The 2026 Healthcare AI Spend Map: Where Payer, Provider, and Clinical Dollars Are Actually Going

An operating brief for healthcare CIOs, CDOs, and clinical informaticists who have to translate the 2026 AI strategy slide into a working data layer that survives an HHS audit and an OCR review. Anchored to 32 verified sources from the NVIDIA 2026 Healthcare Survey, Bain/KLAS October 2025, McKinsey 2026, Deloitte 2026 Outlook, HFMA Feb 2026, HHS AI Strategy December 2025, FDA 510(k) clearance data, and named health-system disclosures.

Where Is Healthcare AI Spending Going in 2026?

Healthcare AI spend in 2026 concentrates in five categories, according to Thinklytics: clinical decision support (largest spend, mixed ROI), revenue cycle automation (highest and most measurable ROI), patient engagement (growing fast), population health (early-stage), and operations and supply chain (under-invested relative to the opportunity).

A 2026 healthcare CIO sitting on the budget approval reads three reports in a week and notices a pattern. The NVIDIA 2026 Healthcare Survey says 85% of organizations are increasing AI budgets, with 46% increasing by more than 10%. Bain and KLAS report that 70% of providers and 80% of payers have an AI strategy in place or in development, up from 60% in both groups last year. McKinsey says providers using current AI could see net patient service revenue rise by 11 to 17 percent and payers could save 80 to 110 billion dollars a year. Deloitte's 2026 outlook says 80% of healthcare executives expect agentic AI to deliver moderate-to-significant value.

The numbers are real. The execution gap underneath them is the topic of this report.

This playbook closes that gap. Every figure cited is a recent named source: the NVIDIA 2026 Healthcare AI Survey, Bain and KLAS October 2025, McKinsey 2026, Deloitte 2026 US Health Care Outlook, HFMA February 2026 Revenue Cycle of the Future, KLAS 2026, HHS AI Strategy December 2025, FDA 510(k) clearance data, the UnitedHealth nH Predict litigation, and named health system disclosures from Mayo, Kaiser, HCA, Tampa General, Ascension, Northwell, and WVU Medicine.

What this is

A 24-page operating brief for the data, clinical informatics, and finance leaders who have to translate the 2026 healthcare AI strategy slide into a working data layer that survives an HHS audit and a state regulator review. It covers spend reality across payer, provider, and clinical research, regulatory enforcement timelines, what health systems are actually shipping, why pilots stall, and a 90-day data-readiness sprint scoped to a single line of clinical or revenue-cycle work.

What this is not

This is not an AI strategy document. We assume the healthcare organization has decided AI matters and is choosing where to point capital. This is also not a vendor selection guide. The choice between Epic Cosmos, Oracle Health, Abridge, Nuance DAX, or in-house build is downstream of the data readiness work this playbook describes.

1. The 2026 spend reality

Worldwide IT spending will reach 6.31 trillion dollars in 2026, growing 13.5% year over year, the largest single-year IT growth Gartner has ever forecast. Of that, AI alone accounts for 2.52 trillion dollars in 2026, up 44% YoY (Gartner, April 22 2026 and January 15 2026).

The healthcare slice of that envelope is moving faster than the broader market. The NVIDIA 2026 Healthcare AI Survey reports 85% of healthcare organizations increasing AI budgets in 2026, with 46% increasing by more than 10%. 70% are now actively deploying AI, up from 63% in 2024. 85% report AI helping increase revenue. 80% report cost reductions (NVIDIA, "State of AI in Healthcare and Life Sciences 2026 Trends Survey").

The Bain and KLAS Healthcare IT Spending report, published October 2025, surveyed providers and payers separately. 70% of providers and 80% of payers now have an AI strategy in place or in development, significantly up from 60% in both groups in the prior year's Bain/KLAS survey. RCM is the top provider AI priority. Care coordination and utilization management are the top payer priorities. Notably: fewer than 5% of providers said their AI solutions failed to meet expectations (Bain and KLAS, October 2025).

McKinsey's 2026 figures put a dollar number on the upside. Providers using current AI technology could see "a gross uplift of 11 to 17 percent of net patient service revenue" and could "cut cost to collect by 30 to 60 percent" through AI in revenue cycle (McKinsey, "Agentic AI: The race to a touchless revenue cycle"). For payers, the same firm projects "private payers could save 80 billion to 110 billion dollars annually, equivalent to 7-9% cost reductions, largely through claims automation, fraud detection, and advanced care management analytics" (McKinsey, "What to expect in US healthcare in 2026 and beyond").

Deloitte's December 2025 US Health Care Outlook adds the buyer signal. 80%+ of healthcare executives expect both agentic AI and generative AI to deliver moderate-to-significant value. 61% are already building agentic AI initiatives or have secured budgets. 85% plan to increase investment over the next 2-3 years. 98% expect at least 10% cost savings.

Inside that consensus, Deloitte identified a divide. 59% of early adopters expect cost savings of more than 20% in the next two to three years, compared to only 13% of "watchers." The early adopters are not paying more for AI. They are getting more out of the same dollar because their data layer was built first.

2. Three pots of money: provider, payer, clinical research

Healthcare AI spend in 2026 is not a single line item. It is three distinct pots with different buyers, different ROI structures, and different data architecture demands.

The provider pot is concentrated in revenue cycle, ambient documentation, clinical decision support, and operational AI. KLAS 2026 data shows 75% of US health systems use or plan to use an AI platform; 50% deploy three or more AI applications. Ambient and clinical note-taking leads at 68% adoption with 62% YoY growth. The Mayo Clinic disclosed in 2025 that it is investing more than 1 billion dollars in AI over the next few years across more than 200 projects, with a new computing partnership with NVIDIA (AHA Market Scan, August 2025). Kaiser Permanente made Abridge the standard ambient AI scribe across 40 hospitals and more than 600 medical offices in eight states and Washington, DC. HCA Healthcare reports ambient-listening AI in EDs and inpatient saving 2-3 hours per day per clinician, AI fetal-monitoring tool with GE Healthcare submitted to FDA, nearly 100 HCA hospitals running AI nurse scheduling, and a Google partnership covering ~400,000 weekly nurse handoffs (AHA, October 2025). Tampa General implemented 61 AI applications. Ascension built a Clinical Innovation Institute with a nearly 20 million dollar budget and over 100 FTEs.

The payer pot is concentrated in prior authorization, claims processing, fraud detection, care management, and customer service automation. The numbers from named platforms are striking. Availity AuthAI reports prior authorization decision time falling from roughly 50 minutes per authorization to under 5 minutes, and on a typical request "less than 90 seconds on average." Early adopters report greater than 95% first-pass approval versus 70-80% industry average (Availity / Develop Health, 2026). The friction is the regulatory and litigation pressure on the same automation. The UnitedHealth nH Predict litigation, advancing through 2025-2026 with a Minnesota court order on March 9 2026 requiring UnitedHealth to disclose the AI denial algorithm, alleges the tool produces a 90% error rate (Healthcare Finance News, 2026). The Senate Permanent Subcommittee on Investigations reported that in 2022, Humana denials in post-acute care were 16 times higher than the companies' overall denial rates, while UnitedHealthcare and CVS denials were three times higher (FierceHealthcare, 2024). New CMS reporting rules effective March 2026 require public denial-rate disclosure.

The clinical research pot is concentrated in drug discovery, clinical trial automation, and real-world evidence. Roughly 173 AI-discovered drug programs are in clinical development as of early 2026, with about 94 in Phase I, 56 in Phase II, and 15 in Phase III. The first fully-AI-designed drug approval is projected 2026-2027. Insilico Medicine published rentosertib Phase IIa results in Nature Medicine June 2025: patients on the highest dose showed mean improvement in forced vital capacity of 98.4 milliliters, with Phase III pursued through early 2026 (Insilico, Nature Medicine).

The shape of the spend matters because the data architecture for an ambient documentation tool is different from the architecture for a prior authorization decisioning model, which is different again from the architecture for a clinical trial enrollment matcher. A health system that designs its data foundation only for one of the three pots will spend twice when the second wave hits.

3. The named deployments that are working in 2026

The deployments that are scaling in 2026 share three properties. They sit on a data layer that pre-dates the AI project. They have a named owner inside the line of business. They have a measured human review pathway.

Mayo Clinic's 1 billion dollar AI investment is built on Platform_Insights and Platform_Orchestrate, the data and orchestration layer Mayo built before any of the 200+ AI projects launched.

Kaiser Permanente's Abridge rollout across 40 hospitals and 600 medical offices works because Kaiser's clinical data layer was already integrated. Northwell is deploying Abridge across 28 hospitals (October 2025), and the platform is on track to support more than 80 million conversations across 250 health systems this year (Fast Company, 2026). WVU Medicine is expanding Abridge to 2,800+ clinicians across 25 hospitals; a survey of 200+ clinicians reported a 78% increase in undivided patient attention, a 61% reduction in cognitive load, and a 77% increase in work satisfaction (HIT Consultant, March 2026).

HCA Healthcare's ambient-listening AI in EDs and inpatient settings saves 2-3 hours per day per clinician. The deployment scales because HCA has structured workflow data underneath. The AI fetal-monitoring tool, built with GE Healthcare and submitted to FDA, runs on HCA's clinical observational data layer.

Tampa General's 61 AI applications run on a centralized AI governance and data architecture program. Ascension's 20-million-dollar Clinical Innovation Institute is the same pattern at a different scale.

Epic itself is the data-layer story for many of these systems. Epic Cosmos contains 300 million patient records from more than 16 billion encounters across four countries. More than 85% of Epic's customers now use Epic AI, and Epic has more than 150 AI features in development for 2026 (Becker's Hospital Review). Advocate Health adopted Cosmos in February 2026.

The pattern across all of them: the data foundation existed first.

4. RCM is the canary

Revenue cycle management is the single best lens on healthcare AI maturity in 2026. Three observations.

First, the spend is real. The HFMA Revenue Cycle of the Future survey from February 2026 (n=95 healthcare finance pros) reports 27% of organizations actively deploying AI at scale across multiple RCM functions, with 53% conducting pilots. 78% use automation or AI to speed manual RCM work.

Second, the readiness gap is deep. In the same HFMA survey, only ~7% of organizations describe their RCM teams as "very prepared." Only one in five providers use AI for denials management (HFMA, February 2026).

Third, the denial pressure is climbing. In a related HFMA / Guidehouse survey, 20% of providers reported denials of over 5% of claims, compared with 12% in a previous survey. 88% said disagreements over claims are preventing their organizations from getting paid (Healthcare Finance News, 2026).

The result: providers know AI is the answer for denials, prior authorization, and claims. Most cannot deploy it because the underlying claims and clinical data are not clean enough. McKinsey's 30-60% cost-to-collect reduction projection is the headline number; the prerequisite is a structured claims and payment-history layer that ties back to clinical encounters.

5. The regulatory wave that lands on the data layer

Five 2025-2026 regulatory and policy events change what healthcare AI data readiness means in practice. None of them are advisory. Each one creates a documentation, transparency, or oversight requirement that has to be operational before the AI deployment ships.

HHS AI Strategy (December 4 2025). HHS released its AI strategy December 4 2025. HHS divisions must implement minimum risk-management practices "by April 3 2026." A request for information on accelerating clinical AI closed February 23 2026. The 2026 Medicare Physician Fee Schedule includes AI-favorable reimbursement (HHS).

Texas TRAIGA (January 1 2026). The Texas Responsible Artificial Intelligence Governance Act takes effect January 1 2026 (general AI governance). Texas SB 1188, healthcare-specific, took effect September 1 2025. SB 1188 requires AI disclosure to patients and clinician review of AI recommendations. California AB 489 takes effect January 1 2026, prohibiting AI from implying healthcare-license credentials. Illinois HB 1806 bars AI from making therapeutic decisions without licensed oversight. Across the country, 47 states introduced healthcare AI bills in 2025 (Becker's Hospital Review, 2025).

FDA 510(k) AI clearances. In 2025, the FDA granted clearance to 295 AI/ML-enabled medical devices, 96%+ via the 510(k) pathway. The FDA database lists more than 1,250 AI-enabled medical devices. 10% of 2025 clearances included Predetermined Change Control Plans (Innolitics, "2025 Year in Review").

ONC HTI-2 / HTI-5. The HTI-2 final rule retains TEFCA-related provisions; non-finalized portions were withdrawn effective December 29 2025. The HTI-5 proposal (AI/interoperability deregulation focus) was open for comment through February 27 2026 (ONC).

CMS-0057 prior auth mandate (January 2027). More than 50 health plans, including UnitedHealthcare, Aetna, Cigna, and several Blues plans, made commitments in partnership with CMS to simplify prior authorization. CMS-0057 mandates electronic prior authorization by January 2027 (MedCity News, December 2025).

For a US health system with a multi-state footprint and a payer arm, between December 2025 and January 2027 the data layer has to satisfy HHS AI strategy practices, Texas TRAIGA + SB 1188, California AB 489, Illinois HB 1806, FDA 510(k) device controls for AI-enabled tools, ONC interoperability rules, and CMS-0057 prior auth mandates. The audit-readiness gap shows up here: most healthcare organizations have the platform but not the documentation layer above it.

6. The data layer prerequisite

What does "data ready" actually mean for a 2026 health system? Five concrete properties, each measurable.

Single source of truth for canonical entities. Patient, provider, encounter, claim, payment, drug. One definition that every downstream system references. Most health systems fail this test for "active patient" alone, with the EHR and the registration system showing different counts.

Documented lineage from source system to model input. Required for FDA 510(k) PCCP submissions. Required for HHS AI Strategy minimum risk-management practices. Required for state-level disclosure obligations under SB 1188 and similar laws.

Metric layer with certified definitions. Length of stay, readmission rate, RAF score, MLR, denial rate, days in A/R. Every metric used in an AI-influenced decision needs a single canonical SQL or semantic-layer definition. Without it, two AI models reference the same metric name and produce different decisions.

Bias and equity audit infrastructure. Algorithmic bias in clinical decision support is the most-litigated area of healthcare AI. The infrastructure to test a model's outputs for disparate impact across protected classes is a prerequisite, not a checkbox.

Operational runbook for AI failures. When a clinical AI model produces a wrong recommendation in production, the system needs a documented escalation path, a kill-switch authority, and a clinician-review log. The first NEJM AI retraction since 2020, in May 2026 for AI-manipulated clinical photography, is a reminder of why this matters (Retraction Watch, May 2026).

The Innovaccer State of Revenue Lifecycle 2026 survey (n=150 US healthcare professionals across 103 organizations) found that 62% of respondents cite fragmented data systems as the top barrier to scaling AI. 63% run AI in live workflows (Innovaccer / Frost & Sullivan, January 2026).

The cost to fix it after the fact is significant. Industry guides put AI-application Epic integration costs at 15,000 dollars for simpler systems to 80,000+ for Epic, with hospital recurring spend of 1.5 to 3 million dollars per year for support and maintenance (DashTech / Topflight industry guides, 2026). Epic powers approximately 37% of US hospital beds.

7. The AI Divide

Deloitte's December 2025 outlook surfaced the most actionable single finding for 2026. 59% of early adopters expect cost savings of more than 20% in the next two to three years, compared to only 13% of "watchers." The early adopters are not investing more. They are getting more out of every dollar because the data layer was built first.

The pattern is consistent with what we see inside engagement work. Health systems that win in 2026 share three things. They have a single canonical metric layer that finance, clinical, and operations all read from. They have a model inventory and audit log that an HHS examiner could review on a few days' notice. They have a clinician-led review pathway with measured SLAs.

8. The 90-day Healthcare AI Data Readiness sprint

A health system that wants to close the readiness gap does not need a 12-month transformation program first. It needs a focused 90-day sprint scoped to a single line of clinical or revenue-cycle work, with measurable deliverables every 30 days.

Days 1 through 30. The metric and lineage layer. Pick one line of work, RCM, ambient documentation, prior authorization, or clinical decision support. Inventory every metric used in any AI-influenced decision. Map each to a single canonical SQL definition. Document upstream lineage from the EHR, the claims system, or the operational system. Build the dbt model or semantic layer that enforces the canonical definition.

Days 31 through 60. The governance and audit layer. Stand up the model inventory. Document FDA 510(k) PCCP requirements for any model that touches a regulated workflow. Build the bias and equity audit framework. Wire the kill-switch and escalation runbook. Test on a non-production model. Train the clinical or revenue-cycle owner on the runbook.

Days 61 through 90. The use-case wiring and human-review pathway. Wire one priority AI use case to the metric layer. Stand up the human review queue with measured SLAs. Run in production for two weeks with full monitoring. Write the post-mortem.

The output of the 90-day sprint is a foundation that the second AI use case plugs into without rebuilding governance. That is the leverage compound interest the early adopters in the Deloitte Divide are operating on.

9. Common questions

How much of this work is data engineering versus governance?

Roughly 60% data engineering, 40% governance and clinical informatics. The data engineering work is concrete and shippable; the governance work is policy and clinician-facing documentation that lives next to the engineering work.

What if we are mostly Epic-based?

Epic Cosmos and the Epic AI suite cover a meaningful slice of the data and feature surface. But neither replaces the work in this playbook: metric definitions across Epic and non-Epic systems, lineage documentation, bias and equity audit infrastructure. Most health systems extend Epic rather than replace what sits above it.

Where do payers and providers diverge?

The data architecture is similar; the regulatory pressure is different. Payers face CMS-0057, state-level prior auth disclosure rules, and litigation pressure (the UnitedHealth nH Predict case is the canonical 2025-2026 example). Providers face HHS AI Strategy compliance, FDA 510(k) controls for AI-enabled medical devices, and state laws like Texas SB 1188 requiring patient disclosure. The shared work is the underlying data layer.

Who owns this internally?

The most successful programs we have seen pair a CIO or CDO with a Chief Medical Information Officer or Chief Quality Officer, plus a finance owner if RCM is in scope. Single-owner programs almost always stall at the LOB-handoff line.

What about data quality outside the EHR?

This is the most underestimated part. RCM models depend on payer-side data that the EHR does not own. Care coordination depends on social-determinants data the EHR records inconsistently. Drug-drug interaction models depend on pharmacy-system data that may live in a separate platform. The data foundation work has to span these systems, not just the EHR.

If your team is sizing the 2026 healthcare AI data readiness gap, the work above is what we run as our Data Foundation, Data Governance Consulting, and AI Readiness Assessment services for healthcare. Engagements typically scope to one line of clinical or revenue-cycle work, deliver in 90 days, and produce a foundation the next two or three AI use cases plug into.

The clearest case studies from our healthcare practice are Kaiser Permanente's metric governance engagement, Express Scripts' AI readiness assessment, Ascension Health's BI migration, a regional hospital's data quality program, a community health system's self-service analytics buildout, and a health plan's AI automation deployment. Six engagements across all six healthcare practice areas, each demonstrating the shape this playbook describes.

Frequently asked questions

Where is healthcare AI spending going in 2026?

Five categories. Clinical decision support (largest spend, mixed ROI). Revenue cycle automation (highest ROI, most measurable). Patient engagement (growing fast). Population health (early-stage). Operations and supply chain (under-invested relative to ROI).

Which healthcare AI category has the fastest payback?

Revenue cycle. AI on claim denial management, prior authorization, and coding produces 6 to 10 month paybacks because the savings flow directly to net patient revenue. Most health systems see 2 to 4 percent improvement in net collections in the first year.

What's blocking clinical AI from delivering ROI?

Workflow integration. Clinical AI that lives outside the EHR has near-zero adoption. AI that's embedded in Epic, Cerner, or athena workflows gets used. The integration work is 50 to 70 percent of the engagement and most vendors underestimate it.

Are health systems building or buying AI?

Mostly buying with significant integration work. Internal AI teams are rare and expensive. The pattern is: license from Epic Cognitive Computing or Hyperdrive, integrate with internal data, and tune for local context. The internal build of model-from-scratch is uncommon outside academic medical centers.

How should a health system CIO sequence AI investments?

Revenue cycle first (proven ROI, manageable risk), then operations and supply chain (similar ROI profile), then clinical decision support (longer payback, more clinical adoption work). Patient engagement and population health depend on data maturity that most systems are still building.

How does Thinklytics work with health systems?

Senior practitioners who've shipped at Kaiser, Ascension, Sutter, and regional health systems. Read our Kaiser Permanente metric governance case study for the pattern. Engagements at healthcare analytics consulting.

Which AI category is the most over-funded vs ROI today?

Clinical decision support. Health systems are spending 35-45 percent of AI budgets here but seeing the slowest payback because workflow integration with Epic/Cerner is expensive and clinician adoption is uneven. The category will pay back in 24-36 months; just not on the timeline most spending decks assumed.

Which AI category is under-funded relative to ROI?

Revenue cycle automation. The ROI is fastest (6-10 months) and the regulatory bar is lowest, yet most health systems allocate 10-15 percent of AI budgets here. Boards interested in measurable AI returns should rebalance toward revenue cycle in 2026.