Health systems keep pouring money into Epic as margins tighten and AI governance lags behind

Children's Minnesota is mid-way through a $175 million Epic implementation covering 45 distinct change management topics, a figure that captures the full operational weight these projects place on health system IT and operations teams. That number, reported by Becker's Hospital Review, sits at the high end of the range but is hardly an outlier. Becker's separately documented the costs of four other major health system Epic projects this week, finding that large-scale EHR investment remains a consistent line item even as hospital margins stay under pressure.

Margin pressure hasn't slowed EHR spending

According to Becker's Hospital Review, health systems are still committing significant capital to EHR projects despite a financial environment that has forced layoffs and job cuts at dozens of organizations in 2026. The persistence of that spend reflects how deeply Epic and comparable platforms are embedded in clinical operations. Pulling back on a live implementation or deferring a planned go-live carries its own set of risks and costs that often outweigh the short-term budget relief.

WellSpan Health's CIO offered one concrete strategy for managing that burden: moving Epic to Amazon Web Services. The executive told Becker's Hospital Review that the AWS migration is expected to directly accelerate Epic upgrade cycles, reducing the manual infrastructure work that typically delays deployments of new Epic releases. Cloud hosting won't eliminate EHR costs, but it shifts the cost structure and can free internal IT capacity that would otherwise be consumed by routine maintenance.

For rural and smaller systems, the cost equation looks different. Wolters Kluwer and Epic recently announced a bundled offering that packages three clinical decision-support tools specifically for rural hospitals, according to Becker's Hospital Review. The partnership is a direct response to the capital constraints facing smaller facilities, which need proven clinical content integrated at the point of care but cannot absorb the same per-project investment that major academic medical centers routinely approve.

AI deployment is outpacing the infrastructure to govern it

While EHR budgets hold steady, a separate and growing vulnerability is taking shape on the AI side of the IT portfolio. A report cited by Becker's Hospital Review found that most health systems lack dedicated AI testing platforms. That means clinical AI tools are being evaluated and deployed without purpose-built validation environments, a gap that carries direct implications for patient safety, regulatory exposure, and vendor accountability.

The gap is not from lack of interest. Becker's Hospital Review reported this week that health systems are increasingly engaging with what it calls the "AI-informed patient," individuals who arrive at appointments having already researched their conditions using AI tools. Physician leaders, including Northwell Health's senior vice president of neurosurgery, have discussed publicly how AI can support patient education and strengthen clinical relationships when used appropriately. The demand signal from patients is real, and health system leaders are responding.

The problem is that deploying AI at the point of care requires more than a vendor contract. It requires a testing environment where models can be evaluated against real clinical workflows before they touch active patient encounters. Without that infrastructure, governance defaults to policy documents and vendor attestations rather than empirical validation. For CIOs managing both a complex EHR stack and an expanding AI portfolio, that creates compounding risk.

CIOs are navigating both fronts simultaneously

The leadership dimension of this moment is significant. Becker's Hospital Review reported this week that health system CIOs have been leaving their roles at a notable pace throughout 2026, with many moving into vendor, advisory, or consulting roles rather than lateral health system positions. That pattern matters operationally: institutional knowledge of live EHR configurations and nascent AI programs walks out with each departure, and successor CIOs inherit both the capital commitments and the governance gaps.

For IT and operations teams, the Children's Minnesota implementation offers a useful benchmark. Forty-five change management topics across a single Epic project reflects the cross-functional coordination required: clinical informatics, supply chain, revenue cycle, training, and integration teams all have to move in sequence. A $175 million budget number means little if the organizational change management isn't resourced to match, and most health systems running these projects are doing so while managing ongoing operational demands.

What procurement and IT teams should be watching

Three pressure points are converging for health system technology leaders right now. EHR costs are not declining, AI governance infrastructure is lagging behind deployment, and leadership continuity is a live risk as CIO turnover continues. Each is manageable in isolation. Together, they require a deliberate sequencing of investment and governance decisions that many teams have not yet formalized.

The Wolters Kluwer and Epic rural bundle is one example of the market responding to a specific segment's constraints with a pre-integrated solution. Expect more of that pattern: vendors packaging clinical content, AI tooling, and EHR integration together in ways that reduce the per-facility implementation burden. For procurement teams, that means evaluating bundles on total cost of ownership and clinical validation depth, not just license fees.

WellSpan's AWS move points in a parallel direction. Cloud infrastructure for EHR hosting is becoming less of a competitive differentiator and more of a baseline operational decision. The question for health system IT leaders is no longer whether to move Epic to the cloud, but which migration path minimizes disruption to live upgrade cycles and frees internal engineering capacity for higher-value work, including building the AI testing environments that most organizations still don't have.