How Healthcare AI Adoption Jumped from 3% to 22% in One Year

The planning baseline for using AI in healthcare changed quickly: domain-specific AI adoption sat near 3% in 2023, remained near 3% in 2024, and then reached 22% in 2025, according to Menlo Ventures’ survey of more than 700 executives conducted in August and September 2025.[1] That is the kind of curve that turns an innovation update into a budget question. A market that looked stuck for two years suddenly moved sevenfold in one.

The number is also easy to misuse. Menlo’s 22% figure measures organizations using domain-specific AI tools, not every possible AI capability embedded somewhere in a hospital’s technology stack.[1] It should not be blended with a hospital EHR predictive-model statistic, a physician survey, or a vendor count and presented as one giant adoption story. For 2026 planning, the useful question is narrower and more operational: which organizations have moved from experimentation into deployed, healthcare-specific AI tools, and where is capital following that movement?

On that narrower question, healthcare is no longer the sector that can be waved away as too slow to matter. Menlo reports that healthcare’s 22% domain-specific AI adoption rate was 2.2 times the broader economy’s 9% rate.[1] The old assumption that hospitals and adjacent healthcare organizations will trail every enterprise software wave now needs to be retired, or at least heavily qualified.

What the 22% figure does and does not say

Menlo’s adoption figure is valuable because it focuses on domain-specific AI tools: products built for healthcare workflows rather than general-purpose AI use. It is also a venture-backed market view, based on executive survey responses and spending analysis, so it is best read as a signal of organized buyer behavior among AI-engaged institutions rather than as a census of every provider organization in the country.[1]

That distinction matters because another legitimate number is much higher. ONC/ASTP reported that 71% of U.S. non-federal acute care hospitals used predictive AI integrated with EHRs in 2024, up from 66% in 2023, based on a survey of 2,080 hospitals with weighted results and response rates of 51% in 2024 and 58% in 2023.[2] That 71% figure is not evidence that 71% of hospitals have adopted the same category Menlo is measuring. It is evidence that predictive AI inside EHR environments is already widespread across acute care hospitals.

Both findings can be true. A hospital may use predictive AI integrated into its EHR while not yet having adopted newer domain-specific AI applications for documentation, coding, authorization, or patient engagement. Conversely, an organization may be buying AI-native administrative tools that do not fit neatly into the EHR predictive-AI frame. Treating these measures as interchangeable would inflate confidence exactly where executives need precision.

For readers who need a broader statistical base beyond this adoption-and-spending view, the full AI in healthcare statistics compendium is the better place to compare market size, accuracy, ROI, and risk indicators across categories.

Health systems are pulling the market forward

The segment map is where the adoption curve becomes a strategy problem. Health systems led the market at 27% adoption in Menlo’s data, followed by outpatient and post-acute organizations at 18% and payers at 14%; life sciences was described as earlier but accelerating.[1] The spread is not just a curiosity. It tells executives where repeatable procurement, governance, workflow redesign, and vendor relationships are forming first.

The spending distribution makes the same point more forcefully. Total healthcare AI spending reached $1.4 billion in 2025, nearly triple the 2024 level, and health systems accounted for 75% of that spending.[1] In other words, the current wave is not a generic healthcare-sector phenomenon evenly distributed across providers, payers, life sciences, and services companies. It is being disproportionately financed by health systems.

That matters for vendor shortlists. A company that wins enterprise health system buyers in 2026 may gain more than revenue; it may gain the reference architecture for how AI is evaluated, secured, integrated, monitored, and renewed. Menlo also reported that 85% of generative AI spending went to startups and that eight healthcare AI unicorns were created in 2025, more than any other vertical AI segment.[1] For a market long dominated by EHR incumbents and large services firms, that is not a side note.

The startup share should not be read as a permanent outcome. It is a 2025 spending signal, not a guarantee that AI-native companies will keep the same advantage once incumbents bundle features, renegotiate enterprise contracts, or close integration gaps. The startup-versus-incumbent question deserves its own treatment; readers evaluating that competitive layer can go deeper in Startups vs. Incumbents in Medical AI and the broader AI healthcare stack mapping.

Where the money is actually going

The 2025 spending pattern is less mysterious when the categories are named. Ambient clinical documentation attracted $600 million, coding and billing drew $450 million, prior authorization grew 10 times year over year, and patient engagement grew 20 times year over year, according to Menlo.[1] These are not abstract AI ambitions. They sit close to labor pressure, revenue cycle friction, patient access, and clinician documentation burden.

Administrative automation has an obvious gravitational pull. Menlo frames the opportunity against $740 billion in annual U.S. healthcare administration spending and $63 billion in healthcare IT spending, arguing that AI is converting services dollars into software dollars.[1] That framing is useful, but it should not be allowed to do too much work. A large administrative cost pool does not automatically mean savings will be captured by buyers, or that savings will appear inside the budget line where a CFO expects them.

A documentation tool may reduce after-hours charting but require new review policies. A coding product may improve throughput but shift work toward exception handling. A prior authorization system may reduce staff burden in one queue while creating new audit and payer-relations questions elsewhere. The spending categories are credible because they attach to real operational pain. Their financial impact still depends on implementation, measurement, and contract design.

For readers focused specifically on cost and workflow performance, the separate healthcare administration AI benchmark analysis is the better place to test claims about cost reduction, accuracy, and adoption.

Procurement speed is an adoption signal, not just a purchasing metric

One of the more important 2025 signals is not a use case at all. Menlo reports that health system AI procurement cycles shortened from 8.0 months to 6.6 months.[1] A 1.4-month reduction does not sound dramatic until it is placed inside the normal reality of health system purchasing, where legal review, security assessment, clinical leadership, IT integration, compliance, finance, and contracting can each slow a decision.

Shorter procurement cycles suggest that at least some organizations are building repeatable pathways for AI review. That may include standard security questionnaires, model-risk intake processes, approved contracting language, clearer ownership between clinical and IT leaders, or board-level permission to move faster in defined categories. Menlo’s data does not prove which mechanism caused the compression, so the safer conclusion is that organizational readiness is improving among surveyed health systems.

This is where adoption becomes institutional. A pilot can survive on a champion. Enterprise adoption requires a queue, a budget owner, a vendor review path, a governance forum, implementation staff, and a way to decide what happens when the tool performs below expectation. The procurement signal deserves attention because it points to those muscles developing, not because faster buying is automatically better buying.

Governance is especially important for generative AI, where employee experimentation can outrun formal controls. The adoption numbers in this article should be read alongside the separate analysis of generative AI governance and shadow AI in healthcare, because enterprise spending and informal use do not always move through the same channels.

The regulatory backdrop is real, but it is not the whole market

The FDA’s list of AI-enabled medical devices included more than 1,450 authorized devices as of 2026, a reminder that healthcare AI is not confined to enterprise workflow software.[3] Imaging, diagnostics, monitoring, and device-linked software have their own adoption dynamics, evidence requirements, and regulatory pathways.

That context should not blur the center of this adoption story. Menlo’s 2025 inflection is largely about healthcare organizations buying domain-specific AI tools, with especially visible movement in health systems and administrative or workflow-adjacent categories.[1] FDA authorizations show that regulated AI-enabled products are part of the broader environment, but device clearance counts are not a proxy for enterprise AI adoption.

The 80% still untapped will not all look like late adopters

A 22% adoption rate means roughly 80% of the market remained untapped in Menlo’s 2025 framing.[1] It is tempting to treat that as a simple vendor pipeline: the next buyers are just waiting their turn. That is too neat. The organizations outside the first wave may have different economics, weaker data infrastructure, fewer informatics staff, less contracting leverage, or less ability to absorb implementation risk.

The ONC/ASTP hospital brief sharpens that concern. In its predictive-AI-in-EHR measure, small hospitals with fewer than 100 beds, rural hospitals, and independent hospitals lagged significantly behind larger, urban, and system-affiliated hospitals.[2] The precise magnitude of the broader AI divide still requires triangulation across datasets, but the direction is hard to ignore: the same organizations with fewer resources to evaluate AI risk are also less likely to have the institutional capacity that early adoption demands.

This is the part of the market where adoption charts can become dangerous. A large urban system may have an AI governance committee, legal support, cybersecurity staff, physician informatics leadership, analytics engineers, and procurement specialists who have already reviewed several vendors. A small rural hospital may be asked to evaluate model performance, patient safety, privacy, indemnification, workflow fit, and total cost with a much thinner bench.

The consequence is not only slower adoption. It can be worse adoption. A resource-constrained hospital may delay a useful tool because it cannot validate the vendor’s claims, or it may accept unfavorable terms because it lacks negotiating leverage. It may depend more heavily on whatever its EHR, group purchasing organization, regional partner, or parent system makes available. The divide is therefore not just about access to AI products; it is about access to evaluation capacity.

For 2026 strategy, that should change the rural partnership conversation. Large systems that already have AI governance infrastructure can treat it as a shared capability, not only an internal control function. State hospital associations, regional collaboratives, academic medical centers, and payers may all have roles to play, but the evidence here supports the need for capacity-building more strongly than it supports any single model for delivering it.

What executives should take into 2026 planning

The 2025 data justify a more serious AI planning posture. Healthcare moved from roughly 3% to 22% adoption of domain-specific AI tools in one year, health systems reached 27%, total spending hit $1.4 billion, and procurement cycles shortened from 8.0 to 6.6 months.[1] Those are enterprise adoption signals, not just conference-stage enthusiasm.

They do not justify pretending that all healthcare organizations are in the same position. Health systems are supplying most of the spending, startups captured most generative AI spend in the reported period, and the strongest growth categories cluster around documentation, revenue cycle, authorization, and engagement.[1] Meanwhile, ONC/ASTP’s hospital data shows that smaller, rural, and independent hospitals lag in predictive-AI adoption inside EHR environments, even before the newer domain-specific AI wave is fully considered.[2]

The board-level implication is straightforward. AI should no longer sit only in an innovation portfolio. It belongs in capital planning, clinical operations, revenue cycle strategy, cybersecurity review, vendor management, workforce planning, and partnership design. But the first planning document should separate the metrics: domain-specific AI adoption, predictive AI embedded in EHRs, generative AI experimentation, regulated AI-enabled devices, and clinician-level use are different signals.

Once those signals are separated, the 2026 agenda becomes clearer. Which AI categories deserve enterprise funding rather than departmental pilots? Which contracts give the organization leverage if performance, safety, or integration disappoints? Which workflows have enough operational ownership to absorb change? Which smaller hospitals in the network need shared evaluation support before they are asked to make AI decisions on their own?

Healthcare has crossed from pilot-era curiosity into a measurable enterprise adoption wave. The next question is not simply how fast the remaining market adopts. It is whether adoption capacity, governance capacity, and procurement leverage spread beyond the large urban systems that are currently best positioned to turn AI interest into durable operating capability.

References

1. 2025: The State of AI in Healthcare. Menlo Ventures. 2025.
2. Hospital Trends in the Use, Evaluation, and Governance of Predictive AI, 2023–2024. ONC/ASTP. 2024–2025.
3. Artificial Intelligence-Enabled Medical Devices. FDA. 2026.