One caveat applies throughout. Nearly every performance, customer and market figure below came from the vendors in the briefings. None has been independently benchmarked, and an attendee's own roundup said as much.
What the tour is, and what the numbers say
StorageNewsletter's recap says that since the first edition in June 2010, the tour has met 372 distinct companies. It also reports 639 sessions, more than 1,600 C-level executives met, and more than 28,000 articles generated. Those are the organizers' cumulative figures, not audited ones.
An attendee's account in iTWire dates the tour week to October 5. It describes sessions of up to two hours with executives. Its common thread is that the GPU is now the most expensive item in the data centre. Vendors are being judged on whether they keep it busy, by cutting delays from data access, failures, recovery and memory limits. That framing explains why storage, memory and virtualization vendors all pitched the same story.
The Microsoft angle: CTERA and Copilot
CTERA announced Microsoft Copilot support on July 29, 2026, ahead of the tour. In its release, the company says it connects Microsoft 365 Copilot and Teams Copilot to file data it manages across edge locations, data centers and cloud environments. The link runs through the Model Context Protocol and CTERA Content Services.
Its stated design points are:
- Files stay where they are, and existing permissions are preserved.
- Copilot answers draw only on content the requesting user is authorized to access.
- The integration reaches beyond Microsoft 365 repositories such as SharePoint and OneDrive.
- Users get source-cited answers in Copilot and Teams through single sign-on.
All of this is CTERA's description of its own product. The announcement does not offer independent security validation or performance numbers. Admins who are rolling out Copilot should still test permission inheritance on their own file estate before trusting it.
At the tour, CTERA also described a broader "Unify, Protect, Understand and Activate" strategy. Its services include:
- InsightAI: metadata governance and analytics.
- Search: global indexing.
- Classify: labeling and workflows.
- Experts: semantic search and retrieval-augmented generation.
The recap gives anonymous customer examples:
- A manufacturer with more than 1PB across four regions.
- A mortgage lender activating 27TB of Azure Blob data.
- A medical-legal organization turning handwritten documents into case narratives.
CTERA also announced Forward Deployed Engineering. Small teams work inside customer environments, typically for about a month, starting with roughly ten specialists.
The Windows angle: VergeOS 26.2 and failover clusters
VergeIO is pitching VergeOS as a VMware alternative. Its 26.2 release is aimed at Windows clustered workloads that often stall a VMware exit. VergeOS's release notes describe shared (multi-writer) disks that let one disk attach to multiple VMs. That unlocks guest-clustered workloads such as Windows Failover Clustering and Cluster Shared Volumes. Shared disks present NAA/WWN identity and support SCSI-3 Persistent Reservations.
VergeIO says VergeFS pools the drives inside the servers into a vSAN. SQL Server failover instances, clustered file servers and Cluster Shared Volumes can then run on the same storage as other VMs. In practice, the pitch is that you may not need a separate storage array just to serve cluster disks.
Scope and limits
- Write coordination is left to the guest cluster. Windows Failover Clustering still does the arbitration inside the guests, as it would on any platform.
- Support does not establish compatibility with every Windows Server, SQL Server or third-party clustering configuration. Validate your own versions and workloads first.
- This is guest-cluster support on VergeOS. It is not a Microsoft endorsement.
Migration claims
VergeIO claims a 120% improvement in both initial and final VMware transfers over 26.1. It attributes the gain to asynchronous I/O in its network block device transfers. The datasheet says that is 2.2x the speed, measured in VergeIO's own lab over NBD. A separate VergeIO blog post expresses the same gain as about 55 percent less transfer time, so the two framings agree. Changed block tracking still limits the final transfer to blocks changed since the initial copy.
Systems on 26.1.6 or later can upgrade directly, one node at a time, while VMs move to other nodes. VergeIO also says restoring through an oVirt-compatible path keeps VM UUIDs, so existing Veeam jobs continue. The recap adds that VergeOS now integrates with Veeam Backup & Replication 13.1. Replication, continuous data protection and Instant Recovery remain unsupported there.
The tour was also where VergeIO's new CEO, Parag Patel, appeared. He took over on September 8, 2026, after leading worldwide vSAN sales at VMware. He estimates roughly 300,000 VMware customers could look for alternatives. The recap names Nutanix, Proxmox, Red Hat OpenShift Virtualization, Microsoft, HPE and Scale Computing among the rivals chasing them.
Cohesity and Microsoft Fabric
Cohesity described a strategy built on protection, security and AI. The recap cites about 12,000 customers and 200EB of protected data. Its Gaia and Gaia Catalog turn protected backup repositories into a semantic data layer for generative AI. Planned integrations include Databricks and Microsoft Fabric. Cohesity's claims of 80–90% lower token consumption and more than 40% lower AI infrastructure costs are its own. Agent Resilience initially targets Amazon Bedrock, so Microsoft platforms are not yet covered there.
The other briefings in brief
All figures here are company-reported.
| Company | Focus | Headline claims |
|---|---|---|
| Clockwork | FleetIQ monitoring and recovery for GPU clusters | 30–50% higher GPU utilization; 90% fewer disruptive checkpoint restarts; $31M raised, $73M total |
| DDN | EXAScaler for training, Infinia for inference | A 700PB Infinia deployment at a frontier AI lab, about 960 nodes, 6.4TB/s peak throughput |
| Hammerspace | Data orchestration across existing storage, plus "Tier 0" using NVMe inside GPU servers | A 500GB checkpoint in 4.5s locally versus 11.1s over a 400Gb/s network; Vanderbilt's roughly 10PB deployment reports about 48% lower storage cost |
| Phison | aiDAPTIV memory extension and Pascari enterprise SSDs | $2.345B revenue in 2025 and about $3.4B in the first half of 2026; a demo cutting memory use from 23.5GB to 15.9GB so a model runs on a 16GB laptop |
| ScaleFlux | Controllers and firmware instead of finished drives | FC6116 PCIe 6.0 controller and MC600 CXL 3.2 memory controller |
The Phison laptop result is a demonstration. It is not a guarantee for AI PCs generally. The recap says Phison is working with Intel on aiDAPTIV for AI PCs, and that eight PC makers were engaged by September 2026.
What to take from it
- Copilot data reach: Third-party file platforms are racing to become the governed source behind Copilot. If your files sit outside Microsoft 365, ask how permissions are enforced at query time.
- VMware exits: Clustered Windows workloads are the usual last holdouts. Treat shared-disk support as a reason to test, not a reason to commit.
- Vendor numbers: Treat every utilization, speed and savings figure as a hypothesis until you reproduce it.
- The common theme: Storage, memory and recovery vendors are all selling on GPU efficiency, and the memory and NAND squeeze is shaping product roadmaps.
For Windows and enterprise IT readers, the two items worth a closer look are CTERA's Copilot connector and VergeOS 26.2's clustering support. Both are real and documented by their vendors. Neither has been independently proven yet.
References
- Recap of the 70th IT Press Tour in Silicon Valley
- CTERA Extends Microsoft 365 Copilot to the Enterprise File Estate, with Governed, Classified and Searchable Data to Improve AI Quality and Outcomes | CTERA
- 70th IT Press Tour analysis: eight companies chasing the idle GPU, and why Philippe Nicolas and Céline Coustaut's tour has become an institution | iTWire