Healthcare organizations are balancing tight budgets and operational transformation.
While finances remain in a perilous state (market research finds year-to-date operating margin increases are hovering below 0.5%, with nonlabor expenses driving cost inflation), healthcare is ramping up its use of artificial intelligence, with more than two-thirds of organizations actively using AI and close to half assessing or using agentic AI.
Standing at the intersection of these trends is the evolving question of where and how to best deploy AI.
Often, healthcare organizations focus on the balance between running AI workloads on the right infrastructure and managing costs, says Tony Nunes, senior manager for strategic business development in healthcare and life sciences at AMD. While this strategy is necessary, especially with cloud management costs remaining higher than expected, it can neglect the importance of aligning and redesigning workflows to accommodate AI.
“Whether you build something in house or buy something off the shelf, you’re investing in something that fundamentally changes how people, process and technology work together,” Nunes says. “If you’re not prepared, and you just try to implement something without thinking about workflows, you’re going to meet resistance.”
DISCOVER: AMD delivers the benefits of AI, cloud and HPC solutions from edge to cloud.
Three Main Motivations for AI Adoption in Healthcare
To understand why organizations may jump into AI without first considering workflow realignment, it helps to know what’s been motivating adoption. Nunes notes three factors:
- Automation. It’s hard to ignore the potential to significantly drive down healthcare costs by automating operational workloads, patient interactions and time-consuming clinical processes.
- Edge Computing. Shifting AI capabilities from the data center to the cloud to client devices (within individual clinics or even patients’ homes) reduces latency and allows for real-time data analysis.
- The Fear of Missing Out. “Corporate buyers can be just as emotional as consumers,” Nunes adds. “When there’s a chorus of voices talking about what generative AI can do, organizations are afraid of getting left behind because they’re not taking advantage of it.”
When these factors drive decision-making, retrofitting modern AI tools to existing workflows seems like the path of least resistance. That’s where things break down, Nunes says, highlighting MIT research about organizations struggling to see ROI for generative AI investments.
“You run into the people and process side of the equation. Healthcare workflows have been structured in a certain way, and they have certain staffing requirements. If AI doesn’t match well with that, you end up creating new workflows that depend on a different mix of people that no one had mapped out.”
Click the banner below to power employee productivity with AI.
The Challenge of Legacy Infrastructure
The ongoing hurdles for healthcare to move past proof of concept with AI also stems from the persistence of legacy infrastructure, Nunes says. This makes it difficult to optimize performance, which means cost reductions are hard to realize.
With the evolution of the electronic health record, for example, AI orchestration layers have emerged to “break some of the barriers” that organizations face in their efforts to gain insights into data assets.
“When you introduce something new to your infrastructure, you need to make sure everything’s in alignment,” Nunes says. “You need to have the right processors in place to get the right level of performance while minimizing your overall costs. If you don’t have your underlying infrastructure aligned, you’re going to incur more costs.”
Faster refresh cycles for today’s software systems — often to accommodate AI-driven functionality — further complicate matters. If these capabilities are pushed onto new platforms before organizations are ready for them, leadership faces three choices, Nunes says: Refresh infrastructure from the ground up, bolt on another platform, or wait and see.
Given the projected costs of the first two options, most organizations today opt for the third. “A lot of hospitals aren’t planning to adopt until they get a grasp on what it will cost. But they won’t be able to maintain this approach because they need to drive down long-term costs, and leaders have the fear of missing out,” Nunes says.
The Benefits of a Flexible, Open Core
In this regard, organizations benefit from looking at processors as “the first component of modernization,” serving a function similar to a strong foundation for a building, Nunes says. Ideally, this infrastructure will also be flexible (to support seamless expansion as necessary) and secure (built in by default, not bolted on).
“If you can get the same performance with fewer core servers, then you get some real estate back in the data center,” Nunes says. Operating fewer servers also helps organizations address possible rising energy costs not accounted for in long-term budgets.
There’s also a compelling argument for infrastructure built on open standards as opposed to proprietary systems, he adds. AMD has prioritized infrastructure built on open standards and made it possible to migrate at least 90% of workloads to open stacks.
Here, Nunes notes, the pattern is similar to the transition from monolithic mainframes to client-server systems more than 30 years ago: Open standards quickly closed the performance gap while offering organizations access to a broader partner ecosystem. “We’re betting that open standards will provide more opportunity for development and more choice on the part of consumers — and both will lead to reduction of costs.”
Brought to you by:
