Research Finder
Find by Keyword
Google Cloud Next 2026: Storage Re-enters the AI Performance Path
Google Cloud shows that storage is no longer a passive capacity layer in AI environments, as throughput, metadata intelligence, and low-latency data delivery become central to training, inference, and agentic workloads.
04/22/2026
Key Highlights
- Google Cloud introduced new storage capabilities including Managed Lustre delivering up to 10 TB/s of throughput to TPU 8t and A5X environments over RDMA.
- Rapid Storage performance increased from 6 TB/s to 15 TB/s, supporting faster AI training and inference workloads.
- Smart Storage adds semantic understanding to unstructured data, forming part of Google’s emerging Enterprise Knowledge Graph strategy.
- Google introduced the Z4M instance for customers building custom storage architectures, supporting trusted parallel file systems such as VAST Data and Sycomp with up to 168 TiB of local SSD capacity per instance.
The News
At Google Cloud Next 2026, Google expanded the storage layer of its AI Hypercomputer architecture with higher-performance file services, faster shared storage, and new metadata intelligence capabilities. The updates position storage as an active contributor to AI workload performance rather than a background repository. Google highlighted Managed Lustre, Rapid Storage, and Smart Storage as key elements supporting training, inference, and agentic AI environments. For more information, see the official Next 2026 announcements blog.
Analyst Take
Much of the AI infrastructure conversation remains centered on accelerators, model sizes, and networking fabrics. That view misses an increasingly important constraint: accelerators only create value when storage platforms can feed, checkpoint, retrieve, and recover workloads without delay. Bottlenecks can strand expensive compute resources just as quickly as a networking fault or node failure.
Google’s announcements indicate that storage is moving back into the critical path of AI system design. Managed Lustre throughput of 10 TB/s over RDMA to TPU and GPU environments signals continued demand for parallel file architectures capable of feeding large-scale training clusters. The Rapid Storage jump from 6 TB/s to 15 TB/s suggests Google is also tuning shared storage for mixed training and inference workloads where concurrency and responsiveness matter.
Google’s broader briefing also shows the company investing across multiple storage tiers simultaneously, including parallel file systems, cloud object storage with embedded semantic services, and Hyperdisk block storage tuned for transactional and analytics workloads. That breadth suggests Google views storage architecture as a portfolio requirement for enterprise AI, not a single-product category.
The more strategic announcement may be Smart Storage. By applying semantic meaning to unstructured data and tying that capability to an Enterprise Knowledge Graph, Google is moving storage beyond performance into context creation. That points to a future where storage platforms participate directly in metadata enrichment, retrieval workflows, and grounding data for enterprise agents.
This aligns with HyperFRAME Research Lens (1H 2026) findings showing only 14% of organizations report fully AI-ready data architectures, while 50% cite scalability as the primary barrier to expanding AI initiatives. Many enterprises are discovering that compute acquisition is easier than preparing data environments that can sustain production AI workloads.
Google is not alone in this direction. AWS continues to evolve high-performance storage and S3-adjacent data services, Microsoft pairs Azure storage with Fabric and analytics layers, and specialist vendors such as Dell, NetApp, Pure Storage, VAST Data, Hammerspace, and WEKA are all competing to place storage closer to the AI execution path. The market opportunity is shifting from capacity supply toward data delivery, metadata control, and usable throughput.
Google also demonstrated openness in the storage layer through the new Z4M instance, engineered for customers and ISVs integrating parallel file systems such as VAST and Sycomp. That suggests Google recognizes many advanced AI environments will remain heterogeneous and may prefer specialist data architectures over single-vendor standardization.
Google Cloud AI Hypercomputer: Storage Becomes a Semantic Fabric
From our perspective, Google Cloud’s strategic pivot with these announcements indicates that the AI Hypercomputer is evolving from a raw performance engine into a data fabric designed for agent workflows. By integrating semantic metadata directly into the storage layer through Smart Storage and the Object Context API, Google can eliminate the pre-processing tax that typically delays AI agent deployment. This move indicates that the future of storage isn't just about moving bytes at 15 TB/s, but about ensuring those bytes are agent-ready the moment they are written to disk.
While the jump to 15 TB/s in Rapid Storage is notable, the larger innovation may be the Enterprise Knowledge Graph integration. By embedding business meaning closer to stored data, Google is shortening the path between raw information and grounded AI responses, reducing reliance on traditional ETL cycles.
By engineering the Z4M instance specifically for third-party parallel file systems such as VAST and Sycomp, Google is acknowledging that the frontier AI environments remain diverse. This open-stack approach can help Google capture demand while allowing customers to retain preferred data architectures. The Object Context API further suggests that Google views the storage bucket as an active participant in AI workflows, continuously enriching data for downstream use.
We find that the Cross-Cloud Lakehouse on Apache Iceberg represents a tactical strike against data gravity by allowing customers to query data in AWS and Azure without relocation. This helps position BigQuery and Gemini as a universal control plane for AI, based on the premise that once organizations can view more of their data through Google’s semantic lens, they may choose to place a greater share of AI execution on AI Hypercomputer. Together, these developments shift the focus from simple throughput to contextual access and delivery speed, making storage a more strategic foundation for agentic workflows.
What Was Announced
Google used Next 2026 to outline storage enhancements across both its AI Hypercomputer and Agentic Data Cloud portfolios, pairing performance gains with new metadata services built for AI workloads.
Within AI Hypercomputer, Google said Managed Lustre now delivers up to 10 TB/s of throughput to A5X GPU and TPU 8t environments over RDMA connectivity. The positioning centers on high-bandwidth shared file access for distributed training jobs where large numbers of accelerators must read common datasets and checkpoints at speed.
The new Z4M instance is designed to help ISVs and organizations build custom storage solutions. Google said each Z4M instance supports up to 168 TiB of local SSD capacity and can be deployed in RDMA clusters spanning thousands of machines. The company specifically cited support for trusted parallel file systems such as VAST and Sycomp.
Google also disclosed significant performance gains for Rapid Storage, increasing aggregate throughput from 6 TB/s to 15 TB/s. The company tied these gains to both training and inference workloads, indicating Rapid Storage is being positioned for environments that need fast data access across mixed AI pipelines rather than only long-running model training clusters.
Smart Storage adds automated annotation, entity extraction, and semantic search directly within stored objects, allowing agents to locate and use data across spreadsheets, PDFs, and other unstructured formats. The service appears designed to enrich stored files with contextual metadata that can later be used by search, analytics, and agentic workflows. Google linked Smart Storage to its broader Enterprise Knowledge Graph strategy.
Supporting that direction, Google introduced an Object Context API that can tag and enrich files stored in Google Cloud Storage before they are accessed by AI agents or applications. Google also paired this with Knowledge Catalog, a service intended to create a unified context graph across enterprise data sources, business definitions, and relationships.
Google announced a Cross-Cloud Lakehouse built on Apache Iceberg. The offering is intended to let customers query data residing in AWS and Azure without requiring broad data relocation, while extending access to downstream analytics platforms and AI tools.
Google further introduced a Data Agent Kit for Gemini-powered data science workflows across notebooks, IDEs, and terminals. The company positioned the tool as enabling intent-driven development across Python, Spark, and SQL, suggesting tighter integration between storage-resident data and AI-assisted engineering workflows.
Google also highlighted M4N instances paired with Hyperdisk Extreme, positioned to deliver high per-core IOPS and throughput for agent-driven data pipelines, analytics, and mission-critical databases. Google cited more than 20% lower Oracle workload total cost of ownership versus leading hyperscale alternatives.
Looking Ahead
We believe these storage-related announcements are best understood as part of Google Cloud’s broader Next 2026 strategy to present a unified enterprise AI stack spanning infrastructure, data, security, productivity, and agents. In that model, storage becomes the persistent data and context layer that supports every higher-level service Google introduced across Gemini Enterprise, Agent Platform, and the Agentic Data Cloud.
Managed Lustre and Rapid Storage show Google treating throughput and latency as first-order AI infrastructure constraints. Smart Storage, Knowledge Catalog, and Object Context APIs point to a longer-term ambition: turning enterprise data into structured, discoverable, policy-aware context that agents and applications can use directly. That places storage closer to orchestration, retrieval, and workflow execution than traditional infrastructure buyers may expect.
The Z4M announcement also indicates Google is willing to compete through ecosystem depth as well as native services. Supporting specialist partners may prove important in large enterprise and frontier AI deployments where customers prioritize architectural choice. Google is also extending this model into sovereign and on-prem environments through Google Distributed Cloud, where new capacity and performance improvements aim to support AI reasoning close to regulated data.
The wider market is moving in a similar direction. Competitors are increasingly blending storage, metadata, governance, and analytics into broader AI platform offers. Google’s advantage is that it can connect those layers to first-party models, custom silicon, global cloud infrastructure, and a growing enterprise productivity footprint.
In our view, the implication is that future enterprise storage evaluations will need to extend beyond cost, capacity, and throughput. Questions around metadata enrichment, agent access controls, cross-cloud data mobility, and integration with AI development pipelines are becoming equally important.
If Google executes well, these announcements could help strengthen its position from cloud provider to an enterprise AI control surface, with storage serving as a foundational control point beneath that strategy.
Don Gentile | Analyst-in-Residence -- Storage & Data Resiliency
Don Gentile brings three decades of experience turning complex enterprise technologies into clear, differentiated narratives that drive competitive relevance and market leadership. He has helped shape iconic infrastructure platforms including IBM z16 and z17 mainframes, HPE ProLiant servers, and HPE GreenLake — guiding strategies that connect technology innovation with customer needs and fast-moving market dynamics.
His current focus spans flash storage, storage area networking, hyperconverged infrastructure (HCI), software-defined storage (SDS), hybrid cloud storage, Ceph/open source, cyber resiliency, and emerging models for integrating AI workloads across storage and compute. By applying deep knowledge of infrastructure technologies with proven skills in positioning, content strategy, and thought leadership, Don helps vendors sharpen their story, differentiate their offerings, and achieve stronger competitive standing across business, media, and technical audiences.
Ron Westfall | VP and Practice Leader for Infrastructure and Networking
Ron Westfall is a prominent analyst figure in technology and business transformation. Recognized as a Top 20 Analyst by AR Insights and a Tech Target contributor, his insights are featured in major media such as CNBC, Schwab Network, and NMG Media.
His expertise covers transformative fields such as Hybrid Cloud, AI Networking, Security Infrastructure, Edge Cloud Computing, Wireline/Wireless Connectivity, and 5G-IoT. Ron bridges the gap between C-suite strategic goals and the practical needs of end users and partners, driving technology ROI for leading organizations.