AI Is Changing the Economics of the Digital Workspace: Five Shifts IT Leaders Need to Make

Artificial intelligence is changing enterprise computing, but its most immediate impact may have less to do with what happens on employees’ desktops and more to do with the infrastructure behind them. 

As organizations invest in AI, IT leaders face a growing dilemma. They need to support increasingly compute-intensive workloads, provide access to expensive GPU resources and experiment with new technologies, all while controlling infrastructure costs. Traditional approaches to end-user computing, built around standardized desktops and common infrastructure platforms, are becoming harder to justify when employees have dramatically different computing requirements. 

The answer isn’t necessarily to equip everyone with an AI PC or move every workload into the cloud. Instead, organizations need to rethink where computing happens, how resources are allocated and how employees access them. 

Here are five shifts that can help IT leaders build a more flexible and economically sustainable digital workspace. 

1. Stop Treating Every Employee as Having the Same Computing Requirements

For years, standardization has been a central principle of enterprise IT. Organizations have sought to simplify operations by giving employees similar devices, desktop environments and computing resources. 

But AI is widening the gap between what different employees need. An office worker using productivity applications and cloud-based AI assistants may require little additional computing power. A data scientist experimenting with AI models, an engineer running simulations or a creative professional rendering complex graphics may need access to substantial GPU capacity. 

Providing everyone with high-performance computing resources is expensive and wasteful. Conversely, forcing demanding workloads onto infrastructure designed for everyday productivity can undermine performance. 

IT should begin by evaluating computing requirements by workload rather than by department or device category. The objective is to provide sufficient computing power for each task without making the most expensive configuration the organizational default. 

2. Rethink the Economics of Cloud Versus On-Premises Computing

Cloud infrastructure offers an attractive way to access specialized computing resources without purchasing expensive hardware. For short-term AI projects or unpredictable workloads, GPU-enabled cloud instances can provide valuable flexibility. 

However, the economics change when those resources are used continuously. A cloud instance that makes financial sense for a temporary project may become costly when running around the clock. Meanwhile, purchasing dedicated infrastructure for workloads used only occasionally can leave expensive computing capacity idle. 

Other factors complicate the equation. Data location, network performance, storage costs, software licensing and security requirements can all influence where computing should happen. Rather than committing to a cloud-first or on-premises-first strategy, organizations should evaluate where each workload delivers the best combination of performance, cost and operational efficiency. 

Those decisions should also remain open to change. As AI requirements evolve and infrastructure pricing shifts, the most economical location for a workload may change with them. 

3. Make Expensive Computing Resources Available on Demand

One of the biggest opportunities to improve computing economics is to stop treating high-performance resources as equipment that must be permanently assigned to individual employees. 

Consider a team of engineers who need GPU-intensive workstations for certain stages of a project. Rather than purchasing a high-end system for every engineer, an organization could provide access to a shared pool of physical or cloud-based workstations. The same approach can support AI developers who need substantial computing power during model testing but much less capacity for routine development activities. 

Making this model work requires more than simply creating shared infrastructure. IT needs the ability to assign resources based on user requirements, manage workstation pools, control capacity and automate when resources are powered on or off. These capabilities can help improve utilization and reduce unnecessary spending, particularly when organizations are paying for costly GPU resources. 

The goal is to make high-performance computing available when employees need it, without paying for maximum capacity all the time. 

4. Separate How Employees Access Computing from Where It Runs

A flexible computing strategy becomes difficult when employee access is tightly connected to a particular infrastructure platform. In traditional virtual desktop infrastructure (VDI), connection brokers, hypervisors, display protocols and hosting environments may be closely integrated. Moving a workload to another platform can therefore require significant changes to the surrounding architecture and potentially disrupt how employees access their desktops. 

A more adaptable approach separates the access and management layer from the underlying computing resources. Employees can authenticate through a consistent interface and connect to authorized resources, whether those resources are conventional virtual desktops, physical workstations, Linux systems or GPU-enabled cloud instances. 

Behind that experience, IT can select appropriate connection protocols, manage resource assignments, orchestrate provisioning and automate capacity and power management across different environments. This separation makes it possible to move workloads as requirements change without repeatedly redesigning the employee access experience. It also gives organizations greater freedom to use existing infrastructure alongside new computing platforms rather than replacing everything whenever a different technology becomes necessary. 

5. Design for AI Requirements That Haven’t Emerged Yet

Perhaps the greatest challenge with AI infrastructure planning is that organizations are making investment decisions while the technology and its business applications are still evolving. Some AI workloads will run locally on endpoints. Others will depend on cloud services, private infrastructure or specialized GPU systems. The balance between these approaches will change as models become more efficient, applications mature and new computing technologies emerge. 

Building an entire digital workspace strategy around today’s assumptions creates unnecessary risk. Instead, organizations should be able to introduce new computing environments incrementally, evaluate their economics and expand or change deployments as business requirements become clearer. 

A department might initially use cloud GPUs for an AI pilot, then transition to dedicated infrastructure if sustained demand justifies the investment. Another team may discover that its existing physical workstations remain the most economical option. 

The important architectural consideration is ensuring that these decisions can be made independently, without forcing a major infrastructure migration or changing how employees access their computing resources. 

The Real AI Opportunity Is Greater Computing Flexibility 

Much of the discussion about AI in the workplace focuses on how intelligent applications will change the employee experience. But organizations also need to address how AI will change the economics of delivering that experience. 

Not every employee needs more powerful hardware. Not every AI workload belongs in the cloud. And not every new computing requirement should trigger another infrastructure investment. The organizations best positioned for this transition will be those that can match computing resources to workload requirements, manage expensive capacity efficiently and adapt infrastructure decisions as technologies and costs evolve. 

About the Author 

Karen Gondoly is CEO of Leostream, a leading provider of enterprise remote access and desktop connection management solutions. Since joining Leostream in 2008, she has held numerous technical and leadership positions, including product manager and COO, before becoming CEO in 2016. Before joining Leostream, she worked at The MathWorks, where her responsibilities included software development, user interface design and usability. She holds bachelor’s and master’s degrees in aeronautical and astronautical engineering from the Massachusetts Institute of Technology (MIT). Her expertise spans software development, user experience, virtualization, cloud computing and enterprise remote access.