Why 83% of Leaders Say Hospitals Struggle With AI — And the Fix

AI is poised to transform healthcare, but outdated infrastructure is blocking progress.

With 83% of healthcare leaders citing legacy data systems as the top barrier, hospitals must modernize their infrastructure to meet AI’s demands for scale, speed, and compute power.

According to IT Medical, a global software engineering and consulting services provider, there are three main infrastructure requirements needed to support AI:

1. Data Storage: AI requires a centralized data lake or warehouse containing full Electronic Health Record (EHR) data and operational database copies.
2. Compute: Training AI models at scale demands GPU-optimized infrastructure designed for parallel processing, not standard Central Processing Units (CPUs).
3. Cloud vs. On-Premises AI Infrastructure: While cloud services support general AI, specialized healthcare applications often require on-premises systems for compliance, performance, and cost control.

IT Medical creates AI-powered solutions for the healthcare sector. To successfully prepare for AI adoption, it has developed a five-step checklist for CIOs, CTOs, and IT leaders on how to build a scalable, compliant infrastructure that speeds up AI adoption and impact.

Editor’s note: This article was created in partnership with IT Medical.

Building Effective AI Infrastructure

This five-best-practices checklist was designed to strengthen leadership, collaborative processes, minimize risks, and help organizations plan ahead, among other benefits:

1. Secure C-Suite Buy-In and Cross-Functional Collaboration

Healthcare staff often associate AI with issues like inaccurate outputs or hallucinating chatbots.

To address this skepticism, stakeholders should be involved throughout the AI development process, including setting requirements to validating models, and providing clear, consistent leadership to build trust.

2. Design Infrastructure for Security, Privacy, and Compliance

With frameworks like HIPAA, GDPR, and ISO 27001 still applying, AI infrastructure raises significant data privacy and cybersecurity concerns.

Storing all patient data in a single repository makes AI systems vulnerable to cyberattacks.

AI teams often have far more access than regular IT systems, increasing the risk of a serious data breach. To reduce this, anonymization should start early.

The solution? Remove patient identifiers in Extract, Transform, Load (ETL) pipelines and eliminate identity data wherever possible.

3. Design for Scalability and Future Growth

Implementing AI is an ongoing process that requires constant model updates and improvements, so your infrastructure will need to keep up with growing demands.

Don't rule out future cloud migration strategies because AI services will evolve and expand. Plan for the growth of storage and compute resources to handle new data and take advantage of the latest processing technologies.

This can be done by:

• Investing in flexible, scalable storage and compute systems
• Preparing for new data types, like imaging, genomics, and patient-reported outcomes
• Leaving space for future cloud migration, even if starting with on-premises systems

4. Standardize Data Movement
and Preserve Integrity

AI development, which requires copying healthcare data into repositories, can cause small data changes through ETL pipelines.

Before deployment, compare the model’s recommendations with actual decisions made by healthcare professionals. This will ensure it doesn’t affect real-world decision-making.

Differences in data formats across systems can lead to errors that are difficult to detect. Because ETL changes the data, there's no way to be sure the model will perform exactly the same in real-world use as it did during development.

5. Monitor Continuously and Update Periodically

Unlike traditional software development, which relies on fixed code logic, AI models evaluate hidden variables. Factors like a doctor’s shift, changes in hospital operations, new research, or shifts in team dynamics can all impact the AI system's performance.

To keep things running smoothly, the AI team should keep a close eye on performance, gather feedback, and update the model with new data when needed through:

• Ongoing model performance monitoring
• Regular retraining with updated datasets
• User feedback loops to flag unexpected behavior

“At IT Medical, we help healthcare organizations tackle the most pressing challenges of AI adoption, from safeguarding patient data and meeting compliance requirements to designing infrastructure that supports real-time performance.

Our team combines deep technical and clinical expertise to modernize outdated systems, strengthen data privacy, and deliver AI solutions that scale with evolving needs and regulations,” said Aleksandra Buimistere, Head of PR at IT Medical.

Addressing infrastructure challenges is the key to unlocking AI’s potential in healthcare, especially around data storage, compute resources, and security.

For example, IT Medical’s work in reducing hospital discharge delays has proven how these strategies can lead to real improvements.