Surgeons in one country operating on patients thousands of kilometers away. Ambulances functioning as mobile intensive care units, supported by real-time imaging and specialist guidance. Medical students training inside fully immersive virtual environments, propelling their careers forward. These scenarios would have been filed under “science fiction” just one short decade ago, but now, thanks to AI, they are firmly within reach, and in some instances, already happening.
In September 2025, surgeons in China completed what was described as the world’s first 5G-enabled remote robotic surgery performed at extreme altitude, with the operating surgeon located thousands of kilometers away from the patient. The Apollo Hospital in Bengaluru, India, has also introduced one of the world’s first 5G-connected ambulances, fitted with onboard cameras, live diagnostic tools, and paramedic bodycams to allow remote guidance from specialists at the hospital during emergency situations. Meanwhile, Stanford Medicine in California has adopted new software that combines imaging from MRIs, CT scans and angiograms to create 3D models that physicians and patients can see and manipulate in real-time, furthering their understanding of certain medical conditions. The future has arrived.
AI is now reshaping diagnostics, surgical precision, emergency response, and clinical training. But as impressive as the algorithms are, they are not the full story. The success stories previously outlined are to be celebrated, but there’s a reason they are few and far between – and it’s not due to the limitations of AI. Healthcare is now entering a phase where AI systems are no longer isolated tools, but active collaborators embedded directly into clinical workflows. When AI is supporting a diagnosis, guiding a scalpel, or coordinating emergency care, any issues with connectivity, such as latency, become a major risk factor. In the context of healthcare, performance is defined by how fast data can move, how predictably systems can respond, and how seamlessly human expertise can be augmented by machine intelligence. The difference between a viable remote procedure and an unusable one is often not compute power or model accuracy, but whether the underlying infrastructure can deliver ultra-low latency, consistently and at scale.
From model training to medical inference
Much of the public conversation around AI in healthcare still focuses on training. We talk about larger models, richer datasets, and the compute required to build them. That phase is important, but it is not where medicine actually happens. Training is episodic, centralized, and surprisingly quite tolerant of delay. It can take place in controlled environments, far removed from the operating theatre, the emergency room, or the back of an ambulance. Inference, however, where AI is actually used out in the field, is continuous, distributed, and unforgiving. It’s the moment when AI is called upon to interpret an image, flag an anomaly, recommend a course of action, or support a clinician in real time. In other words, inference is where decisions are made – and it’s highly dependent on reliable, low-latency connectivity.
Inference depends on constant, real-time access to data, models, and supporting systems that are usually spread across multiple locations. A surgical robot doesn’t carry all its intelligence locally. A connected ambulance relies on live streams of video and vital signs, and seamless access to medical records and wearable devices. Immersive training environments and remote mentoring systems require continuous synchronization between users, devices, and AI-driven analytics. In each case, delays compound quickly. In these environments, latency puts a limit on what is possible. When inference is slowed, guidance arrives late, feedback loses precision, and confidence in the system erodes. This is why the next phase of AI-driven healthcare will be defined less by how intelligent the models are, and more by how efficiently that intelligence can be delivered, everywhere it is needed, without delay.
Why latency matters
In other industries, latency is a question of optimization or productivity. In healthcare, however, it can literally make or break the function of a potentially life-saving tool. Different medical applications impose very different demands on the network. Take the use of AR/VR systems for remote surgical guidance as an example, these environments rely on precise synchronization between what a clinician sees, hears, and does. Once latency rises much beyond 20 milliseconds, visual misalignment and motion discomfort begin to appear. For a surgeon relying on augmented overlays or remote guidance, even slight disorientation can compromise accuracy and safety. So, what feels like a minor delay in consumer applications becomes unacceptable when a human hand is operating on living tissue from a remote site. Every additional kilometer data must travel, every unnecessary routing hop, adds delay. At small scale, this can sometimes be managed. At national or international scale, it exposes the limits of legacy network architectures that were never de-signed for continuous, real-time clinical interaction. And therein lies the challenge.
The need for connected intelligence
The reality of AI-enabled healthcare is that no application succeeds on its own. A connected ambulance is only as effective as its ability to exchange data with emergency departments in real time. Remote surgery depends on continuous, predictable communication between surgeons, robotic systems, imaging platforms, and patient data. Training and mentoring environments rely on synchronized interaction between people, devices, and AI systems, often spread across multiple locations. In each case, the clinical outcome is shaped less by any single technology than by how efficiently the entire system communicates. How directly and seamlessly the diverse networks and resources are interconnected so that data can flow unhindered. When connectivity is fragmented, delayed, or unreliable, intelligence becomes isolated and decisions arrive too late.
As care becomes more distributed and more data-driven, the ability to move information quickly, securely, and without unnecessary detours becomes the deciding factor in whether the success we’ve seen so far can be deployed at scale.
DE-CIX operates as one of the world’s largest, carrier and data-center-neutral internet exchanges, interconnecting thousands of networks across 80+ countries for low-latency peering, cloud connectivity, and resilient, high-performance interconnection services that power modern digital, AI, and cloud applications worldwide.
