This week the money stopped buying healthcare AI software and started buying the licence: Epsilon took $27.6M to become a radiology practice, Nara took $14M to become a TPA, NVIDIA bought into Verily itself — and the same half-year still shows zero digital-health IPOs
Lay last week's cheques side by side and a shape appears that was not there before. Fierce Healthcare's fundraising tracker logged six rounds between September 8 and 10; add Nara Health's Khosla-led round on September 14 and the largest of them share one trait: they are not selling a model to a hospital, they are becoming the institution being displaced. Epsilon Health took close to $27.6M to run a radiology practice of its own; Nara took $14M to be a third-party administrator; Forus put $150M behind a $3B valuation for sitting in the middle of the prescription. In the same week NVIDIA became a shareholder in Verily — the shovel seller now owns part of the mine. Behind all of it, Rock Health's half-year count says something else entirely: $7.4B, 244 deals, 115 acquisitions, zero IPOs.
01 — Top Stories
Epsilon Health leaves stealth with close to $27.6M — not to sell reading software, but to run a radiology practice
Epsilon Health — founded in 2024 in San Francisco by Rustin Rassoli — came out of stealth on September 10. Fierce Healthcare's tracker logs $27.6M led by AlleyCorp with Uncork Capital, Renegade Partners, SemperVirens and Jack Altman; Axios on September 12 reported a $20M Series A (the gap is most likely earlier capital folded in; the company has not broken the rounds out). The number matters less than the shape: Epsilon is itself a practice staffed by board-certified radiologists, building AI into its own workflow rather than licensing a tool to someone else's reading room.
Sell software to a radiology group and your ceiling is what they will pay in licence fees; be the radiology group and you collect the reimbursement for the read itself. For a decade labour cost blocked that path; AI now compresses the marginal human minutes per report, and the same reimbursement turns into margin. The supply side really is buckling: the ACR's February workforce update finds radiologists leaving practice entirely at more than twice the rate of under a decade ago, subspecialists 37% likelier to exit than generalists, and pediatric radiologists down from 2,190 in 2016 to 2,032 in 2023.
Two outlets report two numbers and the company has not broken them out. More to the point, the public record contains no measured figure at all for reading speed, accuracy or radiologist throughput — "dramatically accelerate interpretations" is the company's own phrasing, with no third-party or peer-reviewed backing, and no disclosure of how the radiologists are paid, which is precisely what decides whether this model can keep them.
Nara Health's $14M: Khosla is not backing insurtech, it is backing an AI-native third-party administrator
Announced September 14: $14M across pre-seed and seed, led by Khosla Ventures, with Long Journey Ventures and Superior Studios. Nara is a TPA for self-insured employers: it pairs alternative plan designs — direct provider contracting, reference-based pricing, Direct Primary Care — with agentic care coordination. The company reports 25,000+ covered lives, $600M in claims processed to date, a five-second average member-support response, same-day prior authorisation against an industry norm of three to five business days, and a 55% year-over-year cost reduction at its client AMPS.
170 million Americans are covered through an employer and the average family premium runs about $27,000 a year, so the administrative friction is itself a large business — one that until now only incumbent TPAs could reach. Nara's bet is that once adjudication, eligibility and prior authorisation can be handed to agents, a new entrant can bid for the work directly at a fraction of the incumbent's headcount, instead of first selling them a tool. Same script as Epsilon, moved to the payer side.
Every figure above is self-reported and not independently audited, and 25,000 covered lives is a very small book in the US self-insured market. The 55% reduction is one client (AMPS); the public record does not give the baseline year, risk adjustment, or how much of the drop came from changing the plan design. How much of that saving is AI and how much is simply steering members out of an incumbent network, the release does not separate.
NVIDIA takes a stake in Verily: the accelerated-computing vendor now owns part of the platform running on it
On September 9 Verily announced that NVIDIA had come on as a new strategic equity investor, with a follow-on from CU Healthcare Innovation Fund II, extending the $300M round Verily closed in March; the amount was not disclosed. The technical coupling is deep: the Verily Pre platform turns fragmented clinical assets into FHIR-native, training-ready datasets and runs on NVIDIA B200 GPUs, NeMo, Parabricks and RAPIDS to deploy Verily's Forecast 1.0 multimodal foundation model.
Verily spun out of Alphabet earlier this year, and its hardest commercial problem is convincing academic medical centres that it is neutral. Putting NVIDIA on the cap table buys priority access to next-generation hardware at the price of binding to one vendor — and avoiding exactly that lock-in is what hospital procurement cares about most. The deal also marks how far the chip vendor's healthcare strategy has moved: from selling cards and SDKs to holding equity in the data platform itself.
With no amount disclosed there is no way to tell whether NVIDIA's stake is symbolic or material. Forecast 1.0 has no published external benchmark or clinical endpoint yet; at this stage "multimodal foundation model" is still roadmap language.
Forus raises $150M at a $3B valuation, tripled in a year, by owning the whole stretch between prescribing and dispensing
On September 8, Bain Capital Ventures led a $150M Series C at a $3B post-money valuation, triple the Series B, with Thrive, General Catalyst, Accel, Redpoint and other existing backers all re-upping; total raised now exceeds $300M. The product links physicians, pharmacies, payers and biopharma, automating prior authorisation, appeals, financial-assistance routing and pharmacy selection, free to doctors and patients. Reported metrics: reach into 85% of US residential ZIP codes, adoption by more than a third of the country's dermatology providers, work with nine of the top 15 global biopharma companies, and a client — Family Allergy & Asthma — whose median prior-auth turnaround fell from over seven days to 1.1.
"Free to doctors and patients" is the whole business model: biopharma pays, because every prescription that does not die in prior authorisation is a real sale. That ties Forus's revenue to pharma marketing budgets rather than hospital IT budgets — the latter being what crushed a decade's worth of health SaaS. The $3B is paying for that position, not for the model.
Every adoption and turnaround figure is company-supplied and unaudited, and dermatology penetration does not extrapolate to other specialties. A deeper conflict goes unaddressed: when biopharma pays, "helping a patient get the drug" and "helping the manufacturer sell the drug" are the same operation, and nothing public explains how Forus keeps pharmacy selection and assistance routing from favouring the payer's product.
Implicity's $40M (€35M): the only round this week that brings a peer-reviewed mortality number with it
On September 9 the French cardiac remote-monitoring company Implicity closed $40M in growth equity led by IRIS, with Five Arrows of Rothschild & Co participating. The platform spans the US, France and Germany, serving over 250 medical centres and monitoring more than 120,000 patients daily. European coverage denominates the round at €35M. Two algorithms hold FDA Class II clearance: SignalHF for heart-failure prediction, and the IM007 ILR ECG Analyzer, which cuts false alerts by up to 76%.
In a week of self-reported metrics, Implicity is the one round anchored to peer-reviewed literature: patients monitored on the platform showed 26% lower mortality than on conventional remote monitoring, from Varma et al. in Heart Rhythm, 2025 (doi:10.1016/j.hrthm.2024.11.029). Set against the other six, it shows capital is currently willing to pay for both clinical endpoints and pure operating metrics — and paying more for the latter: Forus's $150M is nearly four times this round.
The study is observational, comparing populations under different monitoring systems rather than randomising them; the 26% should not be read as a causal claim that switching platforms lowers mortality. The company disclosed no revenue or growth figures.
Inspiren's $70M Series C at over $500M: capital starts treating senior living as a scalable AI market
Announced September 10: a $70M Series C led by NewView Capital with Insight Partners, Primary Venture Partners, Scale Venture Partners and Avenir Growth among others, bringing total funding to $225M at a valuation above $500M — reported as the highest for a technology company built specifically for senior-living operators. Founder and chief clinical officer Michael Wang: "a decade from now, no one will accept that a resident's needs could go unknown."
Senior living has been the least-invested corner of healthcare AI: fragmented payers, thin operator margins, poor data. A $500M-plus valuation means at least one set of investors believes the staffing shortage here bites harder than it does in hospitals — hard enough to support pricing. It also fits this issue's through-line: Inspiren is not selling diagnosis, it is selling coverage where the caregivers are not.
Public coverage gives no customer count, community count, revenue or retention, and "above $500M" is a band rather than a figure. Continuous sensing in a residential care setting also carries privacy questions the coverage does not touch.
Put those seven back into the half-year denominator: $7.4B, 244 deals, 19 companies taking 45% — and still zero IPOs
Rock Health's H1 2026 report: $7.4B across 244 deals, against 245 deals and $6.4B a year earlier, with the median deal rising from $12M to $14M. Nineteen companies took 20 mega deals of $100M or more, absorbing 45% of the half-year's capital — up from 22% in 2024 and 42% in 2025 — and 13 of those 19 were first-time recipients at that scale. There were 115 acquisitions, with Q2's 71 the busiest quarter since Q3 2021, and revenue cycle management the most consolidated category of all. IPOs: zero. The only S-1 filed all half-year came from the wearable maker Oura.
Those three numbers together are this issue's structure: more money, concentrated in fewer hands; acquisition as the only exit; and the hottest acquisition category being administration and billing. Which means the companies that spent last week becoming institutions are betting on one thing — with the IPO window shut, what gets bought is not a model but a running, cash-generating, licensed operating asset.
Rock Health counts US deals only, and only disclosed rounds above $2M, so European companies such as Implicity and a great many undisclosed transactions sit outside the denominator. The report published on July 13 and contains no Q3 data.
02 — Product Analysis
Epsilon Health
AI-native radiology practice · Epsilon Health (US)
Function and position. Not an imaging-AI vendor but a practice employing its own board-certified radiologists, with purpose-built AI embedded in its own reading workflow, taking on interpretation volume for hospitals and imaging centres. Revenue comes from the read, not from a licence.
- Strength : it routes around the hardest gate in healthcare AI — no procurement committee to convince, no IT integration queue, no fight for a separate AI reimbursement code. Interpretation volume already has a mature payment path, and the supply gap is real (ACR: attrition at more than twice the rate of a decade ago).
- Strength : it is not bound to one disease or one modality, which is the sharpest break from conventional imaging-AI products; expanding coverage does not mean running the regulatory and sales cycle again per indication.
- Concern : there is no verifiable performance figure at all. Nothing public gives turnaround per report, reads per radiologist shift, or any comparison against a control; "accelerating interpretations" is the company's own word. For a company whose whole thesis is a productivity delta, that is the missing cell that matters most.
- Concern : the model buys the liability too. A missed finding is the practice's malpractice exposure, not the purchasing hospital's as it would be for a vendor; $27.6M has to carry radiologists, models and insurance at once, and the unit economics at scale are untested.
Forus
Prescription-workflow intermediary · Forus (US)
Function and position. It sits between physicians, pharmacies, payers and biopharma, automating prior authorisation, appeals, assistance routing and pharmacy selection, free to clinicians and patients and paid for by manufacturers, and closed a $150M Series C at a $3B valuation on September 8.
- Strength : the payer choice is excellent. Pharma books this under marketing, not IT; the ROI on one more filled prescription is direct and fast, and it is insulated from hospital capex cycles. Nine of the top 15 global biopharma companies are already customers.
- Strength : it has one operating metric that is quantified and checkable in the field — median prior-auth turnaround at Family Allergy & Asthma falling from over seven days to 1.1. That is days, not accuracy: a clinic can verify it themselves, which makes it much harder to inflate than most healthcare-AI claims.
- Concern : the structural conflict is unexplained. An intermediary paid by manufacturers also decides which pharmacy and which assistance programme a patient is routed to, and nothing public describes algorithmic neutrality, disclosure duties or an audit mechanism.
- Concern : the $3B rests on a policy-sensitive gap. If prior authorisation is substantially simplified by reform — which several US states and federal proposals have been pushing — the friction Forus monetises shrinks directly. No revenue is disclosed, so the multiple cannot be assessed.
03 — Companies & Competition
| Company | Recent state & numbers | Position & moat |
|---|---|---|
| Epsilon Health AI-native radiology practice |
Out of stealth 9/10 with $27.6M led by AlleyCorp; Axios reports a $20M Series A. Founded 2024 by Rustin Rassoli. | The moat is being the billing entity itself, with no AI reimbursement code to fight for. The weakness is scale: Radiology Partners fields over 3,000 radiologists. |
| Radiology Partners PE-backed radiology consolidator |
Acquired Everlight Radiology on 8/25 at roughly $1B enterprise value; serves 3,400+ hospitals with over 3,000 radiologists and about 40 million exams a year, near 10% of US imaging volume. | Plays the same AI-productivity card (MosaicOS claims 20–35% shorter reporting time) but buys its volume. The weak spots are leverage and commercial-payor litigation exposure. |
| Forus Prescription intermediary |
Closed a $150M Series C at a $3B valuation, tripled, on 9/8; over $300M raised in total, reaching 85% of residential ZIP codes and more than a third of dermatology providers. | The moat is distribution bought by making it free to clinicians while manufacturers pay. The weakness is policy: prior-auth simplification thins the very friction it sells against. |
| Nara Health AI-native TPA |
Raised $14M led by Khosla on 9/14; self-reports 25,000+ covered lives, $600M of claims processed and same-day prior authorisation. | No moat yet: that book is tiny in the self-insured market, and the real barrier is contracted networks and actuarial data, not the model. |
| Verily Newly independent precision-health data platform |
NVIDIA took an undisclosed stake on 9/9, extending March's $300M round; the Pre platform runs on B200, NeMo and Parabricks. | The moat is the engineering that turns clinical assets into FHIR-native, trainable datasets. The weakness is neutrality: binding to one hardware vendor undercuts its pitch to academic medical centres. |
| Commure General Catalyst's RCM roll-up |
A $70M round at a $7B valuation, plus $200M of growth financing from CVF. | The moat is acquisition speed plus General Catalyst's health-system relationships. It is the most likely acquirer for RCM startups such as GenHealth and Arintra. |
| Implicity Cardiac remote monitoring (France) |
$40M led by IRIS on 9/9; 250 medical centres, 120,000 patients monitored daily, two FDA Class II clearances. | The moat is vendor-neutral implant data aggregation plus a published clinical endpoint. The weakness is competing head-on in the US with the device makers' own monitoring platforms. |
| Tempus AI Listed precision-oncology company |
Announced a $1.5B acquisition of Personalis on 7/20, targeting MRD testing in what it sizes as a $20B market. | One of the few listed comparables: the moat is testing volume and payer coverage. It demonstrates that the only exits here today are acquisition, or consolidation by those already public. |
Close it in one line: this week's competition is not about whose model is better, it is about which cell of the payment chain you occupy. Epsilon stands in the reimbursement for the read, Nara in the administrative fee, Forus in the pharma marketing budget, Verily in the data infrastructure, Inspiren in the senior-care staffing gap — and not one of the five leads with accuracy. In a half-year with zero IPOs, 115 acquisitions and RCM the hottest category, that is entirely rational: what gets bought is a position with cash flow, not a benchmark score.
04 — Taiwan Angle
(1) Nara's and Forus's business models have almost no market in Taiwan. Both monetise waste specific to American multi-payer insurance: prior authorisation taking three to five business days, $27,000 average family premiums, network mark-ups stacked on mark-ups — and compressing seven days to 1.1 is enough to carry a $3B valuation. Taiwan's single payer squeezed most of that friction out long ago, so a Taiwanese team copying the administrative-automation playbook will find the spread it was meant to harvest is not there. The money here belongs at the other end: NHI claims review, note writing, referral coordination — high-volume, very low unit price, which demands coverage-driven pricing rather than the American share-of-savings model.
(2) The Verily deal is the comparison Taiwan's Ministry of Health should be reading against its "333" policy. What NVIDIA bought into is not a model but the Pre platform's ability to turn fragmented clinical assets into FHIR-native, AI-ready datasets — the operative word being ready, not merely interoperable. Minister Shih Chung-liang's 333 policy targets record interoperability across all medical centres by year end, extension to district and regional hospitals over the following two years, and a standard system for clinics, backed by the NT$48.9 billion Healthy Taiwan Deep Cultivation Plan. If that timeline delivers only "hospital A can view hospital B's chart", the output is readable documents. Supporting domestic model training needs a structurally consistent, annotatable, reliably extractable data layer. The engineering costs differ by an order of magnitude, and the budget lines and acceptance criteria being written now decide which one Taiwan ends up with.
(3) Epsilon's problem exists in Taiwan; its solution does not. Radiology staffing strain is international, and the subspecialty attrition the ACR documents — a 37% higher exit rate — has its Taiwanese counterpart in unfilled overnight and rural reading shifts. But Taiwan's rules on who may establish a medical institution do not accommodate a venture-funded company simply opening its own AI-native practice, so the same gap has only two routes: in-house reading support, or a hospital contracting out to a compliant remote reading service. Which means Taiwanese vendors cannot reach for Epsilon's leverage of being the billing entity, and are left with the hardest path of all — convincing hospital procurement and IT.
05 — Further Reading
-
H1 2026 funding and market overview: Durable roots, shifting routes — Rock Health (2026-07-13)
The denominator for every individual deal in this issue. Read the "45% of capital to 19 companies" figure against "RCM is the most consolidated category" and watch them explain each other.
-
Global Consolidation in AI-Enabled Teleradiology: Radiology Partners' $1B Acquisition of Everlight — healthcare.digital (2026-08-25)
Epsilon's mirror image. Same radiology shortage, answered on one side by founding a practice with venture money and on the other by paying $1B for a follow-the-sun network of 800 radiologists in 40 countries — the cost structures are worth reading side by side.
-
The Radiologist Shortage: A Workforce Update from HPI — ACR Bulletin (2026-02-05)
The shared premise behind both radiology-adjacent cheques this week. It does not give a headcount gap but something more useful: how average practice size went from 9.7 radiologists to 17.9, and how radiology-only practices shrank by about a third.
-
The AI virtual-care consolidation wave — Becker's Hospital Review (2026-07-31)
It reconciles the contradiction of deal volume at a decade low while median size rose 54% year to date, and why buying data and networks beats building them — the strategic premise under all five companies in this issue.
-
Nara Health Raises $14M to Rebuild Self-Insured Health Plans via Agentic AI — HIT Consultant (2026-09-14)
The smallest round here, but the clearest full statement of the AI-native services argument: it itemises what replacing an incumbent TPA requires, leaving the reader to judge which items are genuinely model-driven and which are just a different plan design.
06 — References
- Fierce Healthcare Fundraising Tracker '26: Arintra lands $25M; Happy Health buoyed by $75M round. Fierce Healthcare, 2026-09-10. fiercehealthcare.com
- Epsilon Health, which contracts with radiologists who use its AI to generate image reports faster, emerges from stealth with a $20M Series A led by AlleyCorp (Axios). Techmeme, 2026-09-12. techmeme.com
- San Francisco startup seeks to solve radiologist shortage with new 'AI-native' imaging group. CAPPS, 2026-09. cappsonline.org
- Nara Health Raises $14M to Rebuild Self-Insured Health Plans via Agentic AI. HIT Consultant, 2026-09-14. hitconsultant.net
- Verily Secures Strategic Investment from NVIDIA to Accelerate Precision Health AI Platform. HIT Consultant, 2026-09-09. hitconsultant.net
- Verily secures new investment from Nvidia. R&D World, 2026-09. rdworldonline.com
- Verily banks $300M to accelerate AI road map, transitions to independent company. Fierce Healthcare, 2026-03. fiercehealthcare.com
- Verily Health Secures New Investment | Inspiren Raises $70M Series C. Healthcare IT Today, 2026-09-14. healthcareittoday.com
- Inspiren Rakes In $70M to Scale Its AI-Powered Senior Living Platform. MedCity News, 2026-09. medcitynews.com
- Forus triples valuation to $3B with new $150M funding round. Fierce Healthcare, 2026-09-08. fiercehealthcare.com
- Implicity secures $40M growth equity funding to scale its AI-driven cardiac monitoring platform globally. GlobeNewswire, 2026-09-09. globenewswire.com
- Implicity raises $40M for AI remote cardiac monitoring. MobiHealthNews, 2026-09. mobihealthnews.com
- French cardiac monitoring startup Implicity reports 26% lower mortality as it raises €35 million. EU-Startups, 2026-09. eu-startups.com
- GenHealth.ai raises $16.5M to expand healthcare AI agents. MobiHealthNews, 2026-09-08. mobihealthnews.com
- GenHealth.ai Raises $16.5M to Automate Healthcare Administrative Tasks. MedCity News, 2026-09. medcitynews.com
- H1 2026 funding and market overview: Durable roots, shifting routes. Rock Health, 2026-07-13. rockhealth.com
- Global Consolidation in AI-Enabled Teleradiology: Strategic Analysis of Radiology Partners' $1 Billion Acquisition of Everlight Radiology. healthcare.digital, 2026-08-25. healthcare.digital
- The Radiologist Shortage: A Workforce Update from HPI. ACR Bulletin, 2026-02-05. acr.org
- The AI virtual-care consolidation wave. Becker's Hospital Review, 2026-07-31. beckershospitalreview.com
- Tempus AI to acquire cancer genomics firm for $1.5B. Becker's Hospital Review, 2026-07-20. beckershospitalreview.com
- AI company Commure banks $70M funding round, hits $7B valuation. Fierce Healthcare, 2026-05. fiercehealthcare.com
- Commure Raises $200M in Growth Financing from General Catalyst's CVF to Accelerate AI-Powered RCM Platform. Commure, 2026. commure.com
- 高醫大論壇揭示AI醫療新局！衛福部推「333政策」 國家489億預算力挺. 聯合新聞網, 2026-06-27. udn.com
- Arintra Secures $25M to Scale GenAI Autonomous Medical Coding. HIT Consultant, 2026-08-24. hitconsultant.net