Where Healthcare AI Investment Is Going — And Where It Isn't

A study of 3,807 AI healthcare startups shows two-thirds of investment, totalling $63B, went to just three domains: Clinical decision support, drug discovery, and medical imaging.

This is the most detailed map of the healthcare AI startup landscape to date.

The AI health startups founded between 2010 and 2024 were classified by medical domain, AI complexity tier, funding raised, founding team composition, and geography.

The numbers are illuminating about where capital has gone. Even more so, they are revealing about what has been neglected and remains underfunded.

Image

The Five-Tier Complexity Framework

The five-tier framework classified healthcare AI systems by the sophistication of the underlying technology.

The tiers ran from assistive AI — rule-based tools that support administrative tasks without clinical inference — through perceptual AI, analytical AI, and interactive AI, all the way up to autonomous AI systems capable of independent clinical decision-making.

The framework is important because it separates the complexity of the AI system from the complexity of the clinical problem it addresses.

For example, a highly complex autonomous AI applied to a domain with poor data infrastructure and limited regulatory clarity will not attract capital regardless of its technical ambition.

The central finding from this framework analysis is that investment follows data structure, regulatory clarity, and clinical adoption barriers — not AI sophistication.

Higher AI complexity does not predict higher funding. The domains that raised the most capital are not those with the most technically advanced tools; they are those with the most legible data, the clearest regulatory pathway, and the lowest barrier to clinical uptake.

Where the $63 Billion in Healthcare AI Investment Went

Total venture investment across 3,807 startups reached USD 63 billion (2010 through 2024), while the distribution across domains is highly concentrated.

Clinical decision support, drug discovery, and medical imaging together captured roughly two-thirds of all investment.

Drug discovery alone raised USD 18.5 billion — the single largest domain by funding — driven by the scalability of molecular and genomic data, the established regulatory pathway for novel drug entities, and the willingness of large pharmaceutical partners to fund early-stage AI platforms through licensing and co-development arrangements.

Medical imaging and clinical decision support attracted substantial capital for related reasons. Both domains operate on structured, high-volume data (radiology images, electronic health records, laboratory results) that AI systems can process at scale

Both have relatively clear regulatory pathways through the FDA's AI-enabled device framework. And both connect to workflows within health systems that already have purchasing infrastructure in place.

The domains that did not attract capital tell a much more interesting and concerning story.

Mental health, public health, and rehabilitation raised a combined USD 1.95 billion across the 14-year period — a figure that represents only approximately three per cent of total investment despite these areas accounting for a substantial share of global disease burden.

How can this be explained? Scalability limitations, fragmented data, and the absence of clearly reimbursable AI-driven outputs make these domains structurally unattractive to venture capital, irrespective of clinical need.

Image

The FDA Approval Picture: Depth Concentrated at the Low End

We have to look at U.S. Food and Drug Administration (FDA) approval data for AI-enabled medical devices as a proxy for the regulatory maturity of the sector.

The FDA has approved 1,016 AI-enabled medical devices since 1995.

More than 97% fall into the two lowest complexity tiers of the five-tier framework — assistive and perceptual AI. These are tools that automate documentation, flag abnormalities in images, or assist with structured data classification. They are genuinely useful. But they are not the autonomous clinical intelligence that much of the sector's promotional narrative describes.

The concentration of approved devices at the lower complexity tiers reflects the regulatory challenge of validating higher-tier systems.

Autonomous AI that makes independent clinical recommendations requires a fundamentally more demanding evidence base than a tool that highlights a potential finding for a human reviewer to evaluate.

The FDA's growing body of guidance on AI-enabled devices — including its draft guidance on AI for regulatory decision-making published in January 2025 — is developing the framework for higher-complexity approvals.

But the approval data through 2024 shows that most commercially viable healthcare AI, as evaluated by regulators, operates in assistive and perceptual domains.

Who Is Building Healthcare AI: Clinical Under-Representation

The data on founding team composition of the healthcare AI startups is most significant for the long-term trajectory of the field.

Clinical practitioners — physicians, nurses, allied health professionals — make up less than 5% of founding teams across the entire dataset.

This ratio holds at every funding level. It is not a feature of underfunded startups; it is consistent across seed, early-stage, and late-stage companies.

Technical-only or technical-plus-business founding teams lead 35% of all startups in the dataset; business-only founders lead a further 16.6%, and clinician-led teams, defined as those where a clinical practitioner holds a founding role, sit below five percent.

The practical implication is that the majority of healthcare AI products are being designed by teams without direct clinical experience of the workflows, edge cases, and human factors that determine whether a tool is adopted or abandoned in practice.

Co-founders are a prerequisite for useful products, but this identified clinical under-representation is a structural feature of the ecosystem that shapes which problems get addressed and how solutions are designed — and, equally important, which problems remain unaddressed.

This connects directly to the domain funding gap. Mental health, rehabilitation, and public health are domains where the patient journey is complex, the data is unstructured, and the clinical workflow is highly variable across providers and settings.

These are exactly the domains where clinical insight in the founding team would be most valuable for navigating the design problem. However, they are the domains that capital has most consistently bypassed.

The Healthcare AI Startup Gender Data

The gender composition of healthcare AI founding teams reflects broader structural inequalities in the technology sector, with some sector-specific features of note.

Eighty-four per cent of founders across the dataset were male.

Women account for 10.7% of technical founders specifically (the founding background that dominates the sector and that correlates most strongly with funding success). This means women are underrepresented precisely in the credential category that capital most rewards in healthcare AI.

This shows a team formation gap that compounds the funding disadvantage. Seventy-four per cent of female-only founding teams are solo-founded (a single founder operating without co-founders), compared to 40% for male-only founding teams.

This can be attributed to structural barriers in professional network access rather than individual preference: Women in technical and clinical fields have smaller professional networks that intersect with the venture-adjacent talent pools from which co-founders are typically recruited.

Solo-founded companies raise less capital, take longer to scale, and have narrower skill coverage in the founding team.

The team formation gap therefore creates a compounding disadvantage: The structural barrier to co-founder recruitment produces the funding gap, rather than the funding gap being the primary cause of female under-representation.

Image

The Geography of Healthcare AI Capital

The U.S. accounts for 1,609 of the 3,807 startups in the dataset and more than USD 38 billion of total funding — approximately 60% of global capital concentrated in a single country.

Within the U.S., California and Massachusetts together account for approximately 60% of U.S.-based deals, a concentration pattern consistent with broader venture capital geography.

Africa and South America barely register in the dataset.

This reflects the concentration of venture capital infrastructure, research institutions, and regulatory capacity in high-income countries — and the absence of all three in lower-income markets where AI-enabled health tools could address the greatest absolute clinical gaps.

The dataset does not capture public sector or development finance investment in lower-income markets, which limits the picture, but the venture-funded startup ecosystem is overwhelmingly concentrated in geographies that already have sophisticated health systems.

What Decision-Makers Should Take From This Data

For pharma and health systems leaders assessing the healthcare AI landscape, three implications from the study are strategically relevant.

First, capital concentration in clinical decision support, drug discovery, and imaging means these are the domains with the most developed commercial ecosystems, the most vendor competition, and the most mature evidence bases for evaluating AI tools.

Procurement and partnership decisions in these domains can reference a growing body of real-world implementation evidence. Domains outside this cluster carry higher implementation uncertainty but lower commercial competition.

Second, the absence of clinical founders from most healthcare AI startups is a product quality signal as much as an equity concern.

Tools designed without clinical co-founders embedded in the founding team are statistically more likely to encounter workflow misalignment, clinician adoption resistance, and unanticipated edge-case failures when deployed in practice

Due diligence processes for AI tool partnerships should assess clinical expertise in the founding and product team, not only technical capability and funding credentials.

Third, the finding that AI complexity does not predict funding has direct relevance to R&D investment strategy. The domains that attract capital are those with structured data, regulatory clarity, and defined adoption pathways — not those with the most ambitious AI architecture.

Organisations investing in building or acquiring AI capability should assess these three factors in any target domain before evaluating the technical sophistication of available tools.

Pharmatica tracks the healthcare AI startup ecosystem, investment patterns, and the evidence base on AI tool effectiveness and adoption, giving pharma, biotech, and health system leaders the rigorous, independent analysis they need to make informed decisions about AI capability building, partnership, and procurement in a rapidly evolving and unevenly funded landscape.

Pharmatica: Insight. Connection. Impact.