Top AI use cases for NVIDIA vGPU and Horizon

Every enterprise is racing to build or use AI infrastructure. Most default to Containers and graphics processing unit (GPU) clusters. For application programming interface (API)-driven pipelines and stateless compute, that's the right call. But it leaves a set of high-value workloads unaddressed: workloads that have a desktop dependency, that operate against thick-client applications with no API, or that depend on GPU-rendered visual data. Containers were never built to handle these; no amount of container orchestration solves a problem that starts with needing a rendered screen.

Omnissa Horizon + NVIDIA vGPU is purpose-built specifically for these workloads. Whether or not you're running a virtual desktop infrastucture (VDI) today, it's worth knowing where this architecture applies. For these workload types, it isn't a workaround—it's the only architecture that works.

Ask yourself:

• Does the workload require a persistent desktop?
• Does it interact with a thick-client app that has no API?
• Does it depend on GPU-rendered input data?

If the answer is yes to any of these questions, keep reading! The four use cases below show how Horizon + NVIDIA vGPU applies today, each with an example of how an organization in that space could put it to work.

1. AI developer workspaces

AI engineers need flexible, high-performance environments for experimentation and model training. Today, that usually means static GPU allocations or dedicated bare-metal machines, provisioned through a multi-week procurement cycle, then left idle between experiments.

Horizon + NVIDIA vGPU delivers on-demand, GPU-accelerated workspaces with elastic compute tiers and full governance, so developers can spin up an environment in minutes, scale to a higher GPU tier mid-run, and release it the moment they're done.

Example: Pharmaceutical & life sciences

Consider a global pharmaceutical company: Research  and development  engineers could use Horizon AI workspaces backed by NVIDIA vGPU to fine-tune protein-structure prediction models against proprietary compound libraries, in DLP-protected sessions, and access higher GPU compute profiles during intensive runs and release them when finished. In a scenario like this, provisioning time could drop from weeks to under an hour.

2. Legacy GUI Application Automation

Critical workflows live inside thick-client apps—such as enterprise resource planning (ERP) systems, core banking platforms, and supervisory control and data acquisition (SCADA) human machine interfaces (HMI)—that expose no API. Containers can't automate what they can't run. Horizon + NVIDIA vGPU offers a governed way to deploy AI agents against these systems at scale, with vGPU ensuring pixel-perfect rendering so vision models interact reliably with UI states.

Example: Energy & utilities

Consider a grid operator: AI agents deployed within Horizon sessions could monitor legacy SCADA dashboards, log anomalies, and trigger maintenance work orders around the clock while inheriting the same governance, session recording, and access controls already applied to human operators, with no additional tooling required.

3. Human-in-the-loop hybrid workflows

High-stakes workflows—such as contract review, anti-money laundering (AML) investigation, and security operations center (SOC) triage—benefit from AI handling data aggregation while humans own the decision. The challenge is a seamless handoff. In a Horizon session, a supervisor can join a live-agent session and see exactly what the agent sees—no reconstruction, no context loss. NVIDIA vGPU keeps the session fluid and responsive throughout.

Example: Legal & professional services

Consider a law firm: Senior partners could join live AI contract-review sessions to see the agent's annotated document state in real time. Thus, potentially compressing first-pass merger and acquisition (M&A) review timelines while maintaining full partner oversight, with every action captured for compliance.

4. GPU-rendered visual AI workflows

Some AI pipelines operate on data that only exist as rendered pixels. For example, medical imaging, building information modeling (BIM) models, VFX frames, and simulation output. Without GPU rendering, the source data doesn't exist. NVIDIA vGPU pulls double duty: powering the rendering workload and running vision model inference in the same session, as well as making large-scale deployments economically viable without dedicated physical workstations. 

Example: Architecture, engineering & construction (AEC)

Consider an AEC firm: AI vision models could operate against GPU-rendered BIM views inside Horizon sessions to flag clashes and code-compliance issues across hundreds of models, without dedicated workstations per review, while potentially compressing pre-construction quality assurance (QA) from days to hours. 

Ready to get started?

The organizations that extract the most AI value over the next 18 months won't necessarily have the largest GPU clusters. They'll be the ones that extend AI into the governed, GUI-bound, and human-supervised workflows at their operational core, especially the workflows containers can't reach.

If you've already deployed Horizon for your workforce, your agent infrastructure is closer than you think. If you haven't, now may be a good time to consider it for use cases like the ones above.

Visit the Omnissa + NVIDIA page to see how we are working with NVIDIA to simplify your AI adoption.