Health systems are spending big on EHRs and AI while governance gaps widen

Children's Minnesota is spending $175 million on its Epic transition and had to manage 45 distinct change topics to get there, according to Becker's Hospital Review. That figure is not an outlier. Becker's reporting on the cost of four recent health system Epic projects shows that large-scale EHR modernization has settled into the nine-figure range as a baseline, even as margins across the industry remain under pressure.

The capital commitment is striking on its own. What makes it operationally significant for CIOs and procurement leaders is the context around it: health systems are absorbing these costs at the same moment they are being asked to evaluate, deploy, and govern a fast-growing catalog of clinical AI tools that have no established reimbursement pathway and, in most organizations, no dedicated infrastructure for testing them.

EHR modernization is still the anchor spend

Health systems are not slowing EHR investment despite tighter margins, according to Becker's Hospital Review coverage this week. The Children's Minnesota case is among the more detailed public disclosures: $175 million, 45 change management topics, and a deployment scope that illustrates how an EHR migration has evolved from an IT initiative into a full organizational transformation program.

WellSpan Health is approaching the same challenge from a different angle. The Pennsylvania-based system's CIO told Becker's Hospital Review that a migration to Amazon Web Services will accelerate the pace of Epic upgrades, framing cloud infrastructure as an enabler of faster iteration on the EHR rather than a separate workstream. The logic is straightforward: if upgrade cycles are shorter, the system stays closer to Epic's current feature set and reduces the technical debt that drives up future migration costs.

On the vendor side, Wolters Kluwer and Epic announced a bundle of three clinical tools aimed specifically at rural hospitals, according to Becker's Hospital Review. Bundling reduces the procurement complexity for smaller systems that lack the IT staff to evaluate, contract, and integrate multiple point solutions. For rural health IT leaders, that consolidation has real operational value even before the clinical benefits are measured.

Clinical AI is scaling faster than the governance structures around it

Most health systems lack dedicated AI testing platforms, according to a report cited by Becker's Hospital Review. The finding is a concrete governance gap: without a sandboxed environment to validate AI tools against real clinical workflows before go-live, organizations are effectively deploying on faith. That is a procurement and risk management problem as much as a technology one.

The reimbursement picture compounds the risk. Modern Healthcare reported this month that clinical AI has no established payment model, and experts working on the question have not yet converged on what one should look like. Health systems deploying AI-assisted clinical decision support, ambient documentation, or AI-driven diagnostics are doing so without a mechanism to recover the cost through payer contracts. That math matters to any CFO or VP of operations evaluating whether to expand an AI pilot into production.

The governance question is also surfacing in labor contracts. Modern Healthcare reported that nurses in New York and California have secured multiyear agreements with HCA and Sutter that include explicit provisions giving clinical staff a voice in how AI tools are deployed. For health system operations leaders, that development signals a new stakeholder group in AI procurement decisions. Contracts that skip clinical staff input may face friction at implementation, regardless of what the technology can do.

The AI-informed patient adds another variable

Becker's Hospital Review reported this week that health systems are beginning to engage with what it calls the AI-informed patient: people who arrive at clinical encounters having already researched their symptoms, diagnoses, or treatment options using AI tools. A separate piece in Becker's, attributed to Daniel Sciubba, MD, senior vice president of neurosurgery at Northwell Health, addresses how AI can be used to educate patients and reinforce rather than undermine physician-patient relationships.

For operations leaders, the AI-informed patient is not a consumer trend. It is a workflow and training issue. Clinicians need protocols for evaluating the quality of information a patient has already ingested, and organizations deploying patient-facing AI tools need to think about what those tools are actually telling people before discharge. Neither challenge has a standard answer yet.

What this means for your team

• Audit your AI testing infrastructure now. If your organization lacks a dedicated platform to validate AI tools before clinical deployment, that gap represents a live compliance and patient safety risk. Prioritize building or procuring that capability before expanding any AI pilot to production scale.
• Model the EHR total cost of ownership beyond the contract. Children's Minnesota's $175M figure is a useful benchmark: budget explicitly for change management, training, and workflow redesign, not just licensing and implementation fees, when evaluating or renegotiating an Epic agreement.
• Treat AI reimbursement ambiguity as a budget risk. Until payers establish payment models for clinical AI, every deployment is an unfunded operating cost. Build that assumption into multi-year AI roadmaps and flag it in vendor negotiations where vendors are positioning tools as revenue-generating.
• Include clinical staff in AI procurement governance. The nurse contract provisions at HCA and Sutter are a leading indicator. Organizations that build clinical input into AI evaluation criteria before signing contracts will face fewer deployment obstacles than those that try to retrofit it after go-live.