At its recent Next ’26 conference, Google Cloud advanced its storage portfolio across three fronts: high-performance infrastructure for AI training and inference, intelligent metadata capabilities built directly into the storage layer, and expanded ecosystem integrations. 

The announcements span Cloud Storage Rapid, Google Cloud Managed Lustre with 10 TB/s throughput, Hyperdisk Exapools, Smart Storage automation, and material expansions to the Google Cloud NetApp Volumes service, including a deeper partnership with NetApp.

GCP’s storage portfolio now offers a coherent AI-first architecture that Google Cloud frames as an active component of the AI stack, feeding accelerators during training, serving context during inference, and enabling agentic workflows with enriched metadata.

Beyond its core storage infrastructure updates, the NetApp partnership is the most significant ecosystem among the storage announcements. With Flex Unified service reaching general availability across all Google Cloud regions, Google Cloud NetApp Volumes now supports unified block and file storage, offers a single pool, provides native ONTAP API compatibility in the cloud, and integrates directly with Gemini Enterprise.

In addition, NetApp’s internal adoption of Gemini Enterprise, announced alongside the technical updates, provides a practitioner endorsement that reinforces the business-level narrative.

Google Announcements

The storage announcements at Next ’26 span the full stack of object, block, file, and intelligent metadata, with AI workload performance as the primary design axis. Let’s look at each in turn.

Cloud Storage Rapid

Cloud Storage Rapid is a new family of high-performance object storage capabilities built on Colossus, Google’s internal distributed storage system. The Rapid family consists of two offerings:

• Rapid Bucket targets AI training workloads requiring extreme throughput, delivering more than 15 TB/s of aggregate bandwidth, 20 million requests per second, and sub-millisecond latency within a single zonal bucket. Google says Rapid Bucket reduces GPU blocked time by 50% compared with traditional object storage and accelerates checkpoint writes by 3.2x and checkpoint restores by 5x. The service exposes both high-performance gRPC and S3-compatible APIs.
• Rapid Cache (formerly Anywhere Cache) accelerates bandwidth for bursty workloads, including model loading for inference, delivering 2.5 TB/s of aggregate read throughput for existing buckets without code changes. A new ingest-on-write feature enables up to 2.2x faster checkpoint restores. Rapid Cache includes native integrations with PyTorch and JAX.

Google Cloud Managed Lustre

Google Cloud Managed Lustre is a fully managed implementation of the Lustre parallel file system, targeting AI training and HPC workloads. The Next ’26 announcements expanded its throughput and introduced a new pricing tier:

• Throughput increase: Google’s Managed Lustre now delivers up to 10 TB/s, a 10x increase since 2025 and 4–20x higher than managed Lustre offerings from other hyperscalers for a single instance. The service runs on C4NX VMs and Hyperdisk Exapools.
• Checkpoint performance: Managed Lustre writes and restores checkpoints 2.6x faster than other Google Cloud storage solutions.
• Dynamic tier: A new Dynamic tier, priced at $0.06/GB-month, serves data from persistent disk rather than relying on object-based caching, delivering low latency for training and checkpointing without performance cliffs or the complexity of tiered billing.

Hyperdisk Exapools and Z4M

Google announced general availability for Hyperdisk Exapools and previewed the Z4M VM class, both targeted at large-scale AI training clusters:

• Hyperdisk Exapools: Exapools offer the highest aggregate block storage performance and capacity per AI cluster of any hyperscaler. Hyperdisk ML now delivers 2 TiB/s of aggregate throughput, up from 1.2 TiB/s, providing more than 200x the throughput per disk of what Google characterizes as competitive offerings.
• Z4M (preview, Q3 2026): Z4M virtual machines and bare-metal instances provide up to 168 TiB of local SSD storage, up to 400 Gbps of network bandwidth, and RDMA support. Z4M is designed for distributed parallel file systems and large-scale AI/ML workloads. It also integrates with Cluster Director for co-location with accelerators.
• Z4D (preview): Z4D instances are designed for I/O-intensive database workloads, providing up to 84 TiB of local SSD storage and 400 Gbps of VM-to-VM bandwidth.

Smart Storage and Storage Intelligence

Google Cloud introduced automated annotation capabilities and an MCP server integration as part of its Smart Storage initiative, aiming to make stored objects self-describing and directly accessible to AI agents:

• Automated annotations: Smart Storage can now automatically generate context for objects (including image annotations) at write time, eliminating the need for custom annotation pipelines. Object context, now generally available, provides a structured metadata layer that is IAM-governed, mutable, and shared across downstream systems.
• Cloud Storage MCP server: A new MCP server integration enables AI agents to read, write, and analyze Cloud Storage data directly, thereby connecting storage to the broader agentic workflow stack.
• Storage Intelligence: New capabilities include zero-configuration dashboards and enhanced batch operations for large-scale data management. Google reports that 70% of its largest customers now use Storage Intelligence, and each manages more than 50 billion objects.

Filestore for GKE

Google Cloud updated Filestore for Kubernetes environments, allowing developers to provision shares as small as 100 GiB and scale capacity and IOPS independently. Tighter integration with Colossus provides additional enterprise-scale capabilities for AI workloads running on Google Kubernetes Engine.

Google Distributed Cloud Storage Expansion

Google Distributed Cloud (GDC) has increased its storage capacity, now supporting 6 PB of object storage per zone, a 6x increase over its previous capacity. IOPS performance increased to 30 IOPS/GB per zone, a 10x improvement. These updates extend Google Cloud’s high-performance storage capabilities to edge and sovereign cloud deployments.

The Google-NetApp Partnership

The NetApp-related announcements at Next ’26 cover both technical product updates and the expansion of a strategic partnership:

• Flex Unified: Google Cloud NetApp Volumes Flex Unified is now generally available across all Google Cloud regions. The service provides both block storage (iSCSI, NVMe/TCP) and file storage (NFS, SMB) from a single storage pool, enabling enterprise applications, databases, HPC, electronic design automation, and VMware workloads to run in the cloud without rearchitecting or rebuilding environments.
• ONTAP mode: A new ONTAP-mode capability enables customers to bring existing automation tooling (e.g., Terraform, Ansible, and native ONTAP APIs) directly to Google Cloud NetApp Volumes, reducing migration complexity for organizations with deep ONTAP operational investments.
• NetApp Data Migrator: A multi-cloud data migration service that moves data across environments without specialized expertise, lowering the technical barrier for enterprises migrating workloads to Google Cloud.
• Gemini Enterprise integration: The NetApp Volumes data connector for Gemini Enterprise enables customers to use enterprise data stored in NetApp Volumes directly for agentic AI workloads without moving or duplicating data.
• NetApp internal adoption: NetApp announced it has adopted Gemini Enterprise internally across product development and sales operations. NetApp has also been using Google Security Operations for cybersecurity and threat management.
• Partner award: NetApp received Google Cloud’s 2026 Infrastructure Modernization Partner of the Year for Storage award, the seventh time the company has received that recognition.

Analysis

The Next ’26 storage announcements have direct operational implications for the infrastructure teams building and running AI workloads at scale. The shift from treating storage as passive infrastructure to treating it as an active component of the AI pipeline is well supported by the announced technical capabilities:

• AI training teams: Cloud Storage Rapid Bucket and Managed Lustre address the two most common storage bottlenecks in large-scale training, checkpoint write/restore latency and accelerator underutilization from data starvation. The performance should meaningfully reduce total training time and GPU idle cost.
• Inference and agentic workloads: Rapid Cache and the Smart Storage MCP server directly address latency and data accessibility requirements for inference and agentic deployments. The ability to access enriched metadata at the storage layer without building separate retrieval pipelines reduces the engineering overhead for RAG and agent memory architectures.
• Enterprise IT teams migrating to Google Cloud: The Flex Unified GA and ONTAP mode for Google Cloud NetApp Volumes reduce the operational friction of moving enterprise NAS workloads to the cloud. Organizations with established ONTAP expertise and automation can preserve those investments during migration, with the NetApp Data Migrator GA further simplifying the operational path.
• Kubernetes-native developers: The Filestore for GKE updates reduce minimum provisioning requirements, making Google Cloud storage accessible for smaller AI experiments and dev/test workloads.
• Capacity planning: The Hyperdisk ML throughput increases to 2 TiB/s, and Exapools GA give infrastructure architects higher aggregate storage bandwidth budgets per cluster, affecting how they scale training infrastructure without hitting storage ceilings.

Competitive Landscape

Google Cloud’s storage portfolio competes with AWS and Microsoft Azure across multiple dimensions. The competitive picture varies significantly by storage tier and use case, and Google holds genuine leads in some areas while trailing meaningfully in others. The table below provides a structured comparison across the dimensions most relevant to enterprise cloud storage decisions in 2026.

Overall market context: AWS holds approximately 31% of the global cloud infrastructure market, Azure approximately 24%, and Google Cloud approximately 12%, based on early-2026 industry estimates. The storage market broadly reflects these relative positions, with AWS and Azure each carrying substantially larger installed customer bases. Google Cloud’s storage announcements at Next ’26 are calibrated to close that gap specifically in AI-intensive workloads rather than across the full enterprise IT spectrum.

Legend:  ▲ Ahead  (green) = Google Cloud leads    = Parity  (yellow) = comparable capabilities    ▼ Behind  (orange) = AWS or Azure leads

Final Thoughts

Google cloud storage announcements show the execution of a mature, coherent product strategy.

Google Cloud’s throughput advantages in managed Lustre and AI-native object storage are real, but they apply primarily to training and inference workloads that define its target segment. The broader enterprise storage market, including Windows file integration, hybrid cloud reach, archive pricing, and installed base depth, remains dominated by AWS and Azure.

The NetApp partnership expansion, with Flex Unified reaching GA in all regions and ONTAP mode preserving existing operational investments, adds enterprise file and block credibility that Google Cloud has historically lacked compared with AWS and Azure.

The most significant differentiation in the Next ’26 storage portfolio is the Smart Storage intelligent metadata layer and the MCP server integration. Automated object annotations that make stored data immediately usable by AI agents, without custom pipeline development, address the data accessibility challenge enterprises face when deploying agentic AI.

If that capability proves durable and performant at enterprise scale, it provides Google Cloud with a long-term differentiation that is more defensible than throughput metrics.

For enterprises evaluating AI infrastructure, where Google Cloud is already under consideration, the Next ’26 storage announcements strengthen the case for standardizing on Google Cloud as the primary AI training and inference platform.