FDA AI Medical Device Statistics 2026
By Axis Intelligence Research
Co-author: Jennifer Miller — Digital Health | Last updated: September 2, 2026 | License: CC BY 4.0
The FDA authorized 425 AI-enabled medical devices between January 1, 2025 and March 30, 2026, according to Axis Intelligence Research’s row-level analysis of the agency’s own dataset. Radiology took 76.3% of the 2025 cohort. Only seven of the 425 came through a pathway other than 510(k) — a ratio that shapes what “FDA-cleared AI” actually certifies.
Quick Answer
The FDA’s AI-Enabled Medical Device List recorded 333 authorizations in calendar 2025 and 92 more through March 30, 2026, its most recent decision date. According to Axis Intelligence Research, 98.35% of those 425 authorizations (418 devices) came through the 510(k) substantial-equivalence pathway, not through De Novo or PMA. Radiology accounted for 75.0% of Q1 2026 authorizations.
Key Findings
- According to Axis Intelligence Research, the FDA authorized 333 AI-enabled medical devices in calendar year 2025 and 92 in the first quarter of 2026, based on Date of Final Decision in the agency’s AI-Enabled Medical Device List.
- According to Axis Intelligence Research, 418 of 425 AI device authorizations between January 2025 and March 2026 — 98.35% — used the 510(k) pathway, with four De Novo grants and three PMA decisions accounting for the remainder.
- According to Axis Intelligence Research, radiology’s share of FDA AI device authorizations was 76.3% in 2025 and 75.0% in Q1 2026, showing no meaningful decline across five consecutive quarters.
- According to Axis Intelligence Research, 237 distinct parent organizations received at least one FDA AI device authorization in the 15-month window, and 172 of them — 72.57% — received exactly one.
- According to Axis Intelligence Research, the eight incumbent imaging OEMs took 23.42% of 2025 authorizations but 30.43% of Q1 2026 authorizations, the sharpest concentration shift in the observed period.
How Many AI Medical Devices Has the FDA Authorized?
The honest answer starts with a caveat the headline numbers usually omit: the FDA does not publish a running total, and the list is not a census. The agency states plainly that its roster is not a comprehensive resource, and that devices are identified primarily by AI-related terms appearing in the summary descriptions of their marketing authorization documents, using the vocabulary set out in the FDA Digital Health and Artificial Intelligence Glossary.
So rather than repeat a cumulative figure we did not count, Axis Intelligence Research downloaded the FDA’s AI-Enabled Medical Device List and counted a defined, verifiable window: every authorization with a Date of Final Decision between January 1, 2025 and March 30, 2026 — the latest decision present in the list as published.
| Period | Authorizations | Source |
|---|---|---|
| 2025 Q1 | 83 | FDA AI-Enabled Medical Device List |
| 2025 Q2 | 94 | FDA AI-Enabled Medical Device List |
| 2025 Q3 | 83 | FDA AI-Enabled Medical Device List |
| 2025 Q4 | 73 | FDA AI-Enabled Medical Device List |
| 2025 total | 333 | FDA AI-Enabled Medical Device List |
| 2026 Q1 (through Mar 30) | 92 | FDA AI-Enabled Medical Device List |
| 15-month total | 425 | FDA AI-Enabled Medical Device List |
Snapshot: FDA list page last updated June 16, 2026; latest Date of Final Decision March 30, 2026. Retrieved September 2, 2026.
Q1 2026 came in 10.84% above Q1 2025 — 92 against 83. That is real growth, but it is smaller than the “AI clearances are exploding” framing suggests, and it needs one honest qualification. The FDA states that devices authorized but without published decision summaries at collection time roll into a later update. Recent quarters are therefore floors, not finals. The 92 will probably rise. The 83 already has.
Jennifer Miller writes: There is a reason we count a window instead of quoting a cumulative total, and it is not pedantry. The FDA retroactively adds older devices to this list as it identifies them. That means the “total since 1995” moves in both directions between updates, and any two publications counting on different dates get different answers and both are right. A window with a stated snapshot date is reproducible. A cumulative total without one is a number with a short shelf life.
Which Medical Specialties Get the Most FDA AI Clearances?
The FDA assigns each authorization a lead review panel. That field is the cleanest available proxy for clinical specialty, and it tells a blunt story.
FDA AI device authorizations by review panel, 2025
| Review panel (lead) | Devices | Share | Source |
|---|---|---|---|
| Radiology | 254 | 76.28% | FDA AI-Enabled Medical Device List |
| Cardiovascular | 27 | 8.11% | FDA AI-Enabled Medical Device List |
| Neurology | 19 | 5.71% | FDA AI-Enabled Medical Device List |
| Gastroenterology-Urology | 9 | 2.70% | FDA AI-Enabled Medical Device List |
| Anesthesiology | 8 | 2.40% | FDA AI-Enabled Medical Device List |
| Pathology | 4 | 1.20% | FDA AI-Enabled Medical Device List |
| Hematology | 3 | 0.90% | FDA AI-Enabled Medical Device List |
| Orthopedic | 3 | 0.90% | FDA AI-Enabled Medical Device List |
| General and Plastic Surgery | 2 | 0.60% | FDA AI-Enabled Medical Device List |
| Obstetrics and Gynecology | 2 | 0.60% | FDA AI-Enabled Medical Device List |
| Ophthalmic | 1 | 0.30% | FDA AI-Enabled Medical Device List |
| Microbiology | 1 | 0.30% | FDA AI-Enabled Medical Device List |
| Total | 333 | 100% |
Radiology share by quarter
| Quarter | Radiology devices | Share of quarter | Source |
|---|---|---|---|
| 2025 Q1 | 63 of 83 | 75.90% | FDA AI-Enabled Medical Device List |
| 2025 Q2 | 71 of 94 | 75.53% | FDA AI-Enabled Medical Device List |
| 2025 Q3 | 64 of 83 | 77.11% | FDA AI-Enabled Medical Device List |
| 2025 Q4 | 56 of 73 | 76.71% | FDA AI-Enabled Medical Device List |
| 2026 Q1 | 69 of 92 | 75.00% | FDA AI-Enabled Medical Device List |
Five quarters, a 2.1-point spread, no trend. According to Axis Intelligence Research, radiology’s dominance of FDA AI device authorizations is not eroding at any measurable rate — the specialty has held between 75.0% and 77.1% of quarterly authorizations across the entire observed period.
The Q1 2026 cohort did add three panels that were absent from the same quarter a year earlier: Ear Nose & Throat (Tyto Care’s eardrum bulging detection De Novo), General Hospital, and General and Plastic Surgery. Twelve panels appeared across 2025; ten appeared in Q1 2026 alone.
Cardiovascular is the one panel with visible movement, and it moves erratically: 1 device in Q1 2025, then 10, 11, 5 and 10 in the four quarters that followed. That first quarter is not a typo. Exactly one cardiovascular AI device — Fitbit’s Loss of Pulse Detection — carried a Q1 2025 decision date. Anyone modelling cardiology AI growth off a Q1 2025 baseline will produce a number that means nothing.
Which Companies Get the Most FDA AI Device Clearances?
FDA applicant names are messy. “Ge Medical Systems, LLC,” “GE Medical Systems Ultrasound and Primary Care Diagnostics,” “Ge Healthcare Japan Corporation,” “GE Hualun Medical Systems Co., Ltd.” and “GE HealthCare” appear as five separate applicants. Axis Intelligence Research mapped applicant strings to parent organizations, consolidating national units, named divisions and acquired subsidiaries, then recounted.
Top parent organizations, January 2025 – March 2026
| Rank | Parent organization | Authorizations | Share of 425 | Source |
|---|---|---|---|---|
| 1 | GE HealthCare | 32 | 7.53% | FDA list, Axis parent mapping |
| 2 | Siemens Healthineers | 19 | 4.47% | FDA list, Axis parent mapping |
| 3 | Philips | 18 | 4.24% | FDA list, Axis parent mapping |
| 4 | United Imaging | 14 | 3.29% | FDA list, Axis parent mapping |
| 5 | Canon Medical Systems | 10 | 2.35% | FDA list, Axis parent mapping |
| 6 | Bunkerhill Health | 7 | 1.65% | FDA list, Axis parent mapping |
| 6 | Pearl | 7 | 1.65% | FDA list, Axis parent mapping |
| 8 | Aidoc | 6 | 1.41% | FDA list, Axis parent mapping |
| 9 | Esaote | 5 | 1.18% | FDA list, Axis parent mapping |
| 9 | Fujifilm | 5 | 1.18% | FDA list, Axis parent mapping |
| 9 | iSchemaView (RapidAI) | 5 | 1.18% | FDA list, Axis parent mapping |
The five largest imaging OEMs together took 93 of 425 authorizations — 21.88%. That is concentration, but it is not a cartel. According to Axis Intelligence Research, 237 distinct parent organizations received at least one FDA AI device authorization in the window, and 172 of them — 72.57% — received exactly one.
Jennifer Miller writes: That long tail is the finding that gets misread most often. A single 510(k) clearance is a regulatory milestone, not a commercial one. Nothing in this dataset says a device was ever installed, billed, or used on a patient. Clearance counts measure submissions that survived review; they do not measure adoption, revenue, or clinical uptake. The 172 one-clearance companies include venture-backed startups whose first clearance is their entire commercial story, and they include hardware vendors who added a denoising algorithm to an existing scanner. The list does not distinguish them. Anyone using clearance counts as a market-share proxy is measuring paperwork.
The AI Device Diffusion Index (AIDI™)
Counts alone say how much. They do not say whether AI in medicine is spreading — reaching new specialties and new entrants — or consolidating into the same imaging departments and the same vendors. So Axis Intelligence Research built a metric to track exactly that.
AIDI™ — AI Device Diffusion Index
AIDI = 100 × [ (0.5 × Non-Radiology Share) + (0.5 × Challenger Share) ]
Where, for any period:
- Non-Radiology Share = authorizations outside the FDA Radiology review panel ÷ total authorizations
- Challenger Share = authorizations not attributed to the fixed incumbent OEM bloc ÷ total authorizations
- Incumbent OEM bloc (fixed, named, unchanging): GE HealthCare, Siemens Healthineers, Philips, Canon Medical Systems, Fujifilm, United Imaging, Samsung, Esaote
Weights are equal by design: specialty diffusion and competitive diffusion are treated as equally informative. The bloc is a fixed named list rather than a “top N” so that readings from a 73-device quarter and a 333-device year are directly comparable — a top-10 measure would rise mechanically as sample size falls.
A higher AIDI means AI authorizations are more diffuse. A falling AIDI means they are concentrating.
AIDI readings
| Period | Non-radiology share | Challenger share | AIDI | Source |
|---|---|---|---|---|
| 2025 Q1 | 24.10% | 83.13% | 53.6 | Axis Intelligence Research |
| 2025 Q2 | 24.47% | 72.34% | 48.4 | Axis Intelligence Research |
| 2025 Q3 | 22.89% | 73.49% | 48.2 | Axis Intelligence Research |
| 2025 Q4 | 23.29% | 78.08% | 50.7 | Axis Intelligence Research |
| 2026 Q1 | 25.00% | 69.57% | 47.3 | Axis Intelligence Research |
| 2025 full year | 23.72% | 76.58% | 50.2 | Axis Intelligence Research |
| 15-month baseline | 24.00% | 75.06% | 49.5 | Axis Intelligence Research |
The 15-month reading of 49.5 is the baseline. It carries no historical comparison because none has been computed.
According to Axis Intelligence Research, AIDI fell 6.3 points between Q1 2025 and Q1 2026, from 53.6 to 47.3 — and the two components moved in opposite directions. Non-radiology share rose to its highest observed level, 25.00%. Challenger share fell to its lowest, 69.57%, as the incumbent OEM bloc’s take jumped from 23.42% across 2025 to 30.43% in Q1 2026.
That is the headline this dataset actually supports: AI medical devices are reaching more specialties while being cleared by fewer, larger companies. Diffusion by clinical use, consolidation by vendor.
Robustness. Recomputing AIDI with the bloc cut to the Big Four — GE HealthCare, Siemens Healthineers, Philips and Canon — produces 55.4, 53.2, 53.6, 51.4 and 50.0 across the same five quarters. The level shifts; the direction does not. Both specifications show diffusion falling into Q1 2026.
AIDI™ is published by Axis Intelligence Research under CC BY 4.0 and is reproducible from the accompanying dataset.
Which FDA Pathway Do AI Medical Devices Use?
This is the section that should change how the other numbers are read.
| Pathway | 2025 | 2026 Q1 | Combined | Share of 425 | Source |
|---|---|---|---|---|---|
| 510(k) | 329 | 89 | 418 | 98.35% | FDA AI-Enabled Medical Device List |
| De Novo | 2 | 2 | 4 | 0.94% | FDA AI-Enabled Medical Device List |
| PMA / PMA supplement | 2 | 1 | 3 | 0.71% | FDA AI-Enabled Medical Device List |
A 510(k) clearance establishes substantial equivalence to a legally marketed predicate device. It is not a finding that the device is safe and effective in the sense a drug approval carries, and it does not require a clinical trial. De Novo, used four times in 15 months, creates a new device classification where no predicate exists. PMA, used three times, is the most demanding pathway.
Every non-510(k) AI authorization in the window
| Date | Submission | Device / applicant | Pathway | Panel |
|---|---|---|---|---|
| 2025-04-30 | P230022 | Element Science — Jewel Patch Wearable Cardioverter Defibrillator | PMA | Cardiovascular |
| 2025-05-30 | DEN240047 | Clairity — Allix5 | De Novo | Radiology |
| 2025-07-31 | DEN240068 | Artera — ArteraAI Prostate | De Novo | Pathology |
| 2025-11-21 | P130020/S005 | GE HealthCare — Senographe Pristina 3D | PMA supplement | Radiology |
| 2026-02-11 | DEN250007 | Ultrasound AI — Delivery Date AI | De Novo | Radiology |
| 2026-03-03 | P250008 | Perimeter Medical Imaging AI — Claire OCT System | PMA | General and Plastic Surgery |
| 2026-03-17 | DEN250014 | Tyto Care — Tyto Insights for Eardrum Bulging Detection | De Novo | Ear Nose & Throat |
Source: FDA AI-Enabled Medical Device List. Submission numbers link to the FDA device databases.
Peer-reviewed work makes the consequence concrete. A cross-sectional study of 691 FDA-cleared AI/ML devices published in JAMA Health Forum in September 2025 found that 660 of the 691 devices did not report demographic representation, 323 did not report study design, six reported data from randomized clinical trials, and three reported patient outcomes. Premarket safety assessments were documented for 195 devices; postmarket adverse events, including one death, were reported for 36.
We disagree with the common framing that this is an evidence failure. It is an evidence design. The 510(k) pathway was never built to generate outcome data, and 98.35% of AI devices use it. Criticising individual clearances for lacking trial evidence misdiagnoses the cause: the statute asks for equivalence, and manufacturers supply equivalence. The question worth asking is whether equivalence is the right test for a product class whose behaviour changes after clearance.
Can an FDA-Cleared AI Device Be Retrained After Clearance?
Yes — within limits the FDA now specifies in advance.
The agency’s final guidance, Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions (docket FDA-2022-D-2628), is current as of August 18, 2025. A Predetermined Change Control Plan lets a manufacturer pre-specify modifications, and the methodology for validating them, inside the original submission. Once the FDA authorizes the PCCP, those changes can be implemented without a new marketing submission. The guidance applies across all three pathways — 510(k), De Novo and PMA.
Note the sequencing, because secondary write-ups routinely get it wrong: the Federal Register notice announcing this guidance dates from December 2024, while the FDA’s current version of the document carries an August 2025 date. Both dates are correct for different things. Cite the one you mean.
The FDA has also signalled where the list goes next, stating it will explore methods to identify and tag devices incorporating foundation models — from large language models to multimodal architectures — and encouraging sponsors to include that information in public summaries. As of the March 30, 2026 snapshot, no such tag exists in the published data. Any claim about how many FDA-authorized devices use LLMs is, at this date, unsupported by the FDA’s own dataset.
Methodology
Collection. Axis Intelligence Research retrieved the FDA AI-Enabled Medical Device List and its downloadable dataset on September 2, 2026. The FDA page carried a last-modified date of June 16, 2026; the most recent Date of Final Decision in the published list was March 30, 2026.
Scope. Every row with a Date of Final Decision between January 1, 2025 and March 30, 2026 was included: 425 rows. Rows outside that window were excluded. No cumulative “since 1995” figure is published here because that count was not independently verified against the full file.
Unit of analysis. One row equals one marketing authorization, not one distinct product. Devices appearing multiple times under separate submission numbers — iterative clearances of the same product family — are counted once per authorization. This is the FDA’s own structure and is preserved deliberately.
Pathway classification. Assigned from submission-number prefix: K = 510(k), DEN = De Novo, P = PMA or PMA supplement.
Parent-company mapping. Applicant strings were mapped to parent organizations by Axis Intelligence Research. National subsidiaries, named divisions and acquired entities were consolidated under current parents. The mapping is disclosed in the accompanying dataset and is the only step in this analysis that involves editorial judgment.
AIDI formula. Stated in full above, with both components published separately so any reading can be recomputed or reweighted.
Arithmetic. Every count, share and index reading was computed programmatically from the row-level data and verified before writing. No figure in this article was estimated.
Scope and caveats
Three things this dataset does not do. It is not a complete census of AI-enabled devices — the FDA says so directly, and devices whose summaries omit the agency’s AI vocabulary will be absent. It lags: authorizations without published decision summaries at collection time are added in later updates, so the most recent quarter is a floor. And it records regulatory decisions only — nothing here indicates whether a cleared device was ever deployed, reimbursed or used.
About This Dataset
Contents. 171 rows covering FDA AI-enabled medical device authorizations from January 1, 2025 to March 30, 2026: quarterly and monthly counts, review-panel breakdowns, pathway distribution, parent-company rankings, AIDI readings with both input components, and every non-510(k) authorization individually.
Provenance. Every row carries source_org, source_document, source_url, retrieved_date, is_primary, axis_calculated and, where calculated, a method_note giving the exact formula.
Licence. CC BY 4.0. Free to reuse, including commercially, with attribution.
Citation: Axis Intelligence Research, FDA AI Medical Device Statistics 2026, 2026.
How to Cite This Research
APA Axis Intelligence Research, & Miller, J. (2026). FDA AI medical device statistics 2026: 425 authorizations analyzed by specialty, company and pathway. Axis Intelligence. https://axis-intelligence.com/fda-ai-medical-device-statistics/
MLA Axis Intelligence Research, and Jennifer Miller. “FDA AI Medical Device Statistics 2026: 425 Authorizations Analyzed by Specialty, Company and Pathway.” Axis Intelligence, 2 Sept. 2026, axis-intelligence.com/fda-ai-medical-device-statistics/.
Chicago Axis Intelligence Research and Jennifer Miller. “FDA AI Medical Device Statistics 2026: 425 Authorizations Analyzed by Specialty, Company and Pathway.” Axis Intelligence. September 2, 2026. https://axis-intelligence.com/fda-ai-medical-device-statistics/.
Frequently Asked Questions
Does FDA clearance of an AI device mean the algorithm was tested in a clinical trial?
Usually not. 98.35% of AI device authorizations between January 2025 and March 2026 used the 510(k) pathway, which asks whether a device is substantially equivalent to an existing predicate — not whether it improves patient outcomes. The JAMA Health Forum study of 691 cleared AI/ML devices found six that reported randomized clinical trial data.
Why do published counts of FDA-authorized AI devices disagree with each other?
Because the list is restated. The FDA adds older devices retroactively as it identifies them and appends recent authorizations once decision summaries are published. Two counts taken weeks apart legitimately differ. Any figure without a stated snapshot date cannot be reconciled with any other figure.
What does “Panel (Lead)” mean on the FDA AI device list, and why does radiology dominate it?
Panel (Lead) is the FDA advisory-committee specialty that led the review, not the clinical department that uses the device. Radiology dominates because imaging generates large, labelled, digital datasets and because a dense predicate base makes 510(k) submissions viable — the same reason it held 75.0% to 77.1% of quarterly authorizations across five straight quarters.
If a device already has FDA clearance, can the manufacturer retrain the model without a new submission?
Only if an authorized Predetermined Change Control Plan covers that specific modification. Under the FDA’s PCCP final guidance, current as of August 18, 2025, manufacturers pre-specify permitted changes and their validation methodology inside the original submission. Modifications outside the authorized plan still require a new marketing submission.
Are De Novo grants rarer than 510(k) clearances for AI devices?
Dramatically. Four De Novo grants and three PMA decisions appear across 425 authorizations in 15 months. De Novo is used when no predicate device exists, so it tends to mark genuinely novel categories — Artera’s prostate pathology device and Tyto Care’s eardrum assessment tool being two examples from this window.
Does the FDA list identify devices that use large language models or foundation models?
Not yet. The FDA has stated it will explore methods to tag devices incorporating foundation models, including LLMs and multimodal architectures, and has asked sponsors to include that information in public summaries. No such field existed in the March 30, 2026 snapshot, so no defensible count of LLM-based cleared devices can be derived from this source.
How soon after its decision date does a device appear on the FDA list?
Not immediately. The FDA states that authorizations whose decision summaries have not been published within a collection period are incorporated into a subsequent update. In practice the most recent quarter in any snapshot undercounts, which is why quarter-over-quarter comparisons should use closed quarters.
Does an authorization on this list mean the device is being used in hospitals?
No. The list records marketing authorizations only. It carries no information about installation, procurement, reimbursement or clinical use — which is why the 172 organizations holding exactly one authorization in this window cannot be read as 172 commercially active AI vendors.
Can I reuse these figures and the AIDI readings?
Yes. The dataset and the AIDI™ methodology are published under CC BY 4.0. Attribute to Axis Intelligence Research with the snapshot date, since the underlying FDA list is revised.
Related Research
For the broader market context around device authorizations, see Axis Intelligence Research’s analysis of digital health statistics, which covers telehealth, wearables and health-AI funding. For how device regulation fits into national AI law, see AI regulation by country. For enterprise AI deployment rates outside healthcare, see companies using AI statistics.