Growth Equity Investment in Healthcare AI Companies

Sector Definition and Market Structure

Scope of the Mandate

A healthcare AI company, as LeverVenture uses the term, is a company in one of the five sectors of the mandate (Biopharma and Therapeutics, Diagnostics and Precision Medicine, Devices and Robotics, Digital Health and Delivery, and Longevity and Neuro) whose product or operating model depends on a machine-learned model. The model is a layer inside the sector, and the sector's regulator, payer and buyer still decide the outcome. Crossing sectors is the thesis. A healthcare-driven company that is also deep tech in its engineering, that is also digital in its data, or that carries a clean tech dimension, with an artificial intelligence layer that makes it work, is the company we look for, and we underwrite the whole company as a single object.

The Authorized Device Record

FDA's List of Artificial Intelligence-Enabled Medical Devices, content current as of September 4, 2026, contains 1,614 entries: 1,553 cleared through 510(k), 40 granted through De Novo and 21 approved through premarket approval. On the same list, radiology is the lead review panel for 1,230 entries, cardiovascular for 154 and neurology for 73. The list carries 335 devices with a final decision date in 2025, compared with 235 in 2024 and 226 in 2023. FDA states that the list is not a comprehensive resource of AI-enabled devices, so these counts measure an identified floor of the market.

In drug development, FDA's Center for Drug Evaluation and Research reports a significant increase in drug application submissions using AI components, spanning the nonclinical, clinical, postmarketing and manufacturing phases of the drug product life cycle.

Capital Formation

Rock Health's 2025 year-end overview reports $14.2 billion of venture funding for U.S. digital health startups in 2025, with AI-enabled companies capturing 54 percent of total funding, up from 37 percent in 2024, and commanding a roughly 19 percent premium in average deal size. Rock Health's H1 2026 overview reports $7.4 billion raised across 244 deals in the first half of 2026, and it states that Rock Health stopped labeling startups as AI-enabled because AI has become sufficiently ubiquitous that it no longer distinguishes a product or strategy.

The same 2025 overview counts 195 digital health acquisitions in 2025, with digital health companies as the acquirer in 66 percent of them and private equity firms in 10 percent. In short, capital for healthcare AI comes from venture investors, strategic acquirers and private equity alike, and the label alone no longer differentiates a company; the record behind the model does.

Investment Criteria and Underwriting

Stage and Position

LeverVenture invests at the commercial inflection, between venture capital, which underwrites whether the science can work, and private equity, which underwrites the optimization of a business that already works. For a healthcare AI company, that stage means the model's technical risk is substantially retired, the first real buyers exist, and the next constraint is commercial rather than scientific. We hold minority positions, and we lead or co-lead where the company wants an operator at the table.

The LeverRating Applied to the AI Layer

Every opportunity is scored on the six dimensions of the LeverRating: Team, Market, Product, Traction, Financial and Thesis fit. For a device, a diagnostic or a therapeutic, the house framework expands to ten dimensions, so that regulatory position, clinical evidence, reimbursement and freedom to operate each carry their own weight. An AI layer changes what each of those four must prove.

Regulatory position asks whether the authorized indication matches the marketed one and whether a predetermined change control plan covers the retraining the operating plan assumes. Under section 515C of the Federal Food, Drug, and Cosmetic Act (21 U.S.C. 360e-4), a change consistent with an authorized plan requires neither a supplemental premarket approval application nor a new premarket notification. Clinical evidence asks whether performance held on data from sites other than the development sites. Reimbursement asks who pays for the output, which is a different question from who licenses the software. Freedom to operate reaches the training data, including documented rights to each dataset and a lawful basis under 45 CFR 164.514 or 164.508 for any protected health information.

Regulatory and Payment Gates

As of October 2026, FDA's guidance on predetermined change control plans for AI-enabled device software functions is final, in an August 2025 version with content current as of August 18, 2025. Its lifecycle guidance for AI-enabled device software functions, dated January 7, 2025, and its guidance on AI used to support regulatory decision-making for drugs and biological products, dated January 2025, both remain drafts. The Quality Management System Regulation at 21 CFR Part 820 took effect February 2, 2026 (89 FR 7496).

As of October 10, 2026, no FDA rule treats a laboratory-developed test as a device: a federal court vacated FDA's May 6, 2024 rule on March 31, 2025, and FDA restored the prior text of 21 CFR 809.3(a) effective September 19, 2025, so a laboratory-developed test answers to the CLIA certification conditions of 42 CFR Part 493. Decision support in certified health information technology remains governed by 45 CFR 170.315(b)(11) as revised by the HTI-1 final rule (89 FR 1192). A proposed rule published December 29, 2025 (90 FR 60970) would revise that criterion, and no final rule had published in the Federal Register as of October 10, 2026.

On payment, CMS's CY 2027 physician fee schedule proposed rule (91 FR 43842, July 16, 2026) proposes to rename the algorithm-driven services it previously called Software as a Service as Software as a Medical Service, and it notes that certain such analyses performed on laboratory tests are paid today as clinical diagnostic laboratory tests. No final CY 2027 physician fee schedule rule had published in the Federal Register as of October 10, 2026. Performance claims carry their own exposure, because section 5 of the FTC Act (15 U.S.C. 45) declares unfair or deceptive acts or practices unlawful, and the FTC's 1984 policy statement requires a reasonable basis for an objective advertising claim before its initial dissemination.

For a company that sells into Europe, Regulation (EU) 2024/1689, the EU Artificial Intelligence Act, is a comparative gate. Its Annex I lists the Medical Devices Regulation (EU) 2017/745 and the In Vitro Diagnostic Medical Devices Regulation (EU) 2017/746, and the European Commission states that the rules for high-risk AI systems embedded in Annex I products have a transition period until August 2, 2028, following the AI Omnibus amendments that entered into force on July 27, 2026.

ROI² and the Impact Measure

ROI² underwrites the second return on the same evidence as the first. For an AI layer that claims to take cost out of a care pathway, the thesis requires that a payer or a provider be measurably better off. One impact measure is named at entry with a baseline, an owner and a reporting cadence, and it reports alongside revenue, gross margin and cash.

Artificial Intelligence as the Accelerant

The accelerant changes the underwriting in three ways. First, it compresses timelines, and we credit that compression only when it appears in a dated record such as an authorization, a coverage decision or a signed contract. Second, it changes the durability question, because a model that changes after authorization either sits inside an authorized change control plan or turns each update into a regulatory event.

Third, it moves the moat. Rock Health's H1 2026 overview reports that investors and buyers now ask which company has something AI alone cannot provide, and our underwriting answers with the combination of a regulatory record, documented data rights and a clinical workflow that a competitor cannot rebuild from a general-purpose model. The model is rarely the asset by itself; the record around it is.

Reference Entries

• AI Diligence in Life Sciencesthe evidence-based assessment of whether an AI claim is supported by its data, validation, regulatory record and quality system, with the full diligence question set.
• Software as a Medical Device (SaMD)the definition of software with a medical purpose that runs outside a hardware device, and the 510(k), De Novo and premarket approval routes it can take.
• Laboratory-Developed Tests and CLIAhow a test performed within a single laboratory is regulated under CLIA after the 2025 vacatur of FDA's rule, and why reimbursement evidence is often the binding constraint.
• Management Services Organizationthe administrative company that serves a physician-owned practice under a written agreement, and the state laws enacted in 2025 that now limit investor control of that structure.

Related Analysis

• Growth EquityGrowth Equity for AI Healthcare: The 2025 Playbooka founder-level guide to how growth equity investors assess AI healthcare companies, from diligence frameworks and valuation drivers to reimbursement risk and exit pathways.
• Life Sciences and HealthcareFDA Digital Health Guidance: Clarity Accelerates Innovationan analysis of how the predetermined change control plan lets an AI-enabled device keep learning after clearance, and what that change means for underwriting.
• Life Sciences and HealthcareDigital Health: From Telemedicine to AI Diagnosticsan argument that model accuracy is table stakes and that prospective validation, workflow integration and liability for a missed finding decide which AI diagnostic companies endure.

Governing Authority and Sources

• FDA, List of Artificial Intelligence-Enabled Medical Devices (content current as of September 4, 2026)
• FDA CDER, Artificial Intelligence for Drug Development (content current as of May 1, 2026)
• FDA, Predetermined Change Control Plan for AI-Enabled Device Software Functions, final guidance (August 2025)
• FDA, AI-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations, draft guidance (2025)
• FDA, Considerations for the Use of AI to Support Regulatory Decision-Making for Drug and Biological Products, draft guidance (2025)
• FD&C Act section 515C, 21 U.S.C. 360e-4 (2022)
• FDA, Quality Management System Regulation, 89 FR 7496 (2024), effective February 2, 2026
• FDA, Laboratory Developed Tests (2025)
• FDA, Medical Devices; Laboratory Developed Tests, final rule restoring 21 CFR 809.3(a), 90 FR 45134 (2025)
• ONC, HTI-1 final rule, 89 FR 1192 (2024)
• 45 CFR 170.315(b)(11), decision support interventions (current 2026)
• ASTP/ONC, Deregulatory Actions to Unleash Prosperity, proposed rule, 90 FR 60970 (2025)
• CMS, CY 2027 Physician Fee Schedule proposed rule, 91 FR 43842 (2026)
• 45 CFR 164.514, de-identification and limited data sets (current 2026)
• 45 CFR 164.508, authorizations (current 2026)
• FTC Act section 5, 15 U.S.C. 45 (current 2026)
• 42 CFR 493.1, CLIA laboratory requirements (current 2026)
• FTC, Policy Statement Regarding Advertising Substantiation, 49 FR 30999 (August 2, 1984)
• European Commission AI Act Service Desk, Regulation (EU) 2024/1689, Annex I (2024)
• European Commission, AI Act regulatory framework and AI Omnibus timeline (2026)
• Rock Health, 2025 Year-End Digital Health Funding Overview (2026)
• Rock Health, H1 2026 Funding and Market Overview (2026)