The debate over AI and inequality has largely focused on people—who will have access to the technology, who will benefit from it and who risks being left behind. But inside healthcare, another AI divide is already taking shape. This one is between the hospitals that have the money, talent and infrastructure to put AI to work and those that don't.
The numbers are striking. According to federal data on hospital AI adoption, 71% of U.S. hospitals used predictive AI integrated with their electronic health records in 2024. But adoption reached 96% among large hospitals compared with just 59% among small hospitals. Among hospitals belonging to multi-hospital systems, 86% used predictive AI. Among independent hospitals, it was only 37%.
That gap matters because AI is quickly becoming part of how hospitals survive, not simply an innovation project. The same federal data show some of the fastest-growing hospital AI applications are decidedly practical, including billing and scheduling. These are exactly the kinds of labor-intensive functions where financially stressed hospitals need to find efficiencies.
Yet those hospitals may be the least equipped to make the investment.
Hospitals remain extraordinarily dependent on people to operate. According to the American Hospital Association, workforce spending accounted for about 60% of hospital expenses in 2025, exceeding $1 trillion. Total hospital expenses grew 7.5% that year, more than twice the growth in hospital prices.
This creates a dangerous catch-22. The hospitals with the greatest need to use technology to redesign labor-intensive operations frequently have the least capacity to invest in doing so.
A large health system can afford an AI pilot that fails. It can hire an innovation team, absorb a lengthy implementation, build governance and compliance infrastructure, and try again if the economics don't work. A safety-net, community or independent hospital often has far less room for error. A multi-year software contract that might eventually produce savings still requires money today.
And buying the technology is only the beginning.
Successful AI deployment requires workflow redesign, change management, integration, measurement and governance. Even among hospitals already using predictive AI, the federal data show meaningful variation in how comprehensively models are evaluated for accuracy, bias and performance after implementation. Smaller, rural and independent hospitals also continue to lag their larger counterparts in adoption.
If we leave the market to work this out one hospital at a time, we should expect that divide to compound. Systems that successfully automate expensive workflows can reinvest those savings into their next technology investment. They build institutional knowledge. Their employees get better at implementation. They become more attractive partners to technology companies looking for sophisticated customers and proving grounds.
The hospitals that wait don't simply remain where they are. They lose ground. The answer cannot be telling every hospital to develop a better AI strategy. We need to change the ownership and financing model for healthcare AI.
Hospitals already know how to collaborate when acting collectively creates an advantage. Group purchasing organizations allow them to aggregate purchasing power rather than negotiate for every product independently. Clinical research networks allow institutions to pool patients, expertise and infrastructure to answer questions no single organization could address as effectively alone.
We should apply the same instinct to AI. Instead of requiring every health system to independently buy, implement and govern variations of the same technology, health systems should increasingly pool their purchasing power, operational expertise and capital to build common AI capabilities. And smaller, independent and safety-net organizations need a path into those collaborations rather than simply another vendor contract.
The venture community has a responsibility here as well. We need financing structures that acknowledge the economic reality of the organizations that really need to adopt technologies. That could mean performance-based models where payment is tied to verified savings rather than large upfront commitments. It could mean consortium models in which multiple health systems share implementation resources and governance.
And increasingly, it should mean giving health systems ownership in the companies and infrastructure they help create. Ownership matters because the economics of AI are changing quickly. As the technology itself becomes increasingly commoditized, enduring value will come from distribution, proprietary workflows, real-world implementation knowledge and access to the environments where healthcare actually happens. Health systems contribute enormously to each of those things. Under the traditional vendor model, however, they can provide the workflows, clinical expertise and proving ground and then pay someone else an annual fee for the resulting product.
That bargain deserves to be reconsidered. Collective ownership can help address the trust problem, too. AI adoption isn't only a technology decision. It requires decisions about validation, monitoring, clinical safety and who is accountable when something goes wrong. Federal researchers found that hospitals increasingly rely on multiple parties to evaluate AI, with 74% reporting more than one entity accountable for evaluation. Smaller hospitals should not each have to recreate that infrastructure from scratch.
Shared infrastructure could include common validation standards, clinical oversight, monitoring and implementation expertise. That would give more hospitals access not only to AI, but to the capabilities required to deploy it responsibly.
None of this means every hospital needs to own every AI company, or that traditional vendors disappear. It means we should stop assuming that the only scalable model for healthcare innovation is one technology company selling essentially the same capability to thousands of hospitals one contract at a time.
The AI divide between hospitals is already here. If adoption continues to depend primarily on the size of an organization's balance sheet, the institutions best positioned today will keep accumulating advantages while those with the least room for error fall further behind.
AI gives hospitals a real opportunity to break their dependence on increasingly expensive, labor-intensive ways of operating. But the organizations that need that opportunity most cannot be the last ones able to afford it. Preventing that outcome will require more than making AI cheaper. It will require giving more health systems the ability to build, govern and share in the technology's value, transforming them.
AI should become something hospitals build together, not another advantage reserved for those that could already afford to move first.
Amanda DeMano is SVP of venture creation at Abundant Venture Partners.
Editor's Note: Abundant Venture Partners is a strategic investor in the makers of certain AI technologies, including SpendRule and Playback Health.