Tech news in 3 minutes

Veeam Intelligent ResOps for Microsoft 365 Helps Customers Safely Use AI, Cut Storage Costs, and Recover With Precision

8 h ago

The 70th edition of The IT Press Tour took place in Silicon Valley early October. With this edition, new records have been beaten: 372 distinct companies met since the 1st edition in June 2010, organized 639 sessions, met 1600+ CxO for more than 28k articles generated.Designed as a press-focused event, it brings together media representatives and a diverse range of enterprises and organizations for in-depth conversations spanning IT infrastructure, cloud computing, networking, cybersecurity, data management and storage, big data and analytics, and the growing role of AI across all these fields. Eight companies joined that edition, listed here in alphabetical order: Clockwork, Cohesity, CTera, DDN, Hammerspace, Phison, ScaleFlux and VergeIO. Clockwork During its first participation, Clockwork.io presented its vision of Software-Driven AI Fabrics, a programmable software layer designed to improve GPU cluster observability, fault tolerance and utilization. Led by CEO Suresh Vasudevan and a founding team with strong Stanford and VMware roots, the company recently raised $31M, bringing total funding to $73M. Clockwork addresses a critical challenge in large AI infrastructures: GPU failures, network disruptions and performance stragglers can significantly reduce effective computing capacity. Distributed AI workloads synchronize thousands of GPUs, meaning a single faulty component can slow or interrupt an entire job. Industry studies highlight the problem, with Meta reporting a mean time to failure of just 7.9 hours for 1,024-GPU jobs and ByteDance observing significant slowdowns in 42.5% of jobs. Click to enlarge Traditional checkpoint-restart mechanisms can require up to 90 minutes to recover, wasting valuable GPU resources. Clockwork positions fault tolerance as a “goodput multiplier,” maximizing productive GPU processing rather than simply increasing hardware utilization. Its flagship FleetIQ platform, generally available since September 2025, combines detection, diagnosis and automated recovery across AI training, inference and reinforcement learning. Clockwork claims 30–50% higher GPU utilization and 90% fewer disruptive checkpoint restarts. FleetIQ integrates several complementary technologies. FleetAudit validates GPU clusters before deployment, examining networks, accelerators, PCIe topology and communication infrastructure. FleetLens continuously monitors production environments using Clockwork’s highly accurate synchronization technology and an out-of-band probing system, identifying network problems at microsecond granularity and pinpointing faulty links, switches or NICs. On the recovery side, LinkPass, an NCCL plugin, automatically reroutes traffic around failing network interfaces without interrupting workloads. TorchPass introduces live GPU workload migration, transferring running jobs from faulty or degraded nodes to healthy resources without restarting applications. It supports both proactive maintenance and unexpected hardware failures. Another innovation is TorchPass snapshots, which capture distributed workloads without requiring application modifications. Platform snapshots complete in under 20 seconds, while integrated model checkpoints take under five seconds, both with approximately 1% overhead at their respective recommended frequencies. Clockwork also demonstrated growing commercial traction. LinkedIn uses LinkPass in production, reportedly avoiding tens of thousands of GPU-hours of downtime monthly. Together AI is integrating TorchPass and LinkPass into its GPU cloud services, while WhiteFiber uses FleetAudit to detect infrastructure defects before customer deployment. Additional customers and partners include Wells Fargo, Nebius, Nscale and DCAI. The broader market implication is significant. As AI clusters expand to tens of thousands of accelerators, infrastructure reliability becomes as important as raw GPU performance. Improving goodput can deliver substantial economic benefits by reducing wasted computing cycles, recovery delays and operational intervention. Clockwork’s software-driven approach could therefore become an important component of next-generation AI factories, helping operators extract greater value from increasingly expensive GPU investments while improving service reliability and operational efficiency. Cohesity During its fifth participation, Cohesity presented its latest corporate strategy and product developments, with CEO and President Sanjay Poonen and CPO Vasu Murthy outlining a vision centered on three pillars: Protection, Security and AI. The company’s mission has evolved toward protecting, securing and extracting AI insights from enterprise data, under the broader ambition of “Resilience Everywhere”. With approximately 12,000 customers, primarily among Global 2000 organizations, and 200EB of protected data, Cohesity positions itself at the intersection of cybersecurity and data management. Poonen compared its strategy to combining CrowdStrike or Palo Alto Networks with Databricks or Snowflake. The company is profitable, while a potential IPO remains on the horizon without a confirmed timeline. Click to enlarge The first pillar, protection, follows a “no workload left behind” philosophy, covering virtual machines, databases, files, identities, SaaS applications and AI agents. Eighteen months after merging with Veritas, Cohesity claims the industry’s broadest workload connector portfolio, emphasizing rapid recovery as a key differentiator. The second pillar, security, introduces a five-stage cyber resilience framework encompassing comprehensive protection, cyber vaulting, threat scanning, clean-room recovery and the Minimum Viable Company concept. This approach prioritizes restoring critical business operations following cyberattacks rather than relying exclusively on threat prevention. Cohesity is also strengthening its cybersecurity capabilities through advanced AI models. Its participation in Anthropic’s Project Glasswing enables vulnerability analysis using Mythos, complemented by evaluations involving OpenAI, Google and Nvidia technologies. Poonen expects new security products leveraging these models in the coming months. The third pillar, AI, builds on the Cohesity Data Cloud, combining its control plane, NetBackup and DataProtect data movers, and SpanFS architecture. Murthy emphasized that trustworthy enterprise AI requires collaboration between autonomous agents, deterministic systems and human oversight. Three major product initiatives support this strategy. Cohesity Maestro introduces AI-powered operational capabilities for resilience teams. Through MCP integration, agents can identify protection gaps, recommend corrective actions and execute approved operations while maintaining human control over privileged activities Agent Resilience and Data Security Posture Management (DSPM) extend protection to AI agents, including their configurations, memory, tools and vector stores. Initial support targets Amazon Bedrock, with near-continuous backups approximately every five minutes. Future integrations with ServiceNow and Datadog will automate recovery following suspicious activities or destructive events Cohesity Gaia and Gaia Catalog transform protected backup repositories into a semantic data layer for generative AI. Using zero-copy access, Cohesity claims 80–90% lower token consumption and more than 40% reductions in AI infrastructure costs. Planned integrations include Databricks and Microsoft Fabric, while the Dutch government agency Rijkswaterstaat illustrates Gaia’s on-premises deployment capabilities The strategic implication is clear: Cohesity wants to move beyond traditional backup and recovery toward an integrated data security and AI platform. Its extensive protected data footprint represents a competitive asset, potentially turning immutable backup repositories into valuable enterprise knowledge sources while reinforcing cyber resilience in an increasingly agent-driven world. CTera During its ninth participation, CTera presented its latest strategy and innovations, with co-founder and CEO Oded Nagel and co-founder and CTO Aron Brand highlighting the company’s evolution toward an AI-ready enterprise data platform. Founded in 2008, CTera employs approximately 200 people and has raised $180M. The presentation focused on the “AI Value Gap”, the disconnect between massive enterprise AI investments and limited measurable returns. Despite widespread AI adoption, many organizations remain stuck in pilot projects. CTera identifies three major obstacles: data readiness, security and skills shortages. Its central argument is that enterprise AI challenges originate primarily from data management rather than AI models themselves. Click to enlarge CTera advocates bringing AI to governed enterprise data instead of moving sensitive information into external AI environments. This approach preserves existing permissions, strengthens sovereignty and reduces unnecessary data transfers and infrastructure costs. The company’s updated positioning revolves around four pillars: Unify, Protect, Understand and Activate. The first two represent CTera’s traditional strengths in global file systems, distributed collaboration, snapshots and cyber resilience. The newer capabilities, Understand and Activate, introduce metadata intelligence, semantic search, classification and AI-driven content services. The architecture combines the CTera Portal control plane, Edge Filers, endpoint applications and object storage, complemented by several AI services. InsightAI delivers metadata-based governance and analytics, CTera Search provides global indexing, CTera Classify automates labeling and workflows, while CTera Experts enables secure semantic search and RAG. Through MCP integration, these services connect enterprise data to Microsoft Copilot, Claude, Gemini, Cursor and proprietary AI agents. CTera anticipates that AI agents will eventually generate most enterprise data access reques