Sixty-two percent of hospitals running the Epic electronic health record have deployed an ambient AI scribe, the software that listens to a patient visit and drafts the note. A for-profit hospital’s odds of having one: 28.8%. A hospital sitting in the bottom quartile for operating margin: 58.0%, next to 68.7% for one in the top range AJMC, 2026. Adoption isn’t the headline anymore. Distribution is, and it’s breaking along the fault line you’d expect: money.
The Problem
Trade press has spent 2026 treating ambient AI as a solved category, and three vendors, DAX Copilot, Abridge, and ThinkAndor, now account for more than 80% of every ambient documentation deployment inside Epic-using hospitals [AJMC, 2026]. The actual data, a peer-reviewed study of 2,784 Epic hospitals published in the American Journal of Managed Care, tells a narrower story. Of those hospitals, 1,744 (62.6%) had adopted an ambient AI tool as of mid-2025. Adoption rose with staffing-adjusted workload, hospitals in the busiest quartile sat at 73.1% versus 61.7% in the least busy, which tracks with the sales pitch [AJMC, 2026]. But adoption also rose with operating margin, urban location, and nonprofit ownership, and fell off a cliff at government-owned and for-profit facilities (45.0% and 28.8% against nonprofit’s 70.2%), and in the Midwest relative to the South (54.9% vs 69.5%) [AJMC, 2026]. None of those second variables have anything to do with documentation burden. They have to do with who can afford the tool. The compositional gap matches: nonprofit hospitals make up 87.4% of adopters versus 61.4% of non-adopters, and for-profit hospitals are 22.9% of non-adopters but only 5.5% of adopters [AJMC, 2026].
The federal government ran the same test on a different category of clinical AI and got the same answer. Predictive AI models integrated with the EHR, the tools that flag readmission risk or early deterioration, rose from 66% of hospitals in 2023 to 71% in 2024 ONC/ASTP, 2025. Small hospitals under 100 beds sat at 59% in 2024. Large hospitals over 400 beds sat at 96% [ONC/ASTP, 2025]. Two AI categories, two research teams, one identical shape: the hospitals with the least slack get the least AI. Those are the same hospitals carrying the pharmacist and clinician staffing gaps that pushed telepharmacy into the mainstream in the first place, the small, rural, independent, and safety-net facilities already leaning on remote coverage to hold a schedule together. A rushed note written at 2 a.m. by a physician carrying twice the patient load of a peer at a better-staffed hospital is exactly the note a verifying pharmacist has to untangle, no ambient scribe, no time saved, same clinical risk.
The Insight
Here’s the claim a cautious content team edits out before publishing: adoption inequality isn’t the whole problem, because even where ambient AI does land, nobody has proven it’s working.
A commentary in the Journal of Medical Internet Research, co-authored by University of Colorado hospitalists, makes the case directly. Two-thirds of hospitalists already use AI platforms clinically, mostly large language model tools, largely without any health system training or governance behind them JMIR, 2026. The authors compare this moment to the early EHR rollout, a technology adopted widely and credited with real gains, that also increased clinician workload and burnout precisely where implementation was rushed [JMIR, 2026]. A randomized trial they cite found AI assistance alone did not improve physicians’ diagnostic reasoning, while the AI running unsupervised outperformed the clinicians using it, a result that says less about the model’s ceiling than about how badly most deployments integrate it [JMIR, 2026]. The variable was never the model. It was whether anyone designed the workflow around it. The authors frame the real test as whether AI advances the quintuple aim, clinical outcomes, health equity, cost, and clinician experience together, not clinician time saved in isolation [JMIR, 2026].
“The hospitals buying the productivity story fastest are the ones that least need it, and the hospitals that need it most can’t get past the sales call.”
Put the two findings together and the industry narrative collapses. Adoption is concentrated in hospitals that already run the healthiest margins, and even inside that fortunate group, use rates are outrunning any evidence the tool improves an outcome that matters. Ambient AI stops being a burnout fix and becomes something closer to a wealth amplifier: the systems with room to spend get first access to a tool nobody has fully proven works, while the systems running closest to the edge wait for a rollout that keeps getting deprioritized behind capital projects with a clearer return.
Real-World Application
The AJMC data does hand operators one genuine lever, buried in a variable most people would expect to cut the other way. Disproportionate Share Hospitals, the safety-net facilities serving the highest share of low-income patients, actually out-adopted their non-DSH peers, 64.3% to 57.8% [AJMC, 2026]. DSH status carries a dedicated federal payment stream, and it’s one of the only variables in the study where a resource-constrained hospital type beat a wealthier baseline. Targeted funding closes the gap that market forces won’t.
| Hospital profile | Adjusted ambient AI adoption | Lever available |
|---|---|---|
| Large, nonprofit, metro, high-margin | 67-73% | None needed; already first in line |
| DSH / safety-net, mixed margin | 64.3% | Tie AI procurement to existing DSH or 340B revenue instead of a separate capital ask |
| Small, independent, rural | 54-59% | Join or negotiate through a regional consortium or GPO for system-level pricing |
| For-profit, non-metro, Midwest | 28.8-54.9% | Build the case on operating savings, not a burnout narrative that won’t move a for-profit board |
Vendor concentration cuts the same way: DAX Copilot, Abridge, and ThinkAndor already hold more than 80% of the Epic-connected ambient AI market [AJMC, 2026], so a single independent hospital has almost no leverage negotiating against any of the three. A regional consortium of ten small hospitals buying together does, the same math that already justifies group purchasing for drug spend. The AJMC authors reach the same policy conclusion from the outside: they point to HITECH, the 2009 federal program that used financial incentives, not market pressure, to close the original EHR adoption gap, as the template still available for ambient AI [AJMC, 2026]. Their specific suggestions, shared services, scalable pricing, and technical assistance for resource-constrained hospitals, aren’t exotic. They’re the difference between hoping a market with no incentive to reach the bottom quartile eventually gets there, and building the incentive on purpose.
The Bottom Line
The trade coverage of ambient AI in 2026 reads like a technology success story: 62.6% adoption, a clear market leader in three vendors, real productivity gains reported at the hospitals running it. All of that is true, and none of it describes what’s happening to the hospitals outside that number. The AJMC and ONC data, run independently, on different AI categories, in different years, describe the same divide: money and system affiliation predict adoption better than clinical need does. The predictive-AI version of this gap, 59% at small hospitals against 96% at large ones, has already had a year longer to calcify [ONC/ASTP, 2025; AJMC, 2026].
Watch what ONC publishes when it covers 2025 hospital data later this year. If the small-hospital figure hasn’t moved meaningfully past 59%, this stops looking like a rollout lag and starts looking like a permanent tier, the same two-track system EHR adoption produced after HITECH, minus a federal incentive program to force the laggards forward. The DSH finding already proves a targeted funding stream moves the number faster than good intentions do, and a regional consortium can borrow the same leverage without an act of Congress behind it. The hospitals that most need documentation relief are the ones a market-driven rollout was never going to reach first, unless somebody targets the money on purpose.