Healthcare AI Is Booming. So Why Are Providers Still Losing Billions?

Hospitals and health systems are pouring billions into artificial intelligence, yet the average US provider continues to hemorrhage money — losing an estimated $200 billion collectively in 2025. The culprit is not a lack of innovation, experts say, but a profound misdirection of it: the industry has focused automation on the wrong targets.

WHO did WHAT: A growing chorus of health economists and AI researchers argue that healthcare providers are deploying artificial intelligence primarily to milk more revenue from fee-for-service billing and prior authorization — rather than reducing actual care costs, cutting clinical waste, or improving patient outcomes. WHERE: Across the United States, from large academic medical centers to rural community hospitals. WHEN: This pattern has accelerated over the past three years, as generative AI tools flooded the market. WHY NOW: With operating margins still razor-thin (the median hospital lost 1.2% on patient care in 2024), the disconnect between AI hype and financial reality has become impossible to ignore.

The context is stark. US healthcare spending surpassed $4.5 trillion in 2024, yet nearly one in three hospitals ran negative margins. Venture capital and corporate R&D funneled an estimated $12 billion into healthcare AI last year alone — much of it aimed at automating documentation, coding, and revenue cycle management. These tools undoubtedly speed up billing but do little to reduce the underlying cost of delivering care.

Key players include Epic Systems, which embedded GPT-4 for automated note-taking; Cerner (now Oracle Health), which launched an AI-powered prior authorization copilot; and Google Health, whose Med-PaLM 2 showed strong clinical reasoning but has seen limited real-world deployment. A 2025 study by the National Bureau of Economic Research found that hospitals using AI to automate administrative tasks saw a 3% revenue boost — but no improvement in total episode cost. Meanwhile, AI tools that actually reduce length of stay or prevent readmissions remain underinvested.

Analysis: The roots are structural. The US fee-for-service payment model rewards volume, not value. Automating billing workflows directly boosts revenue, creating a strong financial incentive for health systems to prioritize coding AI over clinical AI. “We’re using a Ferrari to run errands around the parking lot instead of driving it on the highway,” said Dr. Ziad Obermeyer, a leading AI-in-medicine researcher at UC Berkeley. The result: providers drown in data but drown no less in costs. Until reimbursement models fundamentally change, AI will likely continue to optimize a broken system rather than fix it.

Outlook: Several Medicare Accountable Care Organizations are experimenting with capitated payments that could flip incentives. New entrants like Hippocratic AI and Ambience Healthcare are building clinical-first AI for nurse triage and discharge planning. If value-based care continues its slow expansion — currently covering about 20% of Medicare beneficiaries — the demand for cost-reducing AI may finally match the supply. The next two years will be critical: either health systems pivot their AI budgets toward outcomes, or the bill for automation without transformation comes due.