers compete for dominance in the infrastructure market powering generative AI and agentic systems. Demand in this segment has shifted from a sole focus on raw compute power toward a holistic combination of silicon, networking, storage, orchestration software, and operational efficiency. Google’s position in the Magic Quadrant validates its integrated approach to AI infrastructure and reinforces its strategic emphasis on custom silicon as a foundational competitive advantage.