By Jeff Koyen

getty

Enterprise IT leaders spent the last two years trying to figure out how to bring AI into the business. As organizations move past experimentation, they must navigate the operational strain AI deployments can create at scale.

The puzzles don’t end with model selection; they start there. How do you support the network and infrastructure demands AI suddenly creates? Where should particular workloads run? How do you manage 50,000 employees with different devices and varying needs?

Reducing this operational complexity is the job for IT teams. The question of how to best achieve that goal defines the work of technology leaders like Hasmukh Ranjan, senior vice president and CIO at AMD.

“You have to be very thoughtful that if you are going to spend a lot of money on AI tokens, you are not bothered or worried if this PC is configured correctly,” Hasmukh says. “Those operational layers need to be more abstracted so you’re focused on the bigger-value solutions for your enterprise.”

Below, we explore how AI is complicating this goal and share how an enterprise foundation with cutting-edge performance, security and manageability features can help organizations thrive in the AI era.

Cost And Performance Are Converging Yet Conflicting

AI consumes resources, such as compute, memory, network bandwidth and electricity, at an astonishing scale that older enterprise infrastructure wasn’t built to handle. The bill is coming due. “If you are an enterprise and you have adopted AI, the data has exploded,” Hasmukh says.

Network traffic and bandwidth consumption have surged in step. The typical AI query pulls data from multiple sources, runs it through a model and returns results. As enterprise systems process more data at higher speeds, the underlying storage, data and network layers all need an upgrade.

Faced with rising AI demands, an IT manager’s first instinct may be to solve the problem with additional cloud infrastructure. In Hasmukh’s view, that approach is not only limiting but unnecessary. Instead, he counsels enterprises to optimize their AI for the most cost-effective infrastructure that can actually handle it.

Inside AMD data centers, that means carefully considering where and how compute resources are deployed. Not every AI workload requires water-cooled GPUs or radical server room redesigns, for example. Many run well on air-cooled cards with lower power consumption and less need for infrastructure overhauls.

“We are being thoughtful in making sure that we prepare for the AI enterprise in different segments,” Hasmukh says. “That’s how we are investing.”

Those same pressures are increasingly extending to the endpoint, where employee devices are beginning to handle more AI workloads locally. Enterprises now need PCs that can balance local AI performance with efficiency and low overhead.

At the same time, tech teams need the manageability, deployment scalability and lifecycle consistency to support large device fleets. AMD PRO, an enterprise platform for commercial PCs, is designed to address these complex operational realities, helping businesses balance AI-ready performance with the manageability and long-term support large fleets require.

Security, AI And Data Governance

Even before the cloud era, IT leaders were thinking about data security as a perimeter problem. Keep the wrong people out. Encrypt what moves between systems. Audit what gets logged.

Now, with most AI interactions inside an enterprise leaving the building, maintaining perimeter security becomes even more crucial.

“Today, pretty much 99% of traffic goes to a frontier model,” Hasmukh says. “That is not a solution that will stay at scale and remain efficient.”

As AI agents become more autonomous, that efficiency challenge grows quickly. Unlike traditional chat interactions, agents continuously reason, retrieve information, call tools and execute workflows, driving significantly higher token consumption across the enterprise. AMD recently explored how this shift is pushing organizations toward more hybrid AI architectures, where some workloads run locally or on dedicated infrastructure to improve cost efficiency, latency and control.

For CIOs, these new transactions can create governance risks as well. Sensitive customer data, proprietary code, financial information and internal documents shouldn’t pass through AI workflows outside the IT team’s direct control.

That’s why AMD is rethinking where AI workloads should run. Some belong in the cloud, some in the data center and some on the device itself. Simple, routine queries may stay local, while frontier models are reserved for workloads that truly require them.

This shift further underscores the importance of security and governance at the endpoints. As AI workloads become more distributed, enterprise endpoints must balance local AI performance with robust security and policy compliance at scale. AMD PRO addresses that need by delivering enterprise-level security with built-in, multilayered protections that are always in place.

The Rising Complexity Of IT As A Service Provider

If some AI workloads should run locally for cost, and some should run locally for security, employee devices must be capable of handling them. And IT needs to manage these devices at scale.

“It’s not enough to put a more powerful processor along with a specific graphics processor on a device,” Hasmukh says. “When I think of a device, I think about 50,000 devices I have to manage, and they have to be productive.”

As more AI workloads move onto employee devices, the management burden for IT teams grows with them. This is the environment AMD PRO processors are designed for: helping enterprises balance local AI performance with security, manageability and simplified deployment across large device fleets.

For Hasmukh, the implication is crystal-clear: if AI is going to run at the edge, the edge needs both top-notch performance and centralized management.

Discover AMD PRO systems for today’s enterprise IT challenges