88% of healthcare organizations and 73% of government bodies say: our infrastructure isn’t ready for AI on-premises. Where it gets stuck, and what to do about it.
The ambition is there. Often the budget too. Yet AI initiatives in healthcare and government run aground at the same point: the infrastructure underneath wasn’t built for it.
In healthcare, 88 percent of IT leaders say their own environment isn’t fully ready to run AI workloads on-premises. In government and education, that figure is 73 percent. Both sit at or above the global average of 82 percent, according to the sector editions of the Nutanix Enterprise Cloud Index 2026. We’ll walk you through why AI workloads make different demands than your existing applications, where it goes wrong in practice, and which choices you can make now so your infrastructure doesn’t stand in the way of your AI ambitions.
Why AI makes different demands than your existing applications
Your current application landscape was designed around one assumption: data travels to the application, the application runs centrally, and the user waits a fraction of a second. For an EHR screen or a permit application, that works fine.
AI workloads break that assumption on three points.
- They’re data-hungry. An AI model analyzing images or interpreting real-time signals processes volumes that traditional applications don’t come close to. That changes the math on where you store your data and how you move it.
- They’re latency-sensitive. On a dashboard, half a second of delay is annoying. For an AI model watching along during a procedure, or flagging a fraud signal before a transaction goes through, that same half second is the difference between usable and worthless.
- They move around. A model you train in the cloud today, you want to run tomorrow at the place where the data originates. That only works if your environment supports that move.
That’s exactly why the research names containerization as the core of the solution. In healthcare, 81 percent expect containerization to increase, and 80 percent already build new applications in containers. In government, it’s higher still: 87 percent expect to lean more on containers over the next three years. Containers aren’t an end in themselves, they make workloads portable.
Healthcare: AI comes to the bedside
In care organizations, AI is shifting from the data center to the place where care is delivered. The research expects that up to 75 percent of healthcare data will soon originate at the bedside.
The figures make concrete what that means. A single-patient room generates up to 7 TB of data a year. An ICU bed has fifteen to twenty connected devices. Sending all that data to a public cloud, having it analyzed, and sending the answer back costs time you don’t have when it comes to clinical decisions. And on top of that: if your external connection drops, your AI support drops with it.
Run those workloads closer to the source, and you solve three things at once. The delay disappears. Patient data stays within the hospital’s walls. And clinical continuity holds up, even if your WAN connection fails.
That explains why 54 percent of care organizations already run containerized applications on-premises or in a private cloud today, against 47 percent in the public cloud. Not out of principle, but because clinical practice demands it.
The (semi-)public sector: service delivery that can’t falter
Government organizations face the same question, with a different weight to it. There, AI is used for grant applications, fraud detection, permit processes, and services to citizens. Applications where the question “why did the system decide this” has to get an answer you can explain to a citizen, a journalist, or a judge.
That makes where the processing happens a governance question, not just a technical one. Who has access to the data. Under which jurisdiction the environment falls. Whether you can demonstrate what happened.
Government organizations are keenly aware of this. 91 percent of IT leaders in government and education call uncontrolled AI use a serious risk to the mission and to security. That’s the highest percentage of all the sectors studied, higher than healthcare (83 percent) and higher than the global average (87 percent).
Enthusiasm without guardrails creates risk, and shadow AI isn’t just a security issue but a governance gap. The approach is to combine infrastructure modernization with clear digital policy frameworks, so employees can innovate within a safe framework.
Two sectors, the same problem, opposite blind spots
Line up the figures and a striking picture emerges.
Healthcare lags furthest behind on infrastructure (88 percent not fully ready) but raises the least alarm about uncontrolled AI use (83 percent). In government, it’s the other way around: less of a gap on infrastructure (73 percent) but the sharpest risk awareness of any sector (91 percent).
That means something different for each sector. Care organizations are probably underestimating the risk of what’s already running uncontrolled. Government organizations do see the risk, but have to translate that awareness into infrastructure that makes safe use possible. Knowing the risk isn’t the same as managing it.
What “ready for AI” concretely means
Readiness isn’t an on-off switch. It’s a set of properties your environment either has or doesn’t.
- Workloads can move. You need to be able to run an application in your data center, in a private cloud, or at a location, without rebuilding it. That’s what containers give you.
- Compute close to the data. Not all processing has to be local, but the latency-sensitive kind does. That calls for capacity in the place where it matters.
- One management layer across everything. If workloads run spread across data center, cloud, and locations, you don’t want three management teams with three toolkits. Without a single overview, every outage becomes a hunt.
- Demonstrable control. Who has access, where the data sits, what happened. For NEN 7510, GDPR, and public accountability, you have to be able to show that, not just promise it.
- Recoverability. An AI application supporting clinical or public service delivery has become business-critical. Your backup and recovery strategy has to match that.
Why hybrid is the outcome, not the compromise
Going fully to the public cloud doesn’t solve the latency problem and collides with your compliance requirements. Running everything on-premises yourself is expensive and holds back your scalability.
The research shows that in practice, organizations choose the middle ground. In healthcare, 63 percent run AI applications through a managed service provider, where an external party hosts or manages the container infrastructure so that workloads can run both centrally and locally.
That’s not a stopgap. It’s the recognition that different workloads need different places, and that you need a party to keep the whole thing manageable.
How we approach this
At co-one, this starts with workload placement: determining, per application, where it should run, based on latency, data classification, compliance requirements, and cost. Not as a one-off architecture choice, but as a decision you can make again when the requirements change.
Around that, we build the rest of the foundation. Network and connectivity that reliably connect locations and cloud environments. Observability so you can see what’s running and where it’s straining, before a user notices. Cyber resilience so that an incident doesn’t mean an outage in care or service delivery.
With Nutanix as a strategic partner, we build environments where the same workload can run in your data center, in a private cloud, or at a location, with one management layer across the top. For care organizations, that means AI support that keeps working when the connection drops. For government organizations, that you can demonstrate where data sits and who can access it.
Where you start
You don’t have to replace your entire landscape to make progress.
- Take inventory of which AI applications are running now or planned, and what each of them needs in terms of latency, data volume, and data classification.
- Determine, per application, where it should run. Often it turns out that only part of it really has to be local.
- Check whether your current environment supports that division. Can you move a workload without rebuilding it?
- Start with one application where the gain is clearest. Prove it there, then scale.
At Curaçao Medical Center and Saxenburgh Medical Center, we’ve walked that route: get the foundation right first, then make room for what goes on top.
Want to know where your environment stands?
We run a workload placement analysis: which applications run where, what that costs in performance and compliance, and what could be better. Concrete, within a few weeks, without you having to commit to a migration project first.
Sources: Nutanix Healthcare Vertical Enterprise Cloud Index Report and Nutanix Public Sector Enterprise Cloud Index Report (June 2026), and the Nutanix Enterprise Cloud Index. Research conducted by Wakefield Research in November 2025 among 1,600 IT, cloud, and engineering leaders at organizations with more than 500 employees across fourteen markets, including the Netherlands.
