San Jose, CA, March 16–19, 2026
NVIDIA GTC 2026, the world’s premier conference for AI, high-performance computing, and the technologies shaping our future, descended on San Jose, California last week. VDURA showed up ready to make some noise. From groundbreaking product announcements to conversations that stopped attendees in their tracks, here’s how our week unfolded at Booth #3422.
The AI Storage Problem No One Was Solving Until Now
Every year, AI clusters get faster. GPUs get more powerful. Model sizes grow. But for years, one silent bottleneck has been quietly strangling AI performance: the data path between storage and the GPU. Every time a GPU needed data, it had to wait. It had to route through the CPU. It had to fight for bandwidth. In the era of trillion-parameter models and real-time inference, that latency isn’t just inconvenient it’s costing organizations millions in wasted GPU cycles. At GTC 2026, VDURA declared that era over.
Introducing RDMA: GPU-to-Storage at the Speed of Thought
On March 16th, VDURA unveiled Remote Direct Memory Access (RDMA) support for the VDURA Data Platform now generally available across all V5000 and V7000-class systems. What does that mean in practice? GPU servers can now access storage directly, bypassing the CPU entirely. No overhead. No bottlenecks. No waiting. This is the low-latency, high-throughput data path that AI training and inference workloads have always demanded and VDURA is now delivering it at scale. As Ken Claffey, CEO of VDURA, put it:
“Today’s announcements at GTC 2026 reflect our commitment to delivering the AI storage platform that spans the full data hierarchy from memory to long-term retention with no compromises on performance.”
Paired with AMD EPYC Turin processors and NVIDIA ConnectX-7 high-speed networking adapters, VDURA’s optimized infrastructure configurations represent a new benchmark for GPU-native AI infrastructure.
Context-Aware Tiering: Your Data, Exactly Where It Needs to Be
VDURA didn’t stop at RDMA. The team also unveiled the first phase of Context-Aware Tiering a technology that intelligently places data across the storage hierarchy based on what the workload actually needs, not a static policy.
Phase 1 includes:
- DirectFlow buffer extension into local NVMe SSDs for hot data placement
- KVCache writeback controls for selective persistence of critical inference data
- A Unified Context Cache Tiering framework spanning DRAM and local SSD for LLM inference support
Future phases will extend this further adding application-aware data placement, improved cache coherence, and NVIDIA BlueField-4 DPU support.
This is VDURA’s vision made real: one platform that unifies memory, flash, and disk under HYDRA the only high-performance distributed architecture purpose-built for modern AI.
The Mountain Comes to GTC
If the tech announcements weren’t enough to make heads turn, VDURA brought something or rather, someone that no one at GTC expected. Hafþór Július Björnsson. “Thor.” The Mountain.
The world’s strongest man, VDURA’s brand ambassador, and the man who deadlifted 510kg in a VDURA hat, was on the show floor and the reaction was exactly what you’d expect. Analysts, engineers, and AI researchers who spend their days debating GPU architectures suddenly found themselves lining up for a photo with the most recognizable strongman on the planet.
But there was a deeper message here. VDURA has always believed that raw power, properly channeled, changes everything whether you’re lifting half a ton of iron or moving petabytes of training data at nanosecond speeds. Thor’s presence wasn’t just a spectacle.
It was a statement: Strength without bottlenecks. Power without limits. Data without compromise.
WATCH THE VIDEO: Thor and Jensen Huang
The GPU Savings Challenge
We also brought a little friendly competition to our booth. Visitors were invited to take on the VDURA GPU Savings Challenge: Where event attendees could guess how many extra GPUs they could buy with the savings from VDURA’s Mixed Fleet storage approach.
Some guessed conservatively. Some swung for the fences. Almost nobody guessed right, and the reveal never got old. It was a simple question with a powerful answer, and it sparked exactly the kind of conversations we came to GTC to have.
The Conversations That Mattered
Beyond the booth, GTC 2026 was a week of real conversations with customers, partners, analysts, and innovators pushing the boundaries of what AI infrastructure can do.
Industry watchers took notice. Marc Einstein, Research Director at Counterpoint Research, called GTC 2026 “an amazing show” and singled out VDURA as one of the companies that caught his attention, noting they have “a very interesting AI and HPC storage and management platform.”
That’s exactly what VDURA came to GTC to demonstrate: not just a product, but a platform one that’s purpose-built for the AI era and ready for whatever comes next.
What’s Next
GTC 2026 was a milestone, not a finish line.
Context-Aware Tiering Phase 1 is on track for general availability later this year. RDMA is live and ready to deploy. And the VDURA team is already heads-down on what comes next BlueField-4 DPU support, application-aware data placement, and a continued push to eliminate every remaining bottleneck between your GPUs and your data.
If you stopped by Booth #3422 last week thank you. If you didn’t make it to the show, reach out to the VDURA team and let’s talk about what GPU-native AI storage can do for your organization. The future of AI runs on fast storage. And fast storage runs on VDURA.