At GTC, NVIDIA introduced NVIDIA STX, a modular reference architecture for AI storage designed to support the next generation of AI infrastructure, including the long-context reasoning required for agentic AI.
The announcement reinforces a critical shift already underway in large-scale AI environments: AI performance is increasingly dependent on data performance.
As models grow more capable and GPUs more powerful, the limiting factor becomes the ability to ingest, orchestrate, and serve massive data streams in real time. Agentic systems require persistent context, rapid retrieval, and continuous interaction across tools and datasets—all of which place unprecedented demands on the data layer.
But at GTC this year, the story didn’t stop with architecture diagrams or reference designs. Just outside the convention center, the future of AI infrastructure was running live—on a bus.
From Reference Architecture to Working AI Factory
While NVIDIA unveiled the underlying infrastructure innovations powering next-generation AI systems, DDN, Supermicro, and NVIDIA demonstrated what those architectures look like when deployed as a fully integrated AI factory.
The Driving AI Breakthroughs experience — the industry’s first AI Factory on Wheels — debuted at GTC as a fully operational, end-to-end AI environment running live workloads.
Inside the mobile AI factory:
- DDN’s AI data platform
- NVIDIA accelerated compute
- High-performance networking
- Pre-integrated AI software pipelines
Operating together as a single production-ready system.
Instead of showing theoretical infrastructure, visitors saw real AI pipelines executing in real time, from enterprise RAG to genomics workflows.
The result is something the industry has rarely seen before: a tangible example of what production AI infrastructure actually looks like.
Why Agentic AI Demands a New Data Architecture
The introduction of the NVIDIA STX reference architecture highlights a core challenge facing modern AI environments.
Traditional storage architectures were designed for capacity and durability — not the low-latency, high-throughput data access patterns required by AI agents operating across long contexts.
Agentic systems require infrastructure capable of:
- Rapid ingestion of massive datasets
- Persistent contextual memory
- Real-time retrieval across knowledge stores
- Continuous interaction between data pipelines and GPU compute
Without this capability, even the most powerful GPU clusters will suffer from:
- Idle compute cycles
- Slower inference response times
- Fragmented AI pipelines
That is why NVIDIA STX reference architecture moves data processing closer to the network and the GPU pipeline using technologies like:
- NVIDIA BlueField-4
- NVIDIA Vera Rubin
- NVIDIA ConnectX-9 SuperNIC
- NVIDIA Spectrum-X Ethernet
Together, these technologies help eliminate traditional bottlenecks that can slow AI systems as context sizes and data volumes expand.
But architecture alone is not enough.
The infrastructure still requires a data platform capable of operating at AI-factory scale.
The AI Data Platform: The Engine Behind the AI Factory
As an AI data platform provider, DDN works with many of the world’s largest AI environments where GPU clusters now scale to tens of thousands — and increasingly hundreds of thousands — of accelerators.
In these environments, the data layer must:
- Feed GPUs continuously with high-throughput data streams
- Orchestrate data across training, inference, and analytics pipelines
- Maintain massive namespaces for AI datasets and vector indexes
- Deliver sub-millisecond access to contextual knowledge
This is exactly the architecture visitors are seeing in action inside the AI Factory on Wheels.
Four live demonstrations aboard the bus illustrate how modern AI factories operate:
- Enterprise RAG: A live retrieval-augmented generation pipeline processing enterprise knowledge bases in real time.
- Financial Services Risk Intelligence: Continuous financial risk simulation replacing overnight batch analytics with near real-time intelligence. · Genomics & Drug Discovery: Accelerated biomedical AI workflows integrated with NVIDIA BioNeMo.
- Intelligent Video Analytics: Low-latency video ingestion and AI analysis pipelines running simultaneously at the edge and in the data center.
Each pipeline demonstrates the same principle:
AI value emerges when compute, networking, and data operate as a unified system.
From Concept to Production AI
The response to the AI Factory bus at GTC reveals something important about where the industry stands today.
Enterprises understand the potential of AI — but many still struggle to deploy infrastructure capable of delivering real business outcomes.
According to DDN’s 2026 State of AI Infrastructure Report:
- 65% of organizations cite infrastructure complexity as a major barrier to AI ROI
- More than half have delayed or cancelled AI projects because of it
That gap between AI ambition and production readiness is exactly what the AI Factory model is designed to solve.
Instead of assembling fragmented infrastructure components, organizations can deploy validated AI factories engineered for real workloads from day one.
The Future of AI Infrastructure
NVIDIA STX signals an important direction for the industry: AI infrastructure must become data-accelerated from the ground up.
The AI Factory on Wheels shows what that vision looks like when fully realized.
A tightly integrated environment where:
- GPUs remain fully utilized
- Data flows continuously across pipelines
- AI workloads move seamlessly from training to inference
In other words, a system where data performance drives AI performance.
As AI adoption accelerates across industries—from finance to life sciences to robotics— organizations will increasingly require infrastructure built around this principle.
The next generation of AI innovation will not be defined by GPUs alone.
It will be defined by AI factories, powered by integrated compute, networking, and AI data platforms working together at scale.
And sometimes, as visitors to GTC discovered this year, those AI factories might just arrive on wheels.