The use of artificial intelligence in healthcare is expanding, with applications ranging from improving diagnostics for rare diseases to easing clinical documentation burdens. However, the focus has often been on the type of AI models used or the number of graphics processing units, rather than whether organizations are ready to support and secure AI workloads.
Jeremy Support, senior vice president and general manager of compute at Cisco, states that the key issue is ensuring that the right workloads are relying on AI. Support notes that “local computing, networking, security — all those elements are part of a hospital’s infrastructure that is leveraging AI as a normal course of business, not as a sidecar.”
Support discusses the need for healthcare leaders to think about workloads, rather than just infrastructure, when it comes to AI. He believes that most healthcare leaders have assumptions about AI infrastructure that are focused on the wrong aspects, such as where AI technologies are taking place, rather than how they can be integrated into the clinical workflow.
Healthcare Leaders’ Assumptions
Support says that he wants to have conversations with healthcare leaders about workloads, not just chatbots or infrastructure. He notes that the challenging part of healthcare is that all aspects of AI must work together, including performance, cost, security, and compliance.
Support believes that the conversation around AI infrastructure is often looked at as a procurement decision, but it should be viewed as a placement decision. He considers factors such as what the clinical workflow is, where the data is created and lives, and what the costs are of moving data between different areas.
In the future, Support thinks that people will not be challenging healthcare organizations about which AI model they chose for a certain job, but rather about how they built their foundation and architecture to support AI workloads.
Transformation for Providers
Many healthcare organizations work with legacy systems and may not be prepared for AI workloads. Support notes that the transformation does not require a complete overhaul of the system, but rather a refresh cycle that can be a catalyst for starting AI projects.
Support believes that no healthcare organization is starting from zero, and that everyone has infrastructure that can be leveraged to improve operations and clinical workflow by using AI. He notes that when organizations replace outdated systems, they can use the opportunity to improve their operations and clinical workflow.
Support notes that there is a tension between adhering to strict compliance and security measures and deploying flexible, scalable compute for modern AI workloads. He believes that questions about workload placement are imperative, and that organizations need to consider the cost and operational expense of moving data around.
Support gives the example of a cardiac MRI, which is imaging done locally that then gets processed in the cloud and returns a bit of data. He notes that this type of workload works at the edge, and that organizations need to consider this when creating a solution for a workflow end to end.
Support wants providers to start thinking about their infrastructure needs now, so they don’t have to stall their strategic goals.
Support thinks that the top healthcare IT priorities for 2027 will be around AI doing more inferencing to support operations and leveraging agents. He notes that this will generate a tremendous amount of traffic on the network, and that organizations will need to manage their WAN and connected pieces effectively.
When it comes to compute, power, and space, Support believes that everyone can plan for and have access to GPUs, but how they pull together the architecture is what will ultimately make them successful. The winners in healthcare AI will be the ones who build the right foundation and architecture, rather than just buying the most powerful GPUs or leveraging the smartest models.
Over the past 10 to 15 years, there has been a push towards edge infrastructure, but AI has now become a driving force in making it happen. Support notes that this is where the data is being generated, and that companies are trying to replatform their operations. He believes that Cisco is poised to take healthcare providers to the next level, especially for partners who have already been designing at the edge.
Agentic AI queries generate up to 25 times more network traffic than a chatbot, and this changes the pattern of what that traffic looks like, with AI agents working 24/7, requiring a much higher level of bandwidth that will be consistently used up.
