97% of US Database Pros Run Outdated Software as Healthcare AI Piles On

A 2026 survey found 97% of U.S. database professionals admit their organizations still run at least some outdated software versions — and healthcare is now pushing AI onto that aging stack faster than anyone is checking whether it holds.

That number comes as OpenAI teams up with Epic, the largest electronic health record (EHR) company in the U.S., while major national insurers embed solutions from the same AI companies onto their platforms. Health systems keep stacking chat overlays and AI layers on top of it all.

The math underneath is older still. Reporting from 2023 cited Dell estimating that 70% of software used by Fortune 5000 companies was developed at least 20 years ago. Effie Carlson, CEO of Watershed Health, laid out the risk in stark terms: healthcare is "essentially testing the bridge's weight limit by driving more trucks across it instead of checking the bridge's rating."

What happens at 2 a.m. in the ER?

Carlson opens with a concrete scene. It's 2 a.m. in the ER, and a physician asks the hospital's new AI assistant to summarize a patient's chart. The cursor spins for minutes while the AI tool waits on the decades-old database underneath to answer — something that database was never designed to do.

Her analogy: it's like asking a 50-year-old runner to sprint with a 25-year-old. A pulled muscle or torn ligament is coming.

In another industry, a slow or sometimes-crashing system creates headaches for the sales team. In healthcare, patients are at risk. The doctor may have forgotten how to find the information "the old way" while the cursor spins.

Every piece of healthcare software sits on data architecture built for one clinician clicking through one screen at a time. An AI chatbot queries that data in real time and decides what output the doctor sees. Old or misread data could easily become a confident chatbot response to medical staff.

The takeaway: the underlying systems were never designed to be asked that quickly or that often.

Why isn't anyone fixing the plumbing?

Rebuilding data architecture is expensive, complicated and genuinely risky. Upgrading a production database carries operational risk, including compatibility problems and service disruption. Running unsupported software creates security and reliability risks.

It's like painting a house every few years without fixing what's behind the walls — the aging plumbing or electrical wiring — until something breaks. Companies and governments struggle to fund modernization when it has no direct line to revenue, especially when the only return is avoiding a failure. It seems easier just to repaint.

Scale changes the stakes. If one hospital runs a small AI pilot and it goes wrong, the damage stays small. If every hospital in a health system across a state moves at once, it could trigger a crisis of its own making — without ever knowing it was possible.

Leaders worry about cybersecurity and hacking, per the American Hospital Association's 2025 cybersecurity year review, but few are looking underneath the hood.

Who owns the risk?

Carlson argues stakeholders should share responsibility before fallout arrives:

• Health system technologists must confirm infrastructure can support a new tool in production before deploying it.
• Vendors adding AI to existing healthcare platforms must know their own systems' capabilities, because that creates risk for customers.
• AI builders bolting onto someone else's underlying system will push that system's limits — so what's the plan when it freezes up?

What should leaders do now?

Carlson prescribes three steps:

1. Audit reality. Health systems and insurers must understand what their infrastructure can handle when AI comes calling — how old systems are, and how they perform when AI and staff query simultaneously.
2. Invest. Leaders should fund infrastructure upgrades with the same urgency they bring to AI rollouts, supporting or replacing underpinning technologies before it's too late.
3. Commit to a roadmap. Modernization is a multiyear effort, and the long-term return is AI that actually works.

The pressure isn't staying contained, either. Multiply AI queries by thousands of users, hitting the EHR day and night alongside normal staff. A slowdown under that duress doesn't stay inside the AI tool — it drags down every workflow across the entire health system.

And the 50-year-old runner? If he keeps sprinting, he won't have just one injury. He'll have them over and over again. Decades-old architecture breaks down the same way when it's asked to keep up indefinitely.

The value of AI, Carlson writes, cannot be realized until the industry addresses what nobody wants to talk about — "that AI adoption is moving faster than anyone's understanding of the risk beneath it."