Survey Shows Healthcare AI is Reaching Bedsides but IT Systems Need a Boost

An AI agent at ECU Health now reads incoming patient-transfer records and distills each case into a three-sentence summary, saving the rural North Carolina healthcare organization roughly 20 hours of chart review per week in its first month of use. That kind of result is becoming less of an outlier as AI moves fast from experimentation into daily operations for many healthcare providers. But as AI enters clinical care it puts greater pressure on the IT infrastructure needed to power it. 

That’s according to the 2026 Healthcare Enterprise Cloud Index (ECI), a Nutanix-supported survey of healthcare IT leaders. Results showed that 57% expect to adopt agentic AI or autonomous agents over the next three years, but 88% do not consider their current infrastructure fully ready for on-premises AI workloads.

“AI in healthcare is past the hype phase. It’s here,” said Leah Gabbert, marketing director of global solutions at Nutanix.

The ECI shows that healthcare providers are deploying AI often on infrastructure built for a different era. Among healthcare respondents, 62% expect to use generative AI over the next three years, 57% expect agentic AI or autonomous agents, and 55% expect predictive analytics or machine learning models, according to the ECI . Those numbers reflect a sector that has moved well past the question of whether to adopt AI and is now wrestling with how to do it at scale.

AI Reaches the Point of Care

The quickest returns, Gabbert says, are coming from routine work that eats into clinicians’ time.

“Right now, the quickest wins we’re seeing are on the operational side,” she said. “Think about doctors spending a few hours each day writing notes. We’re seeing a huge uptake in ambient scribes that listen to visits and draft clinical notes automatically, which is a lifesaver for the physician experience.”

Other applications are moving directly into patient care. NHS virtual wards allow some seriously ill children to receive hospital-level care at home. Wearable devices and other connected tools transmit vital signs to clinical teams, which can use software to identify changes that may require attention.

In another example, the Cleveland Clinic has expanded an AI system that reviews patient data for signs of sepsis across its hospitals. Early deployments helped clinicians identify more cases, reduce false alerts, and receive alerts earlier.

Deploying an application for one department or clinical team is one thing. Supporting similar systems across an entire health system and patients’ homes is a much bigger lift for IT.

Infrastructure Requirements and Gaps

ECI data points to three requirements that determine whether a provider can support AI at scale: raw compute capacity, data quality and system stability.

The first is capacity. Some workloads run right where the patient is. A single intensive care room may contain 15 to 20 connected devices and generate up to 7 terabytes of data per year, according to Nutanix. Sending every signal to a distant cloud can introduce delays and make the application dependent on an external network connection, which can get costly.

“But to do that safely, you have to run that intelligence right at the Point of Care, or the Edge,” Gabbert said. “You can’t wait for data to travel back and forth to a distant cloud.”

Hospitals might train a model in a public cloud or central data center, then run time-sensitive analysis closer to the bedside.

“If an AI is monitoring an ICU patient, you can’t risk an internet outage dropping that connection,” Gabbert said. “The latency is too high, and the stakes are too big.”

The second requirement is data quality. AI systems in healthcare draw on EHRs, imaging platforms, bedside devices, laboratories and other sources, much of which remains siloed across separate systems and organizations. At the HIMSS26 Global Health Conference, Mayo Clinic Chief Clinical Systems and Informatics Officer Edwina Bhaskaran said providers preparing for AI agents need to examine data quality, where information is stored and whether agents can retrieve it across systems.

“This happens quite often,” Bhaskaran said, quoted in an article by Healthcare Dive. 

“You still have to have someone call over and say, ‘Can you fax me records?’” So, while there’s a lot of data that we do share, there’s still quite a bit of data that we don’t.”

The third requirement is stability. Newer AI applications are landing alongside EHRs, imaging systems, and other established software clinicians use every day, without destabilizing them. Those applications often run in virtual machines (VMs) that healthcare IT teams have secured and maintained for years.

“VMs give hospitals a secure, isolated environment they already trust,” Gabbert said. “They aren’t stubbornly clinging to the past; they’re using VMs as a safety net.”

IT Sovereignty Requirements

The ECI report found that 72% of healthcare IT leaders consider data sovereignty a top infrastructure priority, and 54% feel pressure to keep infrastructure within one country because of regulatory, security or stakeholder requirements. Nutanix defines data sovereignty as the principle that data is governed by the laws of the jurisdiction where it resides. But location is only part of the issue.

Justin Sherman, a nonresident senior fellow at the Atlantic Council’s Cyber Statecraft Initiative, wrote that U.S. cross-border data rules can affect the entire healthcare AI supply chain, including training data, models, APIs, and software development kits. The requirements are becoming more concrete. Privacy attorneys Kate Black and Mason Fitch wrote in an April 2026 analysis for the International Association of Privacy Professionals that healthcare and life sciences organizations face data localization mandates, remote access restrictions, vendor reviews, audits, and recordkeeping obligations. They pointed to Florida, which requires providers using certified EHR systems to keep off-site patient information within the United States, its territories, or Canada.

Meeting those requirements starts with visibility. Gartner says AI sovereignty depends on controlling both data and models, using metadata to track how information moves across an organization and its supply chain. That visibility also helps expose shadow AI. The ECI found that 79% of healthcare organizations have encountered unauthorized AI applications, while 83% consider them a serious business and data risk.

Technology then enforces the boundaries set by IT, security, compliance, and clinical leaders. Nutanix says hospitals can place sensitive AI applications on premises or in a private cloud, while access controls, encryption, audit records, and workload policies govern who reaches the data and where processing occurs. Sovereignty is control over where patient data and AI models operate, who can access them, and which laws govern every party in the chain.

Containers as a Solution

Placing an application on premises or in a private cloud sounds simple until you have to do it without tearing out systems that already work. Containers enable healthcare organizations to thread that needle.

By definition, they package applications with their dependencies so they can run consistently in a hospital data center, private cloud, public cloud or edge location. That portability lets providers place workloads where they make the most sense — training in the public cloud, sensitive or latency-dependent processing closer to the point of care.

The ECI found that 86% of healthcare IT leaders say AI is increasing their use of containers . But containers have limits. They cannot create clean data, supply computing capacity or establish governance on their own. Nutanix says readiness still requires a flexible hybrid model: infrastructure that can support both VMs and containers, manage applications across locations and apply common security and data policies.