No Roads , No Revolution: Why Healthcare’s Broken Data Infrastructure Is Undermining AI Innovation
In the age of generative models and multimodal clinical assistants , healthcare AI is often heralded as the next frontier — one poised to solve physician burnout , accelerate diagnosis , and optimize health system efficiency . Yet beneath the optimism lies a sobering reality : the infrastructure needed to power scalable , clinically meaningful AI simply does not exist in most of U.S. healthcare .
This is the warning issued by Mitesh Rao , MD, MHS, CEO of OMNY Health, in a recent interview with MedCity News . According to Rao , without interoperable and representative data at scale , even the most advanced AI algorithms will struggle to deliver value beyond surface-level automation .
“We haven’t built the roads — we’re trying to put Ferraris on dirt ,” Rao said . “ Until we address the fragmented , inaccessible nature of healthcare data , innovation in AI will hit a ceiling .”
AI Growth, Infrastructure Stagnation
Investment in healthcare AI has surged . In 2024 alone , U.S. digital health startups raised more than $4.3 billion for AI- driven platforms , according to Rock Health. Companies like Abridge ( ambient documentation ), Ambience (AI scribe systems ), and Hippocratic AI ( LLMs for clinical communication ) have crossed the billion-dollar valuation mark . Regulatory agencies , including the White House, are actively promoting frameworks to integrate AI into critical infrastructure sectors like healthcare .
But the explosion in funding and hype has outpaced progress in solving the most fundamental constraint : access to clean , standardized , and interoperable data .
Most successful healthcare AI applications today operate in narrow domains — revenue cycle automation , clinical documentation , or administrative workflows — precisely because they depend on structured data that's relatively easy to access . When it comes to more ambitious use cases , such as predictive diagnostics , AI- assisted clinical decision-making , or population-level disease modeling , innovation consistently runs into a wall : incompatible systems , siloed records , and a lack of common data standards .
Siloes , Standards, and Status Quo Incentives
The root of the problem is the fragmented nature of healthcare’s IT ecosystem . Patient data is scattered across a multitude of proprietary EHRs , claims systems , and clinical repositories — each with its own taxonomy , access controls , and institutional ownership .
Despite federal efforts such as the 21st Century Cures Act and CMS’ Interoperability and Patient Access Rule , progress remains slow . While application programming interfaces ( APIs ) like FHIR (Fast Healthcare Interoperability Resources) have made inroads , widespread adoption is uneven , and real-world implementation is patchy .
“The incentives are not aligned ,” Rao emphasized . “ Large incumbents don’t gain financially by making data more portable . In many cases , the business model depends on vendor lock-in .”
Vendors like Epic and Oracle Cerner , which together cover the majority of hospital EHR market share in the U.S., have faced longstanding criticism for limiting data portability , imposing high interface fees , or offering incomplete access through proprietary APIs . While both have made public commitments to interoperability , data liquidity often remains tightly controlled in practice .
Policy without Penalties Falls Flat
Although CMS and ONC continue to introduce interoperability initiatives — including TEFCA ( Trusted Exchange Framework and Common Agreement)— many experts argue that current policies lack teeth . Without strict enforcement mechanisms or direct financial consequences for noncompliance , health systems and vendors have little motivation to change .
The issue isn't only technical — it's also economic and political . Without reimbursement models that reward data sharing or penalize information blocking , the burden falls on innovators to “ work around ” systemic fragmentation .
“ If you’re building AI in retail , you get access to millions of rows of clean transaction data . In healthcare , you get 20 Excel files from five systems in three different formats , none of which talk to each other ,” noted a senior data engineer at a health tech startup , who requested anonymity due to contractual restrictions .
The Clinical Cost of Inaccessible Data
The consequences of this data fragmentation are not merely operational — they are clinical .
AI models trained on limited , non-representative , or outdated datasets risk entrenching bias , missing rare conditions , or failing to generalize beyond a narrow setting . A 2022 JAMA Network Open study found that most clinical AI models in use were developed using data from fewer than five healthcare systems — raising serious concerns about generalizability and safety .
Moreover , lack of real-time data access impairs AI tools designed for dynamic clinical settings , such as inpatient monitoring or predictive deterioration scoring . When models can’t “ see ” the latest labs , medications , or care plan changes , their clinical utility diminishes sharply .
Interoperability as National Infrastructure
What’s needed , Rao and other experts argue , is a reframing of healthcare data infrastructure as a form of national critical infrastructure — akin to roads , energy grids , or air traffic control .
Building such an ecosystem would require :
- Standardized , enforced data formats across EHR vendors and health systems
- Mandatory data access APIs with usage transparency
- Real penalties for information blocking , as outlined in the Cures Act
- Neutral data intermediaries to facilitate secure , de-identified exchange
- Public- private partnerships to maintain infrastructure across regions
The U.K.’s NHS Spine and Estonia’s national health record system offer models where centralized data exchange , real-time patient access , and regulatory coordination have enabled greater digital innovation without compromising security .
From Hype to High Ground
The promise of AI in healthcare is not unfounded . But translating that promise into measurable outcomes will require more than technical advances or venture capital — it will demand infrastructure reform , policy resolve , and a willingness to prioritize the invisible foundation behind every clinical algorithm : trustworthy , connected , and actionable data .
“ We’re trying to build the future of healthcare on a foundation that can barely support email attachments ,” Rao quipped . “ Until that changes , the revolution will remain stalled on the runway .”
Sources :
- MedCity News, “ Healthcare’s Poor Data Infrastructure Is Hindering AI Innovation,” Katie Adams , August 2025.
- JAMA Network Open, “ External Validity of Published Clinical Machine Learning Models ,” 2022.
- CMS Interoperability and Patient Access Final Rule , 2020.
- Office of the National Coordinator for Health IT (ONC), TEFCA Framework.
- Rock Health Funding Database, 2024.
- OMNY Health, corporate communications .