The current discourse surrounding artificial intelligence (AI) in health care demands a more rigorous examination of the disparity between prospective capabilities and empirically validated outcomes. As enthusiasm outpaces evidence, health care institutions face a critical inflection point.
From the perspective of a researcher whose work in medical AI predates the current enthusiasm cycle, and whose belief in the technology’s fundamental promise remains unwavering, a significant concern has emerged: health care entities are inadequately prepared for the substantial market correction that is demonstrably underway. The enterprise AI sector is exhibiting characteristics consistent with a speculative bubble nearing its maximum expansion. For health care, the salient inquiry is not the inevitability of this market recalibration, but whether institutional investments have been strategically focused on optimizing patient welfare or merely on maximizing public relations value.
Signs of Market Correction
The enterprise AI market appears to be facing a crisis of expectations. Surveyed CEOs report that only 25 percent of AI initiatives have delivered expected returns on investment over the past three years, with just 16 percent scaled enterprise-wide (IBM, 2025). The share of companies abandoning most of their AI projects jumped to 42 percent in 2025, more than doubling from 17 percent the previous year (Axis Intelligence, 2025). Even as 91 percent of organizations plan to increase AI spending, the timeline for realizing returns has lengthened from months to years (Horton et al., 2025).
Health care AI appears to be following a similar trajectory. Digital health funding looked steady in Q3 2025 at $3.5 billion, but beneath this surface lies structural instability: megadeals now account for 39 percent of total funding, while Series B deal flow has thinned to just 30 raises through Q3 2025, compared to more than 60 annually over the past four years (Knowles et al., 2025). Series B funding marks a critical transition, during which startups must demonstrate they can scale beyond pilot programs into sustainable businesses with real revenue growth, effectively separating promising concepts from viable companies. This decline in Series B activity suggests that many AI health companies are struggling to prove commercial viability.
The pharmaceutical sector tells a similar story. When Recursion Pharmaceuticals and Exscientia merged in August 2024, following a 78.7 percent valuation collapse, it appeared more like triage than strategy. Combined, these companies had consumed hundreds of millions without producing a single US Food and Drug Administration–approved drug (Buntz, 2024). The patterns suggest the occurrence not of isolated failures but of systemic overvaluation across the AI health care landscape.
Patient-Centered vs. Promise-Chasing Strategies
As someone who worked in medical AI before the current enthusiasm cycle, the author has witnessed multiple waves of technological optimism crash against the realities of implementation. What matters now is whether institutions have been building sustainable capabilities aligned with their clinical mission or accumulating technologies aligned with market narratives.
The distinction will become evident when the correction materializes. Organizations built for patients will have infrastructure that continues to advance care, regardless of market sentiment. Organizations that chased promises will have deployed systems that do not materially improve outcomes, spent resources that could have addressed actual clinical problems, and built dependencies on vendors rather than capabilities.
Diagnostic Questions
The following diagnostic questions can help indicate which category an institution falls into.
Can your institution independently validate vendor claims on your own patient population?
If not, you have built dependency on vendor promises. When those promises fail to materialize, you will have neither the tools nor the capability to course-correct. Independent validation capacity represents foundational infrastructure for responsible AI deployment.
Are your AI investments addressing clinical outcomes and needs, or are they primarily focused on workflow optimization?
Studies of over 2.5 million uses of ambient documentation demonstrate time savings of approximately 20 minutes daily and measurable reductions in burnout (Tierney et al., 2024; Tierney et al., 2025). This represents real value with real evidence. However, it is fundamentally the automation of a clerical task. Organizations built exclusively for workflow optimization helped their staff but did not advance their clinical mission. When the correction reveals this gap, they will realize they optimized the wrong things.
Have you ever declined to deploy an AI system that worked technically but lacked clinical evidence to support its use?
This is perhaps the most diagnostic question. Organizations that declined systems lacking prospective evidence established that patient outcomes matter more than keeping pace with competitors. Those that approved systems broadly established that deployment was the goal, not patient benefit.
When your AI systems fail, can you identify the reason and determine who is accountable?
Few institutions can specify performance thresholds that trigger reassessment or track performance drift over time. Almost none have clear accountability structures. Organizations without a monitoring infrastructure have not built for patient safety. When regulatory or legal reckoning comes, they will be unable to demonstrate due diligence.
Do your investments address problems patients care about, or problems administrators care about?
Health care AI investment has focused disproportionately on administrative efficiency, while avoiding the more complex challenges of outcomes, equity, and access. The 19 percent of patients expecting affordability improvements and 20 percent expecting better physician relationships suggest something important about misaligned priorities (Nong and Ji., 2025). Organizations that made this choice helped their operations but did not advance patient care.
Have you built institutional capacity to evaluate AI or just procurement relationships?
Capacity involves understanding algorithmic approaches, maintaining validation infrastructure, convening cross-functional teams, and establishing governance. Organizations with only procurement relationships will find themselves dependent on vendors when the correction thins the landscape. Organizations with capacity can continue advancing their mission regardless of market conditions.
Can you articulate the stopping criteria for systems currently in production?
If not, you deployed without adequately considering patient safety. Organizations that built for patients established stopping criteria before deployment. Organizations that chase promises assume systems will work and are unprepared when they do not.
Practical Implications
The correction will not primarily harm institutions financially. The harm will be the opportunity cost. Every dollar spent on AI that does not improve patient outcomes is a dollar not spent on interventions that do. Every clinician’s hour spent learning systems that do not change care is an hour not spent with patients. Health care resources are finite; their allocation reflects institutional values.
Organizations built for patients will emerge with validated infrastructure, monitoring capabilities, and institutional expertise that will continue to serve their mission. They will have learned which applications genuinely improve care. They will not be disappointed when inflated promises deflate because they were never investing in promises—they were investing in proof.
Organizations that chased promises will have deployed systems that do not materially improve outcomes, built dependencies on vendors that may no longer exist, and spent scarce resources on technologies that did not advance their mission. Most critically, they will have disappointed their clinicians and patients by celebrating deployment while delivering limited clinical impact.
A Call for Honest Assessment
The broader AI market appears to be teaching a lesson: deployment is not validation, adoption is not evidence, and potential is not performance. Health care, with its fundamental obligation to first do no harm, should have learned this lesson first. Instead, many institutions have followed market enthusiasm rather than clinical evidence.
Organizations should ask themselves one fundamental question: when the bubble bursts and market enthusiasm recedes, will they have infrastructure that continues advancing patient care, or will they have technologies that no longer seem innovative?
Those who built for patients, demanded evidence, validated independently, monitored rigorously, and focused on outcome-relevant problems will find the answer reassuring. Those who chased promises will find it troubling.
The era of frameworks and position statements has ended. Institutions are now making concrete choices about whether AI investments serve their clinical mission or their institutional optics. The correction will clarify which organizations made which choice. The time for honest institutional self-assessment is now, before the market makes its own assessment.
Join the Conversation
New from #NAMPerspectives: The Opportunity Cost of Market Narratives in Health Care AI
Read the commentary: https://doi.org/10.31478/202606a
—
“As enthusiasm outpaces evidence, health care institutions face a critical inflection point.” A new #NamPerspectives commentary discusses the health care AI market bubble and how institutions can keep patient care at the center of their investments.
More: https://bit.ly/3PHaqYs
References
Axis Intelligence. 2025. AI ROI strategy 2025: From $50M investment to measurable returns. Available at: https://axis-intelligence.com/ai-roi-strategy-2025-ultimate-guide/ (accessed November 23, 2025).
Buntz, B. 2024. AI drug discovery space consolidates with $688M Recursion and Exscientia merger. Drug Discovery & Development. Available at: https://www.drugdiscoverytrends.com/recursion-and-exscientia-merge-in-688-m-deal-to-create-consolidated-ai-driven-drug-discovery-platform/ (accessed January 31, 2026).
Horton, R., J. Michalski, S. Winters, D. Gunn, and J. Holland. 2025. AI ROI: The paradox of rising investment and elusive returns. Deloitte. Available at: https://www.deloitte.com/dk/en/issues/generative-ai/ai-roi-the-paradox-of-rising-investment-and-elusive-returns.html (accessed November 23, 2025).
IBM. 2025. IBM study: CEOs double down on AI while navigating enterprise hurdles. IBM Newsroom. Available at: https://newsroom.ibm.com/2025-05-06-ibm-study-ceos-double-down-on-ai-while-navigating-enterprise-hurdles (accessed November 23, 2025).
Knowles, M., J. Rivera, and M. Somaiya. 2025. Q3 2025 market overview: Signals out of sync. Rock Health. Available at: https://rockhealth.com/insights/q3-2025-market-overview-signals-out-of-sync/ (accessed November 23, 2025).
Nong, P., and M. Ji. 2025. Expectations of healthcare AI and the role of trust: Understanding patient views on how AI will impact cost, access, and patient-provider relationships. Journal of the American Medical Informatics Association 32(5):795-799. https://doi.org/10.1093/jamia/ocaf031.
Tierney, A. A., G. Gayre, B. Hoberman, B. Mattern, M. Ballesca, P. Kipnis, V. Liu, and K. Lee. 2024. Ambient artificial intelligence scribes to alleviate the burden of clinical documentation. NEJM Catalyst 5(3). https://doi.org/10.1056/CAT.23.0404.
Tierney, A. A., G. Gayre, B. Hoberman, B. Mattern, M. Ballesca, S. B. Wilson Hannay, K. Castilla, C. S. Lau, P. Kipnis, V. Liu, and K. Lee. 2025. Ambient artificial intelligence scribes: Learnings after 1 year and over 2.5 million uses. NEJM Catalyst 6(5). https://doi.org/10.1056/CAT.25.0040.
The Opportunity Cost of Market Narratives in Health Care AI
The current discourse surrounding artificial intelligence (AI) in health care demands a more rigorous examination of the disparity between prospective capabilities and empirically validated outcomes. As enthusiasm outpaces evidence, health care institutions face a critical inflection point.
From the perspective of a researcher whose work in medical AI predates the current enthusiasm cycle, and whose belief in the technology’s fundamental promise remains unwavering, a significant concern has emerged: health care entities are inadequately prepared for the substantial market correction that is demonstrably underway. The enterprise AI sector is exhibiting characteristics consistent with a speculative bubble nearing its maximum expansion. For health care, the salient inquiry is not the inevitability of this market recalibration, but whether institutional investments have been strategically focused on optimizing patient welfare or merely on maximizing public relations value.
Signs of Market Correction
The enterprise AI market appears to be facing a crisis of expectations. Surveyed CEOs report that only 25 percent of AI initiatives have delivered expected returns on investment over the past three years, with just 16 percent scaled enterprise-wide (IBM, 2025). The share of companies abandoning most of their AI projects jumped to 42 percent in 2025, more than doubling from 17 percent the previous year (Axis Intelligence, 2025). Even as 91 percent of organizations plan to increase AI spending, the timeline for realizing returns has lengthened from months to years (Horton et al., 2025).
Health care AI appears to be following a similar trajectory. Digital health funding looked steady in Q3 2025 at $3.5 billion, but beneath this surface lies structural instability: megadeals now account for 39 percent of total funding, while Series B deal flow has thinned to just 30 raises through Q3 2025, compared to more than 60 annually over the past four years (Knowles et al., 2025). Series B funding marks a critical transition, during which startups must demonstrate they can scale beyond pilot programs into sustainable businesses with real revenue growth, effectively separating promising concepts from viable companies. This decline in Series B activity suggests that many AI health companies are struggling to prove commercial viability.
The pharmaceutical sector tells a similar story. When Recursion Pharmaceuticals and Exscientia merged in August 2024, following a 78.7 percent valuation collapse, it appeared more like triage than strategy. Combined, these companies had consumed hundreds of millions without producing a single US Food and Drug Administration–approved drug (Buntz, 2024). The patterns suggest the occurrence not of isolated failures but of systemic overvaluation across the AI health care landscape.
Patient-Centered vs. Promise-Chasing Strategies
As someone who worked in medical AI before the current enthusiasm cycle, the author has witnessed multiple waves of technological optimism crash against the realities of implementation. What matters now is whether institutions have been building sustainable capabilities aligned with their clinical mission or accumulating technologies aligned with market narratives.
The distinction will become evident when the correction materializes. Organizations built for patients will have infrastructure that continues to advance care, regardless of market sentiment. Organizations that chased promises will have deployed systems that do not materially improve outcomes, spent resources that could have addressed actual clinical problems, and built dependencies on vendors rather than capabilities.
Diagnostic Questions
The following diagnostic questions can help indicate which category an institution falls into.
Can your institution independently validate vendor claims on your own patient population?
If not, you have built dependency on vendor promises. When those promises fail to materialize, you will have neither the tools nor the capability to course-correct. Independent validation capacity represents foundational infrastructure for responsible AI deployment.
Are your AI investments addressing clinical outcomes and needs, or are they primarily focused on workflow optimization?
Studies of over 2.5 million uses of ambient documentation demonstrate time savings of approximately 20 minutes daily and measurable reductions in burnout (Tierney et al., 2024; Tierney et al., 2025). This represents real value with real evidence. However, it is fundamentally the automation of a clerical task. Organizations built exclusively for workflow optimization helped their staff but did not advance their clinical mission. When the correction reveals this gap, they will realize they optimized the wrong things.
Have you ever declined to deploy an AI system that worked technically but lacked clinical evidence to support its use?
This is perhaps the most diagnostic question. Organizations that declined systems lacking prospective evidence established that patient outcomes matter more than keeping pace with competitors. Those that approved systems broadly established that deployment was the goal, not patient benefit.
When your AI systems fail, can you identify the reason and determine who is accountable?
Few institutions can specify performance thresholds that trigger reassessment or track performance drift over time. Almost none have clear accountability structures. Organizations without a monitoring infrastructure have not built for patient safety. When regulatory or legal reckoning comes, they will be unable to demonstrate due diligence.
Do your investments address problems patients care about, or problems administrators care about?
Health care AI investment has focused disproportionately on administrative efficiency, while avoiding the more complex challenges of outcomes, equity, and access. The 19 percent of patients expecting affordability improvements and 20 percent expecting better physician relationships suggest something important about misaligned priorities (Nong and Ji., 2025). Organizations that made this choice helped their operations but did not advance patient care.
Have you built institutional capacity to evaluate AI or just procurement relationships?
Capacity involves understanding algorithmic approaches, maintaining validation infrastructure, convening cross-functional teams, and establishing governance. Organizations with only procurement relationships will find themselves dependent on vendors when the correction thins the landscape. Organizations with capacity can continue advancing their mission regardless of market conditions.
Can you articulate the stopping criteria for systems currently in production?
If not, you deployed without adequately considering patient safety. Organizations that built for patients established stopping criteria before deployment. Organizations that chase promises assume systems will work and are unprepared when they do not.
Practical Implications
The correction will not primarily harm institutions financially. The harm will be the opportunity cost. Every dollar spent on AI that does not improve patient outcomes is a dollar not spent on interventions that do. Every clinician’s hour spent learning systems that do not change care is an hour not spent with patients. Health care resources are finite; their allocation reflects institutional values.
Organizations built for patients will emerge with validated infrastructure, monitoring capabilities, and institutional expertise that will continue to serve their mission. They will have learned which applications genuinely improve care. They will not be disappointed when inflated promises deflate because they were never investing in promises—they were investing in proof.
Organizations that chased promises will have deployed systems that do not materially improve outcomes, built dependencies on vendors that may no longer exist, and spent scarce resources on technologies that did not advance their mission. Most critically, they will have disappointed their clinicians and patients by celebrating deployment while delivering limited clinical impact.
A Call for Honest Assessment
The broader AI market appears to be teaching a lesson: deployment is not validation, adoption is not evidence, and potential is not performance. Health care, with its fundamental obligation to first do no harm, should have learned this lesson first. Instead, many institutions have followed market enthusiasm rather than clinical evidence.
Organizations should ask themselves one fundamental question: when the bubble bursts and market enthusiasm recedes, will they have infrastructure that continues advancing patient care, or will they have technologies that no longer seem innovative?
Those who built for patients, demanded evidence, validated independently, monitored rigorously, and focused on outcome-relevant problems will find the answer reassuring. Those who chased promises will find it troubling.
The era of frameworks and position statements has ended. Institutions are now making concrete choices about whether AI investments serve their clinical mission or their institutional optics. The correction will clarify which organizations made which choice. The time for honest institutional self-assessment is now, before the market makes its own assessment.
Join the Conversation
New from #NAMPerspectives: The Opportunity Cost of Market Narratives in Health Care AI
Read the commentary: https://doi.org/10.31478/202606a
—
“As enthusiasm outpaces evidence, health care institutions face a critical inflection point.” A new #NamPerspectives commentary discusses the health care AI market bubble and how institutions can keep patient care at the center of their investments.
More: https://bit.ly/3PHaqYs
References
Axis Intelligence. 2025. AI ROI strategy 2025: From $50M investment to measurable returns. Available at: https://axis-intelligence.com/ai-roi-strategy-2025-ultimate-guide/ (accessed November 23, 2025).
Buntz, B. 2024. AI drug discovery space consolidates with $688M Recursion and Exscientia merger. Drug Discovery & Development. Available at: https://www.drugdiscoverytrends.com/recursion-and-exscientia-merge-in-688-m-deal-to-create-consolidated-ai-driven-drug-discovery-platform/ (accessed January 31, 2026).
Horton, R., J. Michalski, S. Winters, D. Gunn, and J. Holland. 2025. AI ROI: The paradox of rising investment and elusive returns. Deloitte. Available at: https://www.deloitte.com/dk/en/issues/generative-ai/ai-roi-the-paradox-of-rising-investment-and-elusive-returns.html (accessed November 23, 2025).
IBM. 2025. IBM study: CEOs double down on AI while navigating enterprise hurdles. IBM Newsroom. Available at: https://newsroom.ibm.com/2025-05-06-ibm-study-ceos-double-down-on-ai-while-navigating-enterprise-hurdles (accessed November 23, 2025).
Knowles, M., J. Rivera, and M. Somaiya. 2025. Q3 2025 market overview: Signals out of sync. Rock Health. Available at: https://rockhealth.com/insights/q3-2025-market-overview-signals-out-of-sync/ (accessed November 23, 2025).
Nong, P., and M. Ji. 2025. Expectations of healthcare AI and the role of trust: Understanding patient views on how AI will impact cost, access, and patient-provider relationships. Journal of the American Medical Informatics Association 32(5):795-799. https://doi.org/10.1093/jamia/ocaf031.
Tierney, A. A., G. Gayre, B. Hoberman, B. Mattern, M. Ballesca, P. Kipnis, V. Liu, and K. Lee. 2024. Ambient artificial intelligence scribes to alleviate the burden of clinical documentation. NEJM Catalyst 5(3). https://doi.org/10.1056/CAT.23.0404.
Tierney, A. A., G. Gayre, B. Hoberman, B. Mattern, M. Ballesca, S. B. Wilson Hannay, K. Castilla, C. S. Lau, P. Kipnis, V. Liu, and K. Lee. 2025. Ambient artificial intelligence scribes: Learnings after 1 year and over 2.5 million uses. NEJM Catalyst 6(5). https://doi.org/10.1056/CAT.25.0040.
Estiri, H. 2026. The Opportunity Cost of Market Narratives in Health Care AI. NAM Perspectives. Commentary, National Academy of Medicine, Washington, DC. https://doi.org/10.31478/202606a.
https://doi.org/10.31478/202606a
Hossein Estiri, PhD, is Associate Professor of Medicine at Harvard Medical School and Director of the Clinical Augmented Intelligence (CLAI) Group at Massachusetts General Hospital.
None to disclose.
DISCLAIMER
The views expressed in this paper are those of the author and not necessarily of the author’s organizations, the National Academy of Medicine (NAM), or the National Academies of Sciences, Engineering, and Medicine (the National Academies). The paper is intended to help inform and stimulate discussion. It is not a report of the NAM or the National Academies. Copyright by the National Academy of Sciences. All rights reserved.
Related Perspectives