You know what's fascinating about technology cycles? Just when everyone thinks they understand where the industry is heading, a collection of companies emerge with solutions that make you question everything you thought you knew about enterprise infrastructure.

That's exactly what happened during the 62nd Edition of The IT Press Tour in California, where VMblog met with nine companies that, on the surface, seemed to be solving completely different problems. Lucidity was automating cloud storage management. Hunch was building AI workflow tools. DDN was powering massive GPU clusters. Yet after analyzing all nine briefings, a clearer picture emerged: these companies represent five major shifts that could fundamentally alter how enterprises operate over the next two years.

Here's what we learned, and why it matters for anyone managing enterprise technology infrastructure.

The AI Infrastructure Arms Race Gets Serious

The most striking theme from this year's tour was the sheer scale of AI infrastructure deployment. DDN's Paul Bloch shared numbers that sound almost fictional: they're currently managing 700,000+ GPUs, with expectations to quadruple that within two years. The largest single deployment? Elon Musk's xAI facility in Memphis with 200,000 GPUs-what Bloch described as one of the first true "AI factories."

But here's where it gets interesting for enterprise IT teams. Three companies-DDN, Graid Technology, and Phison-are attacking the same fundamental bottleneck from different angles: storage can't keep up with GPU computing power.

DDN operates at hyperscale, powering the infrastructure behind companies like Meta and Tesla. Their EXAScaler platform delivers sustained performance across tens of thousands of GPUs while maintaining enterprise-grade reliability. When Jensen Huang says "NVIDIA is powered by DDN," that's not just marketing-it's validation of their technical approach.

Graid Technology targets the middle market with their SupremeRAID GPU-accelerated storage. Instead of treating storage and compute as separate systems, they use existing GPU infrastructure to accelerate storage operations. The result? Over 95% of raw NVMe performance while providing enterprise RAID protection. More importantly, they eliminate the complexity that traditionally required vast networking infrastructure-replacing 1,400 cables with just 40 in some deployments.

Phison takes the democratization approach with their aiDAPTIV+ platform. Rather than requiring massive GPU investments, they use flash storage as an extension of GPU memory, delivering 8-10x cost reductions for AI training. Their CEO faced the classic enterprise dilemma: wanting AI capabilities but balking at the $2 million price tag. Their solution makes AI accessible to organizations that could never afford traditional deployments.

The market implications are staggering. We're looking at a trillion-dollar AI infrastructure market where storage architecture becomes the difference between success and expensive failure. Organizations that solve the storage bottleneck will see competitive advantages; those that don't will watch their GPU investments sit idle.

When AI Automates the Automators

Two companies caught our attention with a different approach to AI: using artificial intelligence to eliminate the busy work that consumes most knowledge workers' time. Both Hunch and Lucidity are building automation platforms, but for completely different domains.

Hunch's Overclock platform lets users describe complex workflows in plain English rather than building intricate automation rules. Their CEO David Wilson shared a telling insight: most knowledge work involves "content transformation from one system to another, with very little value add in that process." Instead of building better pipeline tools, they're questioning why we need pipelines at all.

The technical approach involves multi-agent systems with built-in oversight-what they call a "watchdog" agent that monitors execution to ensure tasks follow instructions correctly. When integrated with workplace tools like Slack and Teams, the system can handle failures gracefully and escalate to humans when needed.

Lucidity attacks infrastructure automation with their "no-ops" approach to cloud storage management. Their AutoScaler technology has been eliminating manual disk provisioning for four years, but the new Lumen platform tackles disk tiering-the complex decisions about which storage tier different workloads should use.

One customer, a major US airline, automated over 7,500 provisioning activities and saved $88,000 monthly while improving disk utilization from 21% to 82%. That's not theoretical ROI-that's measurable operational improvement from eliminating manual tasks.

The broader implication? The "busy work tax" that research suggests consumes 80% of knowledge worker time is finally addressable. Organizations implementing these automation layers will see immediate productivity gains while freeing their teams for higher-value activities.

Data Architecture Gets a Complete Makeover

And then there was the intellectually compelling presentations that came from three companies fundamentally rethinking how organizations handle data: Tabsdata, PuppyGraph, and Cohesity.

Tabsdata proposes something that sounds almost radical: eliminating data pipelines entirely. Their "Pub/Sub for Tables" approach lets domain experts-sales teams, finance departments, customer success organizations-publish specific datasets directly. Data consumers subscribe to these published tables and automatically receive updates when new versions become available.

CEO Arvind Prabhakar's insight resonates: all those complex transformations and joins that data engineers perform are essentially trying to "recreate the reality that those systems are representing." If source systems already represent business reality, why work so hard to recreate it through pipeline transformations?

PuppyGraph eliminates another infrastructure overhead: graph databases. Instead of forcing organizations to replicate data into specialized graph systems, they provide graph analytics capabilities directly on existing data warehouses. Their zero-ETL approach means complex relationship analysis without the operational complexity of maintaining another database.

The performance claims are compelling: 20-70x faster than Neo4j on comparable queries, with the ability to handle 10-hop neighbor queries across half a billion edges in under three seconds. But the operational benefit might matter more-no additional infrastructure to manage, no pipelines to maintain.

Cohesity transforms backup data from passive disaster recovery into active business intelligence. Their Gaia platform applies AI to data that's already being protected, turning backup repositories into query-able knowledge bases. One customer reduced support ticket resolution times by 40% by indexing their IT documentation in backup storage and making it searchable through natural language queries.

These approaches share a common theme: dramatically reducing the time data analysts spend on preparation rather than analysis. Organizations implementing these solutions could see faster time to insight while simplifying their infrastructure.

Industry Fights Back Against Vendor Lock-In

The UALink Consortium represents something you don't see often in technology: major competitors uniting around a common standard. With founding members including AMD, Intel, Microsoft, AWS, Apple, and Google, the consortium aims to standardize AI accelerator interconnects-providing an alternative to Nvidia's proprietary NVLink.

The technical specifications are impressive: 800Gbps per port bandwidth, support for up to 1,024 accelerators in a pod, and power consumption that's one-third to one-half of comparable Ethernet interfaces. More importantly, the standard leverages existing Ethernet infrastructure, reducing implementation costs and simplifying adoption.

What makes this particularly interesting is the participation of companies like AWS and Apple, which rarely join industry consortiums. Their involvement signals what they've pegged as a strategic importance of breaking Nvidia's current 99% market share in AI accelerators.

The market implications extend beyond just competitive alternatives. Standardized interconnects enable data centers to deploy a single switching infrastructure that works with any UALink-compatible accelerator. This separates accelerator choice from interconnect infrastructure, creating competitive pressure that could drive innovation and reduce costs across the entire stack.

With silicon expected by mid-2026, this initiative could mark the first blow to Nvidia's current market dominance.

The Great Democratization Wave

A theme that cuts across multiple companies is making advanced capabilities accessible beyond just hyperscalers and large enterprises. This isn't just about reducing costs-it's about fundamentally changing who can compete with advanced technology.

Phison's aiDAPTIV+ makes AI training affordable for universities and small businesses that could never justify traditional GPU deployments. PuppyGraph eliminates the infrastructure overhead that kept graph analytics locked away from most organizations. Hunch makes sophisticated automation accessible through natural language interfaces rather than complex programming.

Even within infrastructure companies, this democratization trend appears. Lucidity brings enterprise-grade storage management to smaller cloud deployments. Cohesity transforms backup data into business intelligence without requiring separate analytics platforms.

The historical pattern suggests this democratization could accelerate innovation in unexpected ways. When advanced capabilities become accessible to broader audiences, new use cases emerge that larger organizations might never consider. Organizations that couldn't previously afford these capabilities become new sources of competitive pressure.

What This Means for Enterprise IT

Looking across these nine companies, several important shifts emerge for enterprise technology leaders planning their 2025-2026 strategies.

Storage architecture becomes strategic rather than commodity infrastructure. The traditional approach of treating storage as an afterthought will fail in AI-heavy environments. Organizations need to evaluate whether their current storage can actually support the GPU investments they're planning.

Automation reaches knowledge work in measurable ways. Beyond factory automation, AI will eliminate routine cognitive tasks across white-collar roles. The organizations that implement these solutions first will gain productivity advantages that compound over time.

Data pipeline complexity starts becoming optional. The 80% of time data analysts spend on preparation could drop dramatically through zero-ETL approaches and direct data publishing models. Faster time to insight becomes a competitive advantage.

Open standards gain momentum as business risk mitigation. Proprietary lock-in becomes a larger concern as alternatives mature. The UALink Consortium specifically demonstrates how major technology buyers can coordinate to create competitive alternatives.

The companies featured at this 62nd edition of The IT Press Tour aren't offering incremental improvements-they're enabling fundamental changes in how enterprises operate. The question for technology leaders isn't whether these trends will impact their organizations, but whether they'll adapt early enough to gain competitive advantages.

Sometimes the most important technology shifts happen not in grand pronouncements, but in the quiet conversations at industry events where practitioners share what's actually working. This year's IT Press Tour revealed that quiet revolution is already underway.

##