How AI Is Pushing Health Care Costs Higher, Not Lower

The promise attached to artificial intelligence in health care has been remarkably consistent. Automate the paperwork, sharpen the diagnoses, cut the administrative bloat, and a system that consumes roughly 17% of U.S. GDP would finally start to bend its cost curve downward. The early returns suggest the opposite is happening. AI is being layered onto an already expensive system in ways that generate new revenue streams, new billing opportunities, and new administrative friction, and the bill is landing on patients and employers.

What's actually happening

The clearest example is on the insurance side. Payers have adopted algorithmic tools to review claims and prior authorization requests at scale. In theory, automation should speed approvals. In practice, several large insurers have faced litigation alleging their AI systems were tuned to deny claims faster and in higher volume than human reviewers ever could. The dispute is no longer hypothetical. When a model can process and reject thousands of claims with minimal human oversight, the economics of denial shift. Each denied claim that a patient gives up on contesting is money the insurer keeps.

On the provider side, the dynamic is different but points the same direction. Hospitals and physician groups are deploying AI-assisted documentation and coding tools that surface billable services and justify higher-acuity codes. Ambient scribes that transcribe a visit can also flag every reimbursable element of that visit. The result is more complete capture of billable activity, which is good for a provider's margin and bad for whoever pays the claim. "Upcoding" concerns predate AI, but software that does it automatically and consistently removes the friction that previously kept it in check.

Why the savings haven't materialized

The core problem is incentive alignment. AI lowers cost only when the entity deploying it is rewarded for spending less. Most of U.S. health care runs on fee-for-service economics, where revenue scales with volume and billing intensity. Drop a productivity tool into that environment and it optimizes for the existing incentive, which is to bill more, not less.

There is also the matter of who captures the efficiency gain. If an AI tool lets a radiology group read 20% more scans per day, that capacity tends to convert into more billed studies rather than lower prices. The labor savings exist, but they accrue to the vendor selling the software and the practice that owns the productivity, not to the patient. Health systems are spending heavily on these platforms, and that capital expense gets folded into the cost base that ultimately sets prices and premiums.

The market context

Venture and corporate investment in health care AI has run into the billions annually, and the sales pitch to hospital executives is almost never "this will reduce your revenue." It is "this will protect your margin, capture more reimbursement, and reduce labor cost." Vendors are responding rationally to what buyers will pay for. A tool that demonstrably cut a hospital's net revenue would be difficult to sell, regardless of its benefit to the broader system.

Employers, who fund a large share of private coverage, are the constituency most exposed and least equipped to push back. Premium growth has continued to outpace wage growth, and AI-driven administrative tooling on both the payer and provider sides adds cost on both ends of every transaction. The insurer's claims-processing software and the provider's coding software are, in effect, two AI systems negotiating against each other, with the administrative overhead of both baked into the final price.

What it means

The technology is not the problem. The same models that accelerate denials could, under different incentives, identify unnecessary care, flag pricing outliers, or simplify the prior authorization process out of existence. Some integrated systems that bear financial risk for outcomes, such as capitated and value-based arrangements, are beginning to use AI this way, because they keep the savings they generate. That structural difference matters more than the algorithm itself.

For now, the trajectory is set by where the money flows. Until reimbursement rewards spending less rather than billing more, AI deployed across the U.S. health system will keep doing what its buyers ask of it, and what they are asking for is not lower prices. Regulators reviewing algorithmic claim denials and coding practices will shape how aggressive these tools can be, but they will not change the underlying incentive. The expectation that automation would make care cheaper assumed a market that wanted cheaper care. The evidence so far is that the market wants higher throughput, and AI is very good at delivering it.