Healthcare organizations are moving artificial intelligence closer to the bedside, the operating room and the factory floor, and that shift is changing how medical devices and robotics systems are built and managed. In a recent interview with HealthTech, David Niewolny, senior director and global head of business development for healthcare and medical at NVIDIA, said the industry now needs systems that can analyze data at the edge for real-time responses while still connecting back to the cloud for training, coordination and oversight.
Why edge AI is gaining ground in healthcare
Hospitals, medtech companies and research groups are under pressure to make faster decisions from more data, including imaging, patient monitoring streams and robotic sensing. Cloud-based AI can help train models and consolidate data, but many clinical use cases require low latency, local processing and reliable performance even when connectivity is limited.
That is why edge-to-cloud architecture is becoming more common. In practical terms, the edge handles inference where data is created, while the cloud supports larger-scale model updates, fleet management and analytics across sites. For healthcare, the appeal is speed, privacy and resilience.
The trend also aligns with the growing use of connected devices. From infusion pumps to surgical robots and remote monitoring tools, many systems now generate continuous data that can inform care in seconds rather than hours. The more automated those tools become, the more important it is to process information close to the source.
What healthcare leaders need to manage
Niewolny’s central message is that deploying AI is no longer just a model-building exercise. Healthcare organizations have to manage the full lifecycle of the device, including provisioning, software updates, security patches, telemetry, monitoring and retirement.
That means IT teams, clinical engineering staff, cybersecurity leaders and data scientists need to coordinate more closely than they did with traditional equipment. If a hospital rolls out AI-enabled devices across multiple departments, it must know which version is running where, how the system behaves under load and whether model performance changes over time.
Model drift is a particular concern. A system trained on one patient population, imaging protocol or clinical workflow may not perform the same way in a different environment, which is why monitoring and validation remain essential after deployment. The Food and Drug Administration has repeatedly stressed that connected medical technologies require ongoing cybersecurity and software management, not one-time installation.
Security, compliance and interoperability remain the bottlenecks
Healthcare data carries some of the strictest privacy and compliance requirements in any industry, and AI does not reduce that burden. In fact, distributed systems can increase the attack surface because more devices, endpoints and update channels must be protected.
That is why vendors are emphasizing secure boot, identity controls, encryption, network segmentation and audit trails. The FDA has also issued guidance urging manufacturers to build cybersecurity into medical devices from the start, especially as products become more networked and software-dependent.
Interoperability is another challenge. AI platforms must fit into existing hospital systems, including electronic health records, imaging archives and clinical workflows. If a new tool slows clinicians down or creates duplicate work, adoption is likely to stall regardless of the technology’s performance.
Industry analysts have long pointed out that healthcare adoption depends as much on workflow design as on raw computing power. That is especially true for edge AI, where a device may need to respond instantly while still sending data back to centralized systems for review and compliance.
Where the technology is showing up first
The clearest use cases are in environments that demand speed and precision. Imaging, pathology, patient monitoring, operating rooms and robotics are among the areas where edge processing can reduce delay and help teams act on data as it arrives.
Robotics is a particularly important test case because autonomous or semi-autonomous systems often need to interpret sensor inputs in real time. In those settings, even small delays can affect accuracy, safety or throughput.
Remote monitoring is another area to watch. As more patients are managed outside the hospital, edge-enabled devices can analyze signals locally and escalate only the most important data to clinicians. That can reduce bandwidth demands while improving responsiveness.
Cloud services still matter, however, because larger models, centralized governance and cross-facility analytics are difficult to run on devices alone. The likely long-term pattern is not edge versus cloud, but edge and cloud working together across a shared infrastructure stack.
What the shift means for healthcare buyers and vendors
For health systems, the move toward edge AI changes procurement priorities. Buyers now need to ask not only what a device can do clinically, but also how it is updated, how it is monitored and how long it will remain supportable as software evolves.
For vendors, the stakes are equally high. The ability to manage devices at scale, deploy updates safely and maintain predictable performance may become as important as the underlying algorithm. Partnerships between chipmakers, cloud providers, device manufacturers and health systems are likely to deepen as the stack becomes more integrated.
The broader implication is that AI in healthcare is entering an operational phase. The next wave of adoption will depend less on proving that AI can work in a lab and more on proving that it can work reliably across thousands of devices, multiple sites and changing clinical conditions. What to watch next is whether more hospitals standardize edge-to-cloud platforms and whether regulators, vendors and providers can align on the controls needed to make them safe at scale.
Frequently Asked Questions
Why can’t healthcare simply keep using cloud AI if it already works for training and analytics?
Cloud AI is still useful, but many clinical tasks need immediate responses and must keep working even when connectivity is unstable. Edge AI processes data where it is generated, which reduces latency, helps protect sensitive information, and supports use cases like monitoring or robotics that cannot wait for a round trip to the cloud.
How does an edge-to-cloud healthcare system split responsibilities without creating confusion for staff?
The edge usually handles real-time inference at the device, while the cloud supports model training, fleet management, analytics and oversight across multiple sites. In practice, hospitals need clear governance so IT, clinical engineering and cybersecurity teams know which system version is running, where data goes, and who is responsible for updates and monitoring.
If edge devices are making decisions locally, how do hospitals keep them compliant and auditable?
Hospitals need audit trails, telemetry, identity controls and centralized management to track device behavior over time. Local processing does not remove compliance requirements; it changes how they are enforced. The cloud can still collect logs, support validation, and help document software versions, performance changes and security events across the device fleet.
What is the biggest operational risk after an AI medical device is deployed?
Model drift is one of the biggest risks. A system can perform well in testing but degrade when used with a different patient population, imaging protocol or workflow. That is why ongoing validation, monitoring and periodic updates matter. Healthcare organizations should treat deployment as the beginning of lifecycle management, not the end.
Which healthcare use cases are most likely to adopt edge AI first, and why?
Edge AI tends to appear first in settings where speed and reliability matter most, such as imaging, patient monitoring, surgical robotics and connected devices like infusion pumps. These environments generate continuous data and often need decisions in seconds, so local processing can improve responsiveness while still sending information back for review.