人工智能将为数十亿人提供护理，并打破构建它的系统

EMERGING TECH RESEARCHAI Will Deliver Care to Institutional Research Group Billions and Break theSystem That Built It Brian WrightLead Research Analyst,Healthcare Ben ZercherSenior Research Analyst,Biotech & Pharma PitchBook is a Morningstar company providing the most comprehensive, mostaccurate, and hard-to-find data for professionals doing business in the private markets. Ben RiccioResearch Analyst,Industry & Technology Research Adi GeorgeAssociate Data Analyst Key takeaways pbinstitutionalresearch@pitchbook.com AI is already pervasive in care delivery on both the consumer and physician sides ofthe equation. We envision a world where cognitive diagnosis functions are near free,yet the downstream effects are likely to drive higher medical cost trends in the nearterm before seeing improved medical outcomes and lower costs in the long term. Themassive cost burden of higher near-term AI-induced utilization (five to seven years)will bring the system to a financing breaking point with the cumulative increase in Published on April 23, 2026 Contents Key takeaways Executive summary AI in healthcare delivery is pervasive AI clinical-care delivery stack Outcomes data limitations Beyond ambient scribing—optical andsensor information Payment models (reimbursement-based medicine) CMMI ACCESS program economics provideglimpse into long-term reimbursementmechanisms23The near-term value of physician cognitive functions increases with AI-induced higherutilization, but scenarios exist where it is significantly arbitraged away long-term, as CaredeliveryvisionsforthefutureAcknowledgements31focus, whereas traditional value-based care model investments are living on borrowedtime. As true outcome measures are developed and tracked, AI-native behavioral Appendix We advance eight proprietary thesis points that thread throughout this report: We believe fragmentation of thedelivery system, combined withthe irreducibly personal nature ofhealthcare, will produce multiple 1.We believe fragmentation of the delivery system, combined with the irreduciblypersonal nature of healthcare, will produce multiple winning care delivery models—notthe winner-take-all dynamic typical of Big Tech platforms. Provider organizations such 2.On the economic trajectory, we project that utilization will go higher in the near termand lower in the long term, while costs will increase more than utilization in the near termand decrease more than utilization in the long term. However, the dynamics of higherutilization in early years will likely result in payment reforms, as even under the mostaggressive clinical AI displacement scenarios, cumulative NHE remain above baseline 3.Big Tech will be more successful in healthcare long term than it has been historically,driven by vertical integration strategies exemplified by Amazon Health AI and Google’s 4.Last-mile execution is the key differentiator for success—the companies, such asSprinter Health, that solve the final connection between AI capability and patient 5.Personality matters:Clinical AI that lacks a human-centered design ethos will failregardless of its technical sophistication. 6.Winners will understand that the US government cannot fund healthcare at historicallevelsand will build operating and financial models that recognize this fiscal reality.Hospital systems that reduce capital intensity—shifting from billion-dollar campus 7.The Chinese and Indian regulatory frameworks provide an advantage in adoptionspeed.US regulatory friction will result in slower near-term adoption rates. However,clinician supply constraints—with a projected shortage of up to 124,000 physicians by 8.The ultimate bottleneck to a healthier society is a broken healthcare financing systemin the US,where providers are not incentivized to provide preventive care and consumers’ Executive summary The current state of AI clinical capabilities is that AI performs extremely well oncomplex reasoning tasks yet breaks down under uncertainty with missing informationor changing context. This was documented extensively by Stanford’s ARISE researchand the broader Stanford HAI ecosystem. However, with each study highlighting That boundary is moving fast. AI capabilities are advancing at an accelerating rate. Instructured clinical tasks—diagnostic pattern recognition, treatment protocol matching,and drug interaction analysis—AI already supersedes human clinical capabilities.Within the near future, that superiority will extend to broader, less defined clinical This report examines five interconnected dimensions of AI in care delivery: AI adoption in healthcare is pervasive and accelerating.81% of US physicians now use AI in their practices, more than doublethe rate just three years ago. Consumer engagement with AI-generated health information is similarly widespread, with 79% ofUS adults likely to search online for health answers and a majority finding AI-generated responses at least somewhat reliable. The care del