The Fastest Path to the First AI Token: Digital Twins with NVIDIA DSX Air and Keysight Inference Builder
AI inference has quickly become one of the fastest-growing workloads in modern data centers. The surge in demand for GPUs, high-performance networking, memory and storage systems, etc., has created both rapid infrastructure expansion and significant design complexity. Delivering AI inference service is no longer about a single system component — it requires many layers to operate together seamlessly. The inference stack — comprising models and inference engines, routing and switching fabrics, DPU offloads, security guardrails, compute resources, memory hierarchies, and storage platforms — must function as a coordinated whole. Each design decision affects latency, throughput, and ultimately how quickly a system can deliver its first response to a user request.
Compounding this complexity is the reality that hardware procurement cycles and supply constraints often prevent organizations from assembling their entire infrastructure at once. Teams frequently must make architectural decisions and operational preparations long before every component is physically available. To bridge this gap, organizations are increasingly turning to digital twin and simulation environments that allow them to explore architectural choices, validate policies, and prepare operational workflows ahead of deployment. By enabling this early validation, these environments help teams move quickly and confidently toward achieving time to first AI token (a fully operational and optimized Inference stack) once the physical infrastructure is in place.
What NVIDIA DSX Air Enables
NVIDIA DSX Air provides a digital environment where organizations can construct and interact with a simulated representation of their AI infrastructure. Within this environment, users can assemble critical components of the AI stack, including routing and switching systems, SuperNIC and DPU offloads, compute nodes, memory resources, storage services, and security frameworks, observing how they interact as a unified architecture.
This environment allows engineers to design network topologies, configure infrastructure and security policies, and test system interactions before physical hardware is installed. By providing access to realistic representations of these systems in a shared sandbox, DSX Air enables teams to evaluate deployment strategies, validate operational rules, and ensure interoperability across multiple infrastructure layers.
Beyond design validation, the platform allows organizations to experiment with operational workflows such as provisioning, automation, and policy management. Teams can model production-like environments, refine their configuration approaches, and develop operational familiarity with complex AI infrastructure before the systems are deployed in a data center.
Keysight AI (KAI) Inference Builder: Realistic Workload Emulation
Keysight AI (KAI) Inference Builder complements this simulated infrastructure by introducing realistic workload emulation into the environment. Keysight can emulate AI clients that generate diverse prompt patterns reflecting real-world usage across different market segments. These workloads may represent chatbot interactions, financial analytics queries, legal document processing, enterprise copilots, and other application-driven inference scenarios. The platform can also simulate abnormal or adversarial traffic patterns to evaluate how infrastructure and guardrail systems respond under unusual conditions.
In a two-arm emulation configuration, Keysight models both the request side and the response behavior of inference systems. This creates a closed-loop environment that exercises inline infrastructure components such as switches, routers, load balancers, SuperNICs, and security enforcement platforms. By driving the infrastructure with realistic traffic dynamics, organizations can observe how policies, routing rules, and system protections behave under production-like conditions.
Keysight also supports a one-arm configuration, where generated prompt workloads interact directly with real inference systems. This approach allows customers to validate live environments for key performance characteristics, including token throughput, response latency, concurrency handling, and scaling limits. Together, these approaches enable teams to test both simulated and physical environments under realistic workload conditions.
Conclusion
By combining the digital-twin capabilities of NVIDIA DSX Air with workload emulation from Keysight, organizations gain a practical environment for preparing AI infrastructure before it reaches production. Teams can design, validate, and refine their architecture in advance, ensuring that networking, compute, storage, and security systems behave as expected when real workloads arrive.
This preparation significantly accelerates first-time deployments. When the physical components are installed, infrastructure teams can move quickly from installation to operation with confidence that policies, automation workflows, and system interactions have already been validated.
The same environment also supports ongoing operations. Because the simulated infrastructure closely mirrors production behavior, organizations can safely evaluate configuration updates and automation changes within their CI/CD workflows before applying them to live systems. This ability to test and refine changes in advance reduces operational risk while helping teams maintain efficient and reliable AI infrastructure as workloads scale.