Aidoc
B2B enterprise software sold to health systems, not payers. Annual per-site / per-module subscription for aiOS and the BriefCase triage portfolio, deployed on-prem or at the edge into the radiology worklist, PACS and mobile alerting; land one acute use case, expand across indications and sites, and increasingly sell aiOS as the governance layer for AI generally. Revenue is hospital opex justified on throughput, transfer capture, length-of-stay and radiologist capacity — NOT a reimbursed per-study fee, and therefore a different engine from Cleerly or HeartFlow, whose price CMS sets.
Earnings, margins, COGS & capex
Aidoc is the board's cleanest counter-example to the reimbursement-moat thesis Cleerly and HeartFlow embody. Cleerly spent five years winning a Category I CPT code and the rate moved by exactly $0.00, because CMS owns the price. Aidoc never sought that pathway: its triage products are not separately reimbursed, so it is paid what a hospital will pay rather than what a payer schedule allows. That is both the bull and the bear case. Bull: it converted the largest independent regulatory stock in imaging AI — 34 authorisations, 33 software-code — into real distribution, nearly 2,000 hospitals and 60M+ cases a year, and has now cleared a foundation model that adds indications down the same pipe. Bear: a budget-funded sale has no annuity beneath it, and CMS showed in the ContaCT NTAP that it will fund AI triage for two federal fiscal years and then stop. The structural fact the board returns to: the four private imaging-AI vendors — Aidoc 33, RapidAI 17, Viz.ai 12, Qure.ai 9 — hold 71 software-code authorisations, more than GE HealthCare (33) and Siemens Healthineers (32) combined, and all four are private. There is no listed vehicle for radiology-AI software, which is why Aidoc is on the watchlist and not in the book.
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~40¢ is cost of goods and ~0¢ operating expense, leaving ~60¢ of operating profit.
Revenue trend
Margins
software-like, dragged by per-study inference at 60M+ cases/yr, edge deployment across ~2,000 hospitals, per-site integration and human governance cost
a >$150M NVIDIA/AWS compute commitment declared alongside the Jul 2025 financing is not the profile of a business near breakeven
equity-funded; $300M raised across two rounds nine months apart
COGS structure
Per-case and per-site in roughly equal measure — what separates Aidoc from both a pure-SaaS comparator and a single-algorithm vendor. Per-case: GPU inference across large CT volumes, transfer and PHI-compliant storage; at 60M+ cases a year a genuine variable line that deflates as inference cheapens — though a foundation model evaluating a broad indication set on every study burns more compute per study than a narrow detector run on order. Per-site does not deflate: aiOS integration into PACS, EHR and worklist, edge infrastructure, data normalisation across scanner vendors and protocols, go-live, and continuous monitoring and governance — recurring human cost scaling with indications times sites, and the reason enterprise clinical-AI gross margin should be modelled below classic SaaS. Critically, because the sell side is a negotiated contract and not an administratively-set rate, falling compute cost is NOT automatically retained: unlike Cleerly, where CMS fixes the price, Aidoc's customer can ask for the saving at renewal.
Capex
Low conventionally, rising practically. No owned imaging fleet: the hardware generating every study belongs to the OEMs, so Aidoc captures none of the scanner economics. Against that, the >$150M NVIDIA / AWS commitment and edge deployment across ~2,000 hospitals are fixed commitments made ahead of revenue; whether they land as capex, prepaid cloud or opex is not disclosed, but economically they are capital intensity.
Latest earnings
n/a
None. Directional company signals only: nearly 2,000 hospitals; 60M+ cases annually; 110M+ cumulative; ~70M patients supported annually (29 Apr 2026); an intent to invest >$150M with NVIDIA and AWS. No revenue, growth, margin, burn or headcount figure is company-confirmed.
- FDA authorisations
- 34, of which 33 under software product codes (97%) — STATE THE BASIS whenever used. Codes QAS 24 / QFM 5 / QIH 4 / JAK 1; JAK is a scanner code, which is why the software-code figure is 33, not 34 (list N=1,524, decisions through 30 Mar 2026, vintage 2026-06-16)
- Rank on the FDA AI list
- 6th overall, 1st among independents on a single basis: GE HealthCare 130 · Siemens Healthineers 95 · Philips 45 · Canon Medical 43 · United Imaging 40 · AIDOC 34 · RadNet/DeepHealth 29 · Samsung 21 · RapidAI 17 · Hyperfine 13 · Viz.ai 12 · Qure.ai 9 · HeartFlow 7 · Cleerly 4 (its icometrix row of 5 is omitted here because GE closed that acquisition on 7 Nov 2025 and those five are already counted inside GE's 130)
- Four-private-vendor software-code total
- 71 = Aidoc 33 + RapidAI 17 + Viz.ai 12 + Qure.ai 9, more than GE (33) and Siemens (32) combined at 65, on a derived software-code mapping. The Aidoc term is 33, not 34 — mixing bases makes the enumeration sum to 72 against a stated 71
- GE-to-Aidoc ratio
- 3.8:1 (130 GE family against Aidoc 34, same corporate-family roll-up basis —)
- Deployment scale
- Nearly 2,000 hospitals; 60M+ cases annually; 110M+ cumulative; ~70M patients supported annually (Aidoc, 29 Apr 2026)
- CARE clearance
- 510(k) CLEARANCE K252970, Jan 2026 — 11 newly cleared indications plus three previously cleared. A CLEARANCE (substantial equivalence), not a PMA APPROVAL and not a De Novo grant; no outcome-benefit claim attaches
- Breakthrough Device Designations
- Two in under a year — CARE Triage (Sep 2025), First Read (25 Jun 2026). A DESIGNATION is prioritised FDA interaction, NOT a clearance, NOT an approval, NOT permission to market
- Disclosed funding
- $150M (23 Jul 2025, General Catalyst and Square Peg co-led) + $150M Series E (29 Apr 2026, Goldman Sachs Alternatives lead, with General Catalyst, SoftBank Vision Fund 2, NVentures); >$500M cumulative
- Last post-money valuation
- NOT DISCLOSED at either round. No negotiated mark is public, none may be inferred from raise size, and the board excludes Aidoc from its ~$33.05B aggregate of last-round private marks for that reason
- Direct reimbursement
- None asserted. Triage products are not separately reimbursed; hospitals pay from operating budget. Precedent: the ContaCT NTAP (Viz.ai) — max $1,040 per case in the FY2021 IPPS final rule, one-year extension FY2022, zero appearances FY2023
Growth drivers
- CARE turning indication expansion from a per-product regulatory and sales cycle into a platform release — the Jan 2026 clearance (K252970) at a reported mean 97% sensitivity / 98% specificity and a claimed order-of-magnitude cut in false alerts, alert burden being the main cause of quiet non-use after go-live
- Installed-base expansion — nearly 2,000 hospitals and 60M+ cases a year is the cheapest distribution for each new indication, and the reason the regulatory stock compounds commercially rather than sitting idle
- aiOS sold as the vendor-neutral deployment, monitoring and governance layer for clinical AI generally — the position that converts a point solution into infrastructure, and one no CT OEM can hold without conceding neutrality
- Extension into reporting — First Read (chest-radiograph draft reports) received Breakthrough Device DESIGNATION on 25 Jun 2026, attacking radiologist throughput — a far larger budget line than triage
- Health-system investor-customers (Hartford HealthCare, Mercy, Sutter Health, WellSpan Health, Jul 2025) — strategic capital doubling as reference accounts and a channel into peer systems
- Radiologist supply constraint — outpatient imaging interpretation turnaround more than doubled between 2014 and 2023 on the Neiman Health Policy Institute analysis Aidoc cites (25 Jun 2026). Capacity, not accuracy, is the buyer's pain
Bull & bear
Aidoc is the only independent in imaging AI that turned a regulatory stock into a distribution asset: 34 authorisations (33 software-code) installed across nearly 2,000 hospitals reading 60M+ cases a year, with a cleared foundation model that makes every future indication a platform release down an existing integration. Because it never depended on a CPT code, it is immune to the rate risk defining the reimbursed-diagnostics names — and it is the most likely private imaging-AI asset to be taken out by a listed acquirer, at a price set by that distribution rather than by the clearance count.
- The clearance stock is real and it is the largest independent one: 34 authorisations, 33 software-code, sixth on the whole FDA AI list and ahead of every non-OEM peer. Against GE's entire 130-strong family the ratio is 3.8:1 — remarkable for a private company against the largest imaging OEM on earth
- Unlike its private peers the stock has been converted. Each new indication ships down an integration that already exists, already passed security and clinical-governance review, and already holds a worklist position — operating leverage, and why the CARE clearance matters more than its sensitivity numbers
- No CMS rate to lose. The bear case that hollowed out reimbursed diagnostics — Cleerly's Category I upgrade moving the rate by exactly $0.00, HeartFlow's 75580 cut 13.8% in one year — does not apply. Aidoc's price is negotiated with a buyer who can measure throughput and length-of-stay, and who is not waiting on an OPPS rule
- CARE is a genuine regulatory first: on the company's account the first double-digit set of acute indications cleared under one foundation model, at mean 97% sensitivity and 98% specificity with roughly an order-of-magnitude reduction in false alerts. False alerts, not missed findings, are what kill adoption after go-live
- Capital and partners are aligned for the compute-heavy phase: >$500M raised, Goldman Sachs Alternatives leading with SoftBank Vision Fund 2 and General Catalyst, NVentures on the cap table plus a >$150M NVIDIA/AWS build-out, and four US health systems invested as customers. Few private clinical-AI companies can fund a foundation-model roadmap and a national sales motion at once
- It is the most credible acquisition target on the board's watchlist and the board says so — 'watch, likeliest via acquisition'. GE's own filings show distribution is what acquirers pay up for, and Aidoc has it
Strip out the authorisation count and Aidoc is an enterprise software vendor selling a cost-centre product into hospital operating budgets, with no payer annuity, no disclosed revenue, no disclosed mark, a compute bill scaling with every study it reads, and five hardware incumbents who own the scanner and can bundle triage at the console. The count proves the wrong thing: 34 clearances are 34 substantial-equivalence decisions, not 34 units of demonstrated benefit — and CMS has already ruled twice, in Aidoc's own product space, that triage AI misses the substantial-improvement bar.
- The number that gets quoted counts clearances, not value. All 34 are 510(k)-class, not PMA, not De Novo, with no demonstrated efficacy. Clearance stock trades as M&A currency and on the board's reading of GE's chequebook it is the CHEAPER half of a deal: $357M for the 12 authorisations GE prices net of cash against $2,293M for Intelerad, which holds zero; the other 11 acquired authorisations carry nothing in that $357M. No multiple is asserted between them — but the direction is unambiguous
- There is no reimbursement and the precedent says there will not be. The first AI NTAP paid up to $1,040 per case in FY2021, was extended once, and appears zero times in FY2023; NTAP applications for pulmonary-embolism triage and ASPECTS scoring — Aidoc's exact category — were REJECTED for failing to show substantial improvement. A vendor with 34 clearances and no payment identity sells its customer a cost, not a revenue
- Budget-funded revenue has no floor. Every contract is renegotiated against a hospital margin that is not expanding, by a buyer now consolidating AI vendors. Falling compute cost does not accrue to Aidoc as it would to a vendor whose price CMS fixes — an enterprise buyer with alternatives asks for it at renewal and usually gets it
- The economics are heavier than the SaaS framing suggests: a foundation model on every one of 60M+ annual studies burns more inference than a narrow detector run on order; edge deployment across ~2,000 hospitals carries field cost; monitoring and governance is recurring human cost scaling with indications times sites. The >$150M NVIDIA/AWS compute commitment declared with the Jul 2025 financing is the tell
- The distribution advantage sits on someone else's hardware. GE 130, Siemens 95, Philips 45, Canon 43 and United Imaging 40 belong to the firms owning the scanner, the console and the service contract, each able to seat triage in the acquisition workflow at a marginal price near zero
- Nothing here is buyable and nothing is priceable. No post-money at either round, no revenue, no burn, no filings — which is why Aidoc is excluded from the ~$33.05B aggregate of last-round private marks rather than estimated. Six disclosed private marks — last-round marks, not marks-to-market, and not uniformly post-money — dated Jun 2025 to May 2026 and summed, total ~$33.05B against the entire 17-name listed pure-play cohort at ~$33.90B of market cap at the 31 Jul 2026 close. Admiring the asset and owning it are different acts
What it is worth
NO VALUATION IS CARRIED, by decision, and this is the strictest case of that rule on the board. The $150M Series E of 29 Apr 2026 disclosed the amount, the lead and the syndicate but NOT a post-money — as did the $150M round of 23 Jul 2025 before it. Two consecutive rounds without a published price means there is not even a stale negotiated mark to quote and caveat, and no valuation may be inferred from raise size, from the quality of the lead, or from >$500M of cumulative funding (hence the exclusion from the six-name ~$33.05B aggregate of last-round private marks). What CAN be framed is the envelope, and it is unusually opaque: unlike Cleerly, whose price CMS publishes, Aidoc's price is a negotiated enterprise contract nobody publishes. Revenue is (sites) x (indications and modules live) x (annual contract value), of which only the first term is public at 'nearly 2,000 hospitals'. An aggregator ~$73.5M 2026 revenue figure circulates with no method or vintage; it is UNVERIFIED and unused. The only priced reference points are M&A: GE HealthCare paid $2,293M net of cash for Intelerad (closed 18 Mar 2026), which holds ZERO FDA AI authorisations, against $357M on that same basis for MIM Software ($259M) and icometrix (~$98M), carrying 12 of the 23 authorisations GE acquired — the other 11 sit outside that filing scope (BK Medical's two alone cost $1.45B under pre-spin GE). The two sums are struck on DIFFERENT bases and NO MULTIPLE is asserted between them. The honest output is not a number but a rule: if a listed acquirer takes Aidoc in, read the price paid for the DISTRIBUTION line — ~2,000 hospitals and 60M+ annual cases — not for the authorisation count.
Hospital operating margins compress, AI consolidation cuts the other way, and an unreimbursed line item is renegotiated down at renewal by a customer with console-level alternatives from the vendor that sold it the scanner. The clearance count keeps rising and keeps proving nothing: the NTAP rejections for PE triage and ASPECTS scoring showed regulators want outcome benefit, and 510(k) substantial equivalence never supplies it. Compute scales with cases rather than contracts, so growth in studies read is growth in COGS; the >$150M commitment lands ahead of the revenue meant to justify it; and the ContaCT precedent repeats at company scale — a category paid $1,040 per case for two federal fiscal years and then nothing. The exit compresses to a distribution sale priced off the reach line, and with two undisclosed post-moneys behind it there is no public mark to fall from — only a private one that never becomes a print.
Aidoc stays private, keeps compounding indications and sites, and stays unpriceable. Revenue grows materially off an undisclosed base as the footprint deepens, but gross margin settles below classic SaaS on per-study inference plus per-site services plus continuous surveillance, and operating margin stays negative while the >$150M compute commitment runs off. Competition does not kill it — the OEMs bundle 'good enough' at the console and take the price-insensitive tail, Microsoft/Nuance takes part of the reporting layer, Aidoc holds the acute-triage and governance seats. No reimbursement pathway arrives, because the NTAP gate is outcome benefit and the trials that would clear it are not the ones being run. It raises again privately at a mark that is negotiated rather than discovered.
The radiology capacity crisis deepens, aiOS becomes the standard governance layer as health systems consolidate AI vendors, and CARE turns indication expansion into a release cadence shipping into ~2,000 existing integrations at near-zero incremental sales cost. First Read converts its designation into a clearance and opens the far larger reporting budget, where ROI is measured in radiologist FTEs rather than triage minutes. Revenue compounds off an undisclosed base to a scale supporting an IPO, and the listing becomes the first buyable expression of imaging-AI software economics on a US exchange — at which point the first question is not the clearance count but what share of revenue is multi-year contracted and what net revenue retention is. In the likelier branch a listed OEM or imaging roll-up pays up for the distribution, and the board's tranche-5 trigger fires against the acquirer rather than against Aidoc.
Framing, not a recommendation, and not investment advice. The board's position is explicit: 'watch, likeliest via acquisition'. WHAT WOULD MAKE IT INVESTABLE: an IPO, or a listed acquirer taking it in — and the second is far likelier. Aidoc sits at the centre of the standing investability-gap regime: the entities owning the moats this board identifies are private (Epic, Aidoc, RapidAI, Viz.ai, and Cleerly as holder of the Category I plaque code), and Aidoc is not even inside the ~$33.05B aggregate of last-round private marks, having no mark to include. Buying GEHC or RDNT for exposure buys imaging-OEM or sites-of-care economics instead. Per the board's deployment discipline, a private-turned-listed name is sized at 3-5% only on the FIRST print that discloses the AI revenue line separately, never on listing day; a listed acquirer absorbing Aidoc into a $30B+ diversified platform would dilute the asset to noise and would not qualify.
SWOT
Strengths
- The largest independent regulatory stock in imaging AI — 34 authorisations, 33 software-code (97%), sixth overall and first among non-OEM independents, ahead of RapidAI 17, Viz.ai 12 and Qure.ai 9
- Clearance stock actually converted into distribution — nearly 2,000 hospitals and 60M+ cases a year — the asset an acquirer would underwrite and the one hardest to rebuild
- The Jan 2026 clearance (K252970) is, on the company's account, the first double-digit set of acute indications cleared under a single foundation model, at a reported mean 97% sensitivity / 98% specificity
- aiOS as a vendor-neutral deployment, monitoring and governance layer — Aidoc as the operating system other people's algorithms run on rather than one more point solution
- >$500M raised, a Goldman Sachs Alternatives-led Series E, NVIDIA's venture arm plus a compute partnership, and four US health systems invested as customers
Weaknesses
- No reimbursement pathway. Every dollar comes from a hospital operating budget renegotiated at renewal, with no payer annuity underneath
- Unpriceable and unmonitorable from outside — no post-money at either the 2025 or 2026 round, no revenue, no burn, no runway, no filings
- 34 authorisations count clearances, not demonstrated benefit — all 510(k)-class substantial-equivalence decisions, not PMA approvals or De Novo grants
- Compute intensity cuts against the software-margin story — a foundation model on every one of 60M+ annual studies plus a >$150M NVIDIA/AWS commitment, with the customer rather than CMS setting the price, so deflation is shareable rather than retained
- Structural dependence on the OEMs it competes with — GE (130), Siemens (95), Philips (45), Canon (43), United Imaging (40) own the acquisition point and can seat triage at the console at a marginal price near zero
Opportunities
- Reporting is a far larger budget line than triage — First Read points Aidoc at radiologist throughput, and the capacity crunch is documented (outpatient turnaround more than doubled 2014–2023)
- Selling aiOS as neutral AI infrastructure to systems that bought five vendors and can govern none — a durable seat if Aidoc holds it before Microsoft/Nuance, Blackford or the PACS vendors
- Indication compounding — each new finding cleared under CARE reaches ~2,000 hospitals through an existing integration, so incremental gross profit per clearance rises versus the one-algorithm-one-sale era
- Reimbursement optionality it has not needed — the CPT 2026 cycle moved AI-assisted imaging further into the code set, and Aidoc has the widest cleared surface to attach a future payment identity to
- An acquisition priced off distribution — GE HealthCare paid $2,293M net of cash for Intelerad (closed 18 Mar 2026), which holds ZERO FDA AI authorisations, against $357M on that same basis for MIM Software ($259M) and icometrix (~$98M), carrying 12 of the 23 authorisations it acquired — the other 11 carry nothing in that figure
Threats
- The ContaCT precedent — the first AI NTAP was approved in the FY2021 IPPS final rule at max $1,040 per case (65% of an applicant-estimated $1,600 average cost), extended one year for FY2022, and appears ZERO times in FY2023. NTAP is a bridge that expires by design
- NTAP rejections in Aidoc's exact space — pulmonary-embolism triage and ASPECTS scoring were turned down for failing the substantial-improvement test (npj Digital Medicine, 2024). The gate is outcome benefit; clearance counts do not clear it
- OEM bundling at the console — the five hardware vendors hold 353 authorisations between them and control the acquisition point
- Hospital budget cyclicality and AI fatigue — an unreimbursed line item is first to be scrutinised when margin compresses, and systems that bought several AI vendors in 2023–2025 are consolidating
- Generative report drafting invites a harder regulatory and liability posture than triage. First Read holds a designation, not a clearance; a draft report is a clinical assertion, not a queue reorder
Moats, dependencies & bottlenecks
Moats
Strong as a barrier to entry, weak as an exclusivity the stock persists, but so does everyone else's The largest independent portfolio on the FDA AI list, and a real gate: each indication is its own submission, pivotal dataset and predicate argument. But 510(k) clearance is substantial equivalence, not exclusivity — it does not stop a rival clearing the same indication, and 1,524 rows on the list shows it stops very few. Counts are quoted on a single basis only.
The genuinely hard asset. Reaching the worklist requires PACS/EHR integration, security review, clinical-governance sign-off and change management at every site; removal costs the customer real disruption. Switching cost is materially higher than for a per-study send-out model like Cleerly's, and this is the line an acquirer would underwrite.
the fastest-decaying item here 110M+ cumulative cases and a cleared multi-indication model is a real head start, and K252970 is a regulatory first. But foundation-model advantage in imaging has proven shorter-lived than expected, the OEMs hold vastly more raw acquisition data, and Aidoc's own clearance creates a public predicate path for followers. A lead, not a wall.
if held, weak if contested The most strategically valuable position Aidoc holds and the least secured. Being the layer that deploys, monitors and governs everyone's clinical AI is infrastructure; being one more algorithm vendor is not. Microsoft/Nuance, Blackford, deepc and the PACS vendors pursue the same seat, and the OEMs resist a neutrality that commoditises their consoles.
>$500M raised, Goldman Sachs Alternatives leading with SoftBank Vision Fund 2 and General Catalyst, NVentures invested beside a >$150M NVIDIA/AWS commitment, and four health systems invested as customers. Capital is not a moat, but at foundation-model scale a funded compute roadmap bars smaller rivals — and health-system investors double as reference accounts.
Aidoc is the default name in multi-finding acute CT triage as RapidAI is in stroke perfusion and Viz.ai in stroke coordination. Brand matters where the buyer cannot easily evaluate model quality — but it is defended by evidence and by alert burden, both directly attackable.
Dependencies
Sole revenue source — no payer intermediary With no reimbursement, 100% of revenue is discretionary hospital opex justified on throughput, transfer capture, length-of-stay and radiologist capacity — defensible in a good margin year, first reviewed in a bad one. No annuity, no assignment of benefits, no payer schedule; only a renewal conversation.
Siemens Healthineers, Philips, Canon Medical, United Imaging) Upstream input and competitive threat in the same entities Every case Aidoc reads was acquired on hardware it does not own, from vendors holding 130 / 95 / 45 / 43 / 40 authorisations. They supply the studies, own the customer relationship and the service contract, and can bundle triage at the console. United Imaging is a mainland-China-listed OEM, named as a competitive fact only — no mainland-China listing is recommended here.
COGS and capability input A stated >$150M commitment across NVIDIA and AWS, with NVentures also on the cap table. Training plus 60M+ annual inference runs make compute a cost line and a capability constraint. Falling GPU cost helps — but unlike an administratively-priced peer, the customer can claim the saving at renewal.
including generative outputs The line runs on 510(k) clearances plus predetermined-change discipline; the next leg (First Read) holds only a Breakthrough Device DESIGNATION as of 25 Jun 2026 — not a clearance, not an approval, not permission to market. Generative clinical output faces a harder review and liability posture, on a timeline Aidoc does not control.
Distribution rails aiOS reaches the radiologist through someone else's worklist and record. Epic holds zero FDA AI authorisations and needs none — it owns the chart. A PACS or EHR vendor closing its AI marketplace narrows the rail Aidoc rides.
Pre-profit on any reasonable assumption, $300M raised across two rounds in nine months, a nine-figure compute commitment outstanding, and neither runway nor mark observable. A private company forced to raise into a colder market discovers its valuation the hard way — and with no prior post-money there is not even a reference point to reprice from.
Advantages
- Largest independent FDA authorisation stock in imaging AI — 34, of which 33 software-code (97%), sixth overall and first among non-OEM independents
- The only one of the four private imaging-AI vendors whose clearance stock is matched by comparable distribution: nearly 2,000 hospitals, 60M+ cases a year
- A cleared foundation model (K252970, Jan 2026) that turns each new indication into a platform release down an existing integration
- Structural immunity to the CMS rate risk defining the reimbursed-diagnostics names — no OPPS Addendum B exposure, no status-indicator drift, no annual ratesetting
- aiOS as a candidate neutral infrastructure layer for clinical AI, a position no CT OEM can occupy without conceding vendor neutrality
- Funded for the compute-heavy phase — >$500M raised, a Goldman Sachs Alternatives-led Series E, NVIDIA's venture arm plus a >$150M NVIDIA/AWS commitment, four US health systems invested as customers
Weaknesses
- Zero direct reimbursement, and the NTAP record in this category is one expiry (ContaCT, FY2021–FY2022) and two rejections (PE triage, ASPECTS)
- 34 clearances are substantial-equivalence decisions, not evidence of clinical benefit
- No disclosed valuation at either round, no revenue, no burn, no runway: unpriceable and unmonitorable from outside
- Compute-heavy at foundation-model scale, with a >$150M infrastructure commitment against an undisclosed revenue base
- Distribution rides on OEM hardware and third-party PACS/EHR rails, controlled by parties that are also competitors or gatekeepers
- The next growth leg (First Read) is a DESIGNATION, not a clearance, and generative clinical output carries a harder regulatory and liability profile
Bottlenecks
- No reimbursement pathway — every dollar competes against clinical hires and equipment inside a hospital operating budget, with no payer offset
- Evidence of outcome benefit rather than accuracy — the gate CMS applied when rejecting NTAP for pulmonary-embolism triage and ASPECTS scoring, and the gate guideline bodies apply. 34 clearances do not clear it
- Alert burden and clinician trust — a multi-indication model on every study widens the false-positive surface, the mechanism by which deployed imaging AI quietly stops being used. The claimed order-of-magnitude reduction is unaudited outside the FDA submission
- Per-site implementation cost and cycle time — integration, security review, governance and change management at each of ~2,000 hospitals is services-heavy and caps how fast logos convert to revenue
- Compute cost scaling with cases rather than contracts
- Model surveillance at scale — monitoring drift across dozens of indications times thousands of sites is recurring human cost that cannot be cut without weakening the aiOS pitch it underwrites
- No liquidity and no mark — private, no disclosed post-money at two consecutive rounds, and the board's trigger for investability is an IPO or acquisition that has not happened
Top signals & trends
Top signals
A crossover-quality lead nine months after the prior $150M round is a genuine demand signal. It says nothing about price: no post-money was disclosed and none may be inferred.
Two consecutive undisclosed marks removes the only external reference point. Not evidence of a down round — but it is why Aidoc is excluded from the ~$33.05B aggregate of last-round private marks and no estimate is carried.
Parsed from FDA's machine-readable list (N=1,524, decisions through 30 Mar 2026). The competing '17 cleared algorithms' figure is a company/aggregator construction FDA does not publish and is RETIRED — asserted nowhere here.
bullish on the category, bearish on investability · The count rests on a derived software-code mapping. The board's core structural finding: the genuinely AI-software authorisations sit with private companies, which is why no US-listed vehicle for radiology-AI software exists.
A regulatory first for a foundation model in clinical imaging, and the mechanism that makes indication expansion cheap. It is a CLEARANCE (substantial equivalence), not a PMA approval or De Novo grant, so no outcome-benefit claim attaches.
Bullish on ambition and on the throughput budget it targets; the designation confers prioritised FDA interaction, NOT clearance, NOT approval, NOT permission to market. Conflating the two is a specific error.
bearish on near-term economics, bullish on capability · A compute commitment roughly the size of either $150M round on its own — the clearest evidence that foundation-model clinical AI is not capital-light and that the operating-margin assumption stays negative.
Verified at primary in the Federal Register; NTAP requests for PE triage and ASPECTS scoring were rejected outright. The category's only marquee payment pathway lasted two federal fiscal years by design.
Customer capital is the strongest available substitute for disclosed revenue — systems that put equity in generally have the product in production. It is also a governance consideration: reference accounts holding equity are not neutral references.
MIM Software $259M + icometrix ~$98M, both net of cash, carrying 12 of the 23 authorisations GE acquired; the other 11 carry nothing in that figure. No multiple is asserted. Read only as direction — acquirers pay for reach, which is bullish for an exit and bearish for anyone valuing Aidoc off the clearance count.
Trends
Health systems that bought five vendors in 2023–2025 cannot govern them and are consolidating — the demand aiOS is built for, with breadth of clearances as the credential. The same trend is fatal to sub-scale single-indication vendors.
The Jan 2026 CARE clearance is the first regulatory proof that a multi-indication model can be authorised as one device. It compresses the cost of adding findings — for Aidoc first, then for everyone following the predicate it created.
Radiology accounts for the large majority of FDA-cleared medical AI while only a handful of algorithms carry a code that pays; the npj Digital Medicine review counts two ML algorithms with a Category I CPT code and five SaMD algorithms at Category III. The binding constraint is evidence of benefit, not clearance throughput.
NTAP runs two to three federal fiscal years, is per-case capped and expires by design; the Bipartisan Policy Center counts four AI tools holding NTAP in 2025 against one in 2023 (SECONDARY, attributed, not re-grounded against the IPPS rules). More pathways, none permanent.
GE, Siemens, Philips, Canon and United Imaging hold 353 authorisations between them and own the scanner, console and service contract. 'Good enough, already paid for' compresses willingness to pay for third-party software across the category.
Outpatient imaging interpretation turnaround more than doubled between 2014 and 2023 on the Neiman Health Policy Institute analysis Aidoc cites (25 Jun 2026). Capacity is the buyer's actual pain, it is worsening, and it is independent of Aidoc's execution.
A far larger budget pool and the natural next leg via First Read — but it puts Aidoc into a field with Rad AI, Microsoft/Nuance and the ambient-documentation vendors, under a heavier regulatory standard, holding a designation rather than a clearance.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
GE HealthCare / Siemens Healthineers / Philips / Canon Medical / United Imaging The CT and X-ray installed base — the upstream input. Every case Aidoc analyses is acquired on OEM hardware it does not own, and these five hold 130 / 95 / 45 / 43 / 40 FDA AI authorisations respectively. Supply and competition in the same entities. United Imaging is mainland-China-listed and named as a competitive fact only; no mainland-China listing is recommended here.
GPU compute for CARE training and for inference across 60M+ studies a year; NVentures invested in both the Jul 2025 and Apr 2026 rounds and is named in the stated >$150M commitment.
Named alongside NVIDIA in the >$150M commitment — cloud training capacity, PHI-compliant storage and hosting for the non-edge footprint.
validation and monitoring staff Radiologist-labelled training data, pivotal-study readers, and the ongoing performance-monitoring function aiOS is sold on — the COGS line that does not deflate with GPU prices and scales with indications times sites.
Hospitals and health systems (nearly 2,000 deployed) The buyer and the payer. Purchased from operating budget on throughput, transfer-capture, length-of-stay and radiologist-capacity arguments — no reimbursement offset, so the ROI case must survive a CFO review at every renewal.
Health-system strategic investors (Hartford HealthCare, Mercy, Sutter Health, WellSpan Health) Participated in the $150M round of 23 Jul 2025 alongside NVentures. Customer capital is the strongest available proxy for production usage — and a reason to treat these references as non-neutral.
stroke and trauma services, transfer networks The service lines capturing the economic benefit of triage — faster time-to-treatment, better transfer capture, shorter stays. Where the ROI story lives even when radiology signs the contract.
Throughput buyers rather than triage buyers — the constituency First Read targets, and where Rad AI and the reporting incumbents are already established.
Private. The closest strategic analogue: 12 authorisations, care-coordination-first around stroke and cardiac pathways, and holder of the category's only marquee payment history — the ContaCT NTAP, and a CMS rate of $128.90 effective 1 Jan 2025 for CPT 0764T/0765T on AI-enhanced ECG (company-sourced). Competes hardest where the sale is to the service line rather than radiology. No valuation exists on the board for this name either.
Private. 17 authorisations, all software-code — second among independents behind Aidoc. Owns acute stroke and neurovascular on clinical evidence (the RAPID engine selected patients in DEFUSE 3 and DAWN), the credential Aidoc's breadth cannot buy in that indication. Last disclosed capital: a $75M growth investment led by a credit fund, Jul 2023, no valuation disclosed.
Private. 9 authorisations; strongest in chest X-ray and TB screening and in emerging-market deployments Aidoc's US enterprise motion does not reach. Competes directly in the chest-radiograph franchise First Read targets, at a price point built for lower-resource buyers.
130 FDA AI authorisations — leaderboard top on a corporate-family basis, 8.5% of the 1,524-row list — at $68.02 / $30.72B on the 31 Jul 2026 close, $21.27B TTM revenue. Simultaneously the largest supplier of the scans Aidoc reads, the most credible console-level bundler, and the most likely acquirer; the Intelerad deal (18 Mar 2026, $2,293M net of cash, zero AI authorisations) shows it buying reach rather than clearances.
95 authorisations (90 Siemens-named + Varian 5). AI-Rad Companion and the Healthineers digital ecosystem give it a console-and-enterprise path into Aidoc's worklist position, with an existing purchasing relationship at almost every large customer. On a derived software-code mapping it holds 32 — fewer than Aidoc's 33 alone, which is the point of the four-vendor comparison.
45 authorisations (the trade-press tally of 58 does not reproduce against the FDA file and is retired). Enterprise-imaging and radiology-informatics footprint makes Philips a bundler and plausible acquirer more than a head-to-head triage rival; its platform layer competes with aiOS rather than with the algorithms.
43 authorisations. Strong CT hardware franchise, particularly high-end and cardiac, with console-level automation. Competes by making the third-party read look optional rather than by beating it on accuracy.
29 authorisations across the DeepHealth / Quantib / iCAD / Gleamer / See-Mode / Aidence family — the largest non-OEM roll-up of imaging AI, attached to a US imaging-centre operator the board marks at $5.02B with -$14.19M TTM net income. The closest thing to a listed proxy for third-party imaging AI, and the board explicitly rejects it as one: a long is a bet on sites of care, not AI economics.
Not an algorithm vendor but the most dangerous competitor for aiOS's actual position: a distribution and orchestration network riding PowerScribe's incumbency in radiology reporting, able to route third-party AI into the report where radiologists already work. It also owns the reporting surface First Read must enter.
Private. The direct competitor to First Read — AI-generated draft radiology reports and reporting-workflow automation, commercially deployed across US radiology groups while Aidoc's equivalent holds only a Breakthrough Device Designation (25 Jun 2026). Same throughput budget, attacked from the reporting side, and there first.