The healthcare AI company landscape is not a monolith. A radiology AI startup clearing its tenth 510(k) and an ambient documentation company embedded in 200 hospital EHRs are both called "healthcare AI companies," but they operate under different regulatory frameworks, sell into different buyer relationships, and carry entirely different evidence burdens. Treating them as a single category makes it harder, not easier, to evaluate any individual company.
This overview organizes the landscape by primary application domain — the most useful axis for clinicians, procurement staff, and researchers trying to understand where a company actually sits. Within each domain, it notes the regulatory pathway most companies in that segment pursue, the evidence standard they face, and the structural factors that distinguish leading companies from earlier-stage players.
How This Landscape Is Organized
Company profiles on this site are grouped by primary focus area rather than by company size, funding stage, or market cap. That choice reflects how readers actually use the information: a radiologist evaluating AI-assisted chest CT tools needs to compare companies within that domain, not across the entire healthcare AI spectrum.
The domains covered here — medical imaging AI, ambient documentation, clinical decision support, genomics and diagnostics, and drug discovery AI — account for the substantial majority of FDA-cleared AI medical devices and deployed clinical AI tools as of mid-2026. Each has a distinct regulatory posture, evidence base, and deployment reality.
Medical Imaging AI
Imaging AI is the most densely regulated segment of the healthcare AI company landscape. Roughly three-quarters of all FDA-cleared AI/ML-enabled medical devices are in radiology, pathology, or other imaging-dependent specialties. That concentration reflects where the training data existed earliest — large PACS archives, digitized pathology slides, structured DICOM files — and where the regulatory pathway was clearest.
Companies in this segment typically pursue 510(k) clearance for specific detection or triage tasks: pulmonary nodule detection, intracranial hemorrhage flagging, mammography CAD, cardiac function measurement from echocardiography. The intended use statement is narrow by design — it defines exactly what the algorithm is cleared to do and, implicitly, what it is not.
Established Imaging AI Companies
Several companies have accumulated ten or more FDA clearances across imaging subspecialties. Aidoc, Annalise.ai, Viz.ai, and Intelerad (which acquired Mednax Radiology Solutions) represent the tier of companies with broad multi-product portfolios rather than single-algorithm plays. Aidoc's platform approach — integrating multiple cleared algorithms into a single radiology workflow layer — is a structural bet that hospitals will prefer fewer vendor relationships over best-of-breed individual tools.
Viz.ai has pursued a different path, focusing on time-sensitive care coordination workflows: stroke, pulmonary embolism, aortic dissection. The clinical argument is that AI's value in these cases is not just detection accuracy but speed — getting the right specialist notified faster than a traditional read workflow. Their cleared indications reflect that positioning.
In digital pathology, Paige.ai holds a PMA authorization for its prostate cancer detection algorithm — a higher regulatory bar than 510(k) and one that required clinical trial data. PathAI and Proscia operate in the pathology AI space with different evidence postures: PathAI has published peer-reviewed validation studies; Proscia has focused more on workflow tooling for pathology labs.
Imaging AI: Evidence and Deployment Gaps
FDA clearance does not equal clinical deployment, and clinical deployment does not equal demonstrated outcome benefit. Most imaging AI clearances are based on retrospective performance studies — sensitivity and specificity on held-out test sets — rather than prospective RCTs measuring patient outcomes. The external validation gap is real: algorithms trained on data from large academic centers frequently show performance degradation when deployed at community hospitals with different scanner types, patient demographics, or imaging protocols.
Radiologist adoption remains uneven. Studies on AI-assisted chest X-ray reading have shown that radiologists do not uniformly defer to AI recommendations — some over-rely on AI flags, others dismiss them. The workflow integration question (how tightly the AI output is embedded in PACS versus surfaced as a separate alert) significantly affects actual use rates in practice.
Ambient Documentation and AI Scribes
Ambient AI documentation — tools that listen to clinical encounters and generate structured notes without physician typing — has seen faster deployment velocity than almost any other healthcare AI category. The regulatory path is different here: most ambient documentation tools are not classified as medical devices under current FDA guidance, which means they do not require 510(k) clearance. That lowers the barrier to market but also means there is no standardized pre-market evidence requirement.
Nuance (owned by Microsoft) and Suki are among the most widely deployed. Nuance DAX Copilot integrates with Epic and several other major EHRs and has published evidence on documentation time reduction. Abridge, which has a prominent partnership with UPMC, has taken a more academically oriented approach, publishing peer-reviewed studies on its ambient documentation accuracy.
The physician burnout framing — that ambient AI reduces documentation burden — has driven adoption faster than clinical evidence typically would. The practical risk is that errors in AI-generated notes carry forward into the medical record. Unlike a radiology AI that flags a finding a radiologist then reviews, an ambient scribe's output may be accepted with minimal review under time pressure.
Clinical Decision Support
Clinical decision support (CDS) AI covers a wide range of tools embedded in EHR workflows: sepsis prediction alerts, deterioration scores, medication interaction flagging, care gap identification, and risk stratification for conditions like acute kidney injury or readmission. The regulatory picture here is genuinely complicated.
Under the 21st Century Cures Act, certain CDS software is explicitly excluded from FDA device regulation — but the exclusion has conditions. If a CDS tool requires clinician review before action and is not intended to replace clinical judgment, it may fall outside FDA oversight. If it is intended to acquire, process, or analyze medical images or signals, or to support diagnosis, it likely falls within FDA's jurisdiction. The line is not always obvious in practice, and FDA has issued guidance attempting to clarify it.
Epic and Oracle Health (formerly Cerner) both embed proprietary AI models directly in their EHR platforms. Epic's deterioration index and sepsis prediction model are used at hundreds of health systems. These tools operate at scale that no standalone AI company has matched, but their evidence base is primarily retrospective, and external validation studies have produced mixed results — a well-documented finding in the literature on Epic's sepsis model specifically.
Standalone CDS AI companies include Sepsis Express (acquired), Dascena (which faced regulatory scrutiny), and Etiometry. Health Catalyst acquired several CDS-adjacent analytics companies and positions its Ignite platform as a population health AI layer. The acquisition activity in this segment reflects both the difficulty of standalone viability and the strategic value of CDS AI to larger health IT platforms.
Genomics and Molecular Diagnostics AI
AI applied to genomic data — variant classification, polygenic risk scoring, tumor mutational burden analysis, pharmacogenomics — occupies a distinct regulatory space. Many genomic AI tools are regulated as laboratory-developed tests (LDTs) under CLIA oversight rather than as medical devices, though FDA's authority over LDTs has been contested and is subject to ongoing regulatory evolution.
Tempus AI (which went public in 2024) has built a large clinical genomics database and offers AI-assisted tumor profiling and treatment matching. Guardant Health's AI-assisted liquid biopsy analysis is used for early cancer detection and treatment monitoring. Foundation Medicine (owned by Roche) uses AI to interpret comprehensive genomic profiling results.
The evidence standard in genomics AI is evolving. Companion diagnostic claims require PMA-level evidence; prognostic or informational genomic reports face a different bar. The clinical utility question — whether AI-assisted genomic interpretation changes outcomes versus standard interpretation — is an active area of research with few completed RCTs as of mid-2026.
Drug Discovery and Preclinical AI
Drug discovery AI companies occupy a different position in this landscape: their products are not medical devices, are not regulated by FDA as such, and are not deployed in clinical settings. They are included here because they are frequently grouped under the "healthcare AI" label and because their downstream outputs — drug candidates — eventually enter clinical trials and, if approved, clinical practice.
Recursion Pharmaceuticals, Exscientia (acquired by Recursion in 2024), Insilico Medicine, and Schrödinger represent different approaches: Recursion uses large-scale cellular imaging and phenotypic screening; Insilico uses generative AI for molecular design; Schrödinger uses physics-based simulation augmented by machine learning. The common thread is using AI to accelerate target identification, hit generation, or lead optimization.
As of Q2 2026, no drug discovered primarily through AI methods has received FDA approval, though several AI-assisted compounds are in Phase II and Phase III trials. The clinical evidence question for this segment will take years to resolve — the validation event is a regulatory approval, not a study publication.
Comparing Segments: Regulatory and Evidence Posture
| Segment | Primary Regulatory Path | Evidence Standard | Deployment Maturity |
|---|---|---|---|
| Medical Imaging AI | 510(k) / De Novo / PMA | Retrospective validation; some prospective studies | High — hundreds of cleared devices deployed |
| Ambient Documentation | Not regulated (most tools) | Vendor-disclosed; some peer-reviewed studies | High — widespread EHR integration |
| Clinical Decision Support | Varies (CDS exclusion or 510(k)) | Mixed — EHR-embedded tools have large retrospective bases | High — embedded in major EHR platforms |
| Genomics / Molecular Dx AI | LDT / PMA (companion Dx) | Variable — companion Dx require RCT-level evidence | Moderate — concentrated in oncology |
| Drug Discovery AI | Not applicable (preclinical) | No clinical evidence yet; Phase II/III trials ongoing | Low — no approved AI-primary drugs as of Q2 2026 |
Company Stage and Funding Considerations
The healthcare AI company landscape includes a mix of publicly traded companies, private companies at various funding stages, and subsidiaries of larger health IT or pharmaceutical firms. Stage matters for assessing profile stability: a company that has raised a Series B may look similar to one that has raised a Series D in terms of product maturity, but the financial runway and acquisition risk differ substantially.
- Publicly traded companies in healthcare AI include Tempus AI, Veeva Systems (which has AI-assisted clinical data tools), and Invacare (adjacent). Nuance is now a Microsoft subsidiary. Aidoc, Viz.ai, and Abridge remain private as of mid-2026.
- Acquisition activity has been significant: Recursion acquired Exscientia, Microsoft acquired Nuance, Google has invested in or acquired several health AI companies. Acquisitions affect profile continuity — a company's FDA clearances transfer with an acquisition, but product roadmaps and support commitments may change.
- Funding stage as a proxy for evidence is unreliable. Some well-funded companies have thin peer-reviewed evidence bases; some earlier-stage companies have published rigorous external validation studies. Funding stage and evidence quality are independent dimensions.
What This Landscape Does Not Cover
Several categories of companies are frequently labeled "healthcare AI" but are excluded from this site's profiles for specific reasons:
- General cloud or AI infrastructure providers (AWS, Google Cloud, Azure) that offer healthcare-specific APIs but do not have cleared clinical AI products are not profiled here. Their tools may underlie products from companies that are profiled, but the infrastructure layer itself is not a clinical AI company for these purposes.
- Health IT companies where AI is a minor or recent feature addition — scheduling optimization, patient portal chatbots without clinical function — are excluded. The threshold is whether AI is a primary clinical product, not whether the company uses AI somewhere.
- Research-stage startups with no cleared or clinically deployed product are excluded, even if they have published interesting research. The profile format requires a real-world deployment anchor.
- Generative AI tools applied to clinical tasks — LLM-based diagnostic support, AI-generated clinical summaries — are tracked separately in the Generative AI in Medicine Watch section, because none currently hold FDA authorization for those specific uses and the regulatory status requires distinct treatment.
How Profiles Are Updated
Company profiles on this site are updated when material changes occur: a new FDA clearance, a significant acquisition, a published safety event, a major funding round, or a product withdrawal. Profiles are not refreshed on a fixed editorial calendar — the update trigger is a verifiable material event, not a publication date.
Each profile carries a last-reviewed date. If a profile has not been updated in over twelve months, treat the funding and clearance figures as potentially stale and verify against primary sources — the FDA 510(k) database, SEC filings for public companies, and company press releases cross-referenced with independent reporting.
Feedback & Corrections
Corrections, deployment experience notes, and questions from clinicians and procurement professionals are welcome. For formal corrections, use the contact page.
Comments
Join the discussion with an anonymous comment.