Deploying AI factories is not for the faint of heart. They’re complex feats of engineering, involving careful consideration and planning at every level — from the building foundation up through the network cables that make GPU clusters possible. At AI-factory scale, even slight infrastructure misconfigurations can have outsized impacts on overall performance, so preparation is critical.
This is why VAST became an early partner on the NVIDIA DSX Air platform. By creating simulations of data center systems, DSX Air helps users test, optimize, and validate their network configurations before racking their gear or rolling out updates. It has been instrumental in helping our customers generate network topologies and configurations, ensuring the successful deployment of our AI OS into their AI environments and significantly reducing AI factory deployment time.
Recently, we partnered with NVIDIA to incorporate our AI OS and DSX Air into the NVIDIA DSX platform for ground-up AI-factory planning. DSX Air allows for full-stack simulations of AI workloads, including security, storage, orchestration and other software-level functionality, in addition to compute and network configurations. Customers ensure their workloads will run as expected, and also test how certain failures (ranging from a bad cable to a cloud outage) impact performance, or ensure other layers of the stack aren’t starving GPUs of data.
This has been a boon for VAST customers, who can now test their specific workloads using DSX Air to help avoid surprises down the road. For CSP customers, that might involve validating their cloud control plane atop a collection of GPU clusters backed by VAST AI OS storage and data services. Customers deploying their own AI applications can use DSX Air to simulate entire end-to-end AI pipelines, such as KVcache reuse considerations and real-time video reasoning workflows invoking NVIDIA AI models and taking full advantage of AI OS capabilities (e.g., data storage, event triggers, serverless functions, and vector database/RAG functionality).
This level of planning will be critical as more organizations begin deploying AI workloads, whether in their own data centers or in custom environments with AI cloud providers. Infrastructure is expensive, enterprise applications can involve many moving parts and legacy systems, and workloads like AI agents can generate unparalleled amounts of network traffic. So we’re proud to collaborate with NVIDIA on DSX Air and ensure our customers can deploy AI with the confidence that their infrastructure is sound and their workloads will run as expected. If you’d like to learn more about using NVIDIA DSX Sim and NVIDIA Air to help simulate and plan your AI factory, reach out and we’d be happy to help.
