AI healthcare needs low-latency links, says DE-CIX chief

DE-CIX Chief Executive Officer Ivo Ivanov has outlined the network requirements he believes are necessary for wider use of AI in healthcare, where essential health services remain out of reach for 4.5 billion people worldwide.

He argued that medical AI is already delivering results in remote training, telesurgery and emergency care, but that wider adoption depends on low-latency links between data centres, hospitals, cloud systems and mobile networks. His assessment centred on the limits that distance and network performance impose when clinical decisions or procedures must be carried out in real time.

The backdrop is a global shortage of medical staff. Ivanov cited figures showing there are 17.2 doctors per 10,000 people worldwide and fewer than one trained surgeon per 100,000 people in some low- and middle-income countries. The World Health Organisation has estimated a shortfall of 11 million health workers by 2030.

Against that backdrop, investment in generative AI for healthcare is expected to rise sharply, with spending forecast to grow from USD $2.79 billion in 2025 to USD $50.79 billion in 2035, according to Ivanov.

Access limits

AI could extend the reach of health services rather than replace doctors, Ivanov said, pointing to uses including immersive training, remote mentoring, diagnostics, patient monitoring and robot-assisted surgery. In each case, practical deployment depends less on the model itself than on where computing resources are located and how quickly data can move between systems.

For virtual and augmented reality support in surgery, latency should not exceed 20 milliseconds, according to Ivanov. He said that threshold limits data travel to no more than 1,800 kilometres, and only on fully optimised networks, before delays become disruptive for a clinician using the system.

"Did you know that half of the world's population - some 4.5 billion people - still lack access to essential health services?" Ivanov said.

For inexperienced doctors in rural settings, he said, VR and AR systems can provide remote mentoring from specialists as well as AI guidance during procedures. The technology can also support consultations and diagnostics, but its usefulness depends on a stable, low-latency connection.

Telesurgery places even tougher demands on infrastructure. Haptic-enabled force feedback systems, which allow a surgeon to feel resistance through robotic controls, work best at less than 10 milliseconds of latency, Ivanov said. That translates to a distance of no more than 1,000 kilometres unless operators make major infrastructure investments.

He cited studies showing a 25% reduction in operative time, a 30% decrease in intraoperative complications and a 40% increase in surgical precision for telesurgery using force feedback. Those figures suggest the technology could improve outcomes if hospitals can secure the required network conditions.

Ivanov also referred to a long-distance operation carried out between Florida and Angola, in which a cancerous prostate was removed over 11,000 kilometres of fibre-optic cable with network latency of 140 milliseconds. He said such procedures remain limited to planned operations on non-critical patients, with AI-assisted control used to offset the delay.

Emergency response

In emergency medicine, connected ambulances equipped with AI support and robotic tools are even more sensitive to latency, Ivanov said. He put their operating requirement at about three milliseconds, which limits them to a radius of roughly 250 to 300 kilometres from the hospital and the computing infrastructure they rely on.

These systems can connect to patient wearables before arrival, support rapid diagnosis through imaging and AR links, and use traffic management systems to speed the route to the hospital. Ivanov said this has produced a 30% improvement in survival rates for critical patients, a 40% reduction in certain medical errors by paramedics, 20 minutes faster time to definitive treatment and a 50% reduction in response times in congested urban areas.

DE-CIX argues that this puts interconnection at the centre of medical AI deployment. According to Ivanov, inference workloads increasingly need to run close to where they are used, often in edge data centres, while maintaining access to patient records, cloud systems and network status data.

Security is also central, he said, because healthcare systems must move highly sensitive information between multiple environments. Private interconnection arrangements, alongside redundancy across fibre, 5G and low Earth orbit satellite links, are among the measures he identified as necessary to support clinical use.

"Because without infrastructure, without the flowing, exchanging, and processing of data, there can be no AI," Ivanov said.