Guest Post: The Layer Beneath: Healthcare Data Infrastructure in an Age of Commodity Intelligence

AI is creating a paradox in healthcare IT: the more capable the technology becomes, the less defensible the technology itself may be.

The market has moved quickly from skepticism to ubiquity. In 2025, AI-enabled companies captured 54% of U.S. digital health venture funding, up from 37% the year before.¹ By the first quarter of 2026, Rock Health stopped tracking “AI-enabled” startups as a separate category because AI had become table stakes. The capital kept coming: U.S. digital health startups raised $7.4 billion across 244 deals in the first half of 2026, with 45% of all capital concentrated in 20 megadeals.²

Taken together, those facts point to something unusual in the market. AI is becoming indispensable for health startups at the same time that AI itself is becoming insufficient. This paradox changes where enduring value is likely to accrue.

With increasingly intelligent AI available through an API to anyone and everyone, scarcity, and therefore value, shifts towards assets that have not been democratized: trusted access to data, integration into live workflows, regulatory permission, and network reach. In healthcare, the companies best positioned for the next cycle will not necessarily be the ones with the most impressive model. They will be the ones that control the rails and resources the models need to reach.

That is why we are looking one layer down.

The conventional hierarchy in healthcare technology places applications above infrastructure. Infrastructure is like plumbing: necessary, unglamorous, and presumably lower value than everything built on top of it.

But this hierarchy makes less sense when a new entrant can access frontier intelligence on demand.

This does not mean the application layer disappears. Exceptional applications will still win by owning workflows, earning user trust, producing measurable outcomes, or generating feedback loops that improve with use.

But the model itself is unlikely to provide sustainable differentiation. The private markets are paying up anyway: PitchBook’s healthcare IT update for the first half of 2025 found larger transactions generally exceeding 30x EV/EBITDA, with an explicit AI premium at the top of the market.³ A 30x EV/EBITDA multiple assumes some sort of moat, but Menlo Ventures articulates a view that defensive moats largely come from assets such as regulatory and compliance infrastructure, while generative moats come from compounding data and expanding workflow coverage. The strongest companies have both.⁴

For investors, the relevant question is no longer, “How capable is the AI?” It is, “What remains difficult to replicate once every competitor has access to comparable intelligence?” In healthcare, much of the answer lies in the data.

A clinical decision-support product cannot produce a complete recommendation if it sees only a fragment of the longitudinal record. A revenue-cycle agent cannot automate a transaction end to end if a human must still assemble data from the payer, EHR, clearinghouse, and practice-management system. A care-management tool cannot coordinate across settings it cannot reach.

The models reliably generate answers, but infrastructure access determines whether the model has the inputs, permissions, and connections required to generate the correct answer inside the workflow.

Every AI company that must fight for clean, usable data pays an integration tax albeit one that isn’t obvious in the audited financial statements.

It shows up in long implementations, custom interfaces, services-heavy deployments, delayed time to value, and churn when the product cannot perform reliably in the customer’s environment. The more fragmented the underlying data, the more application economics begin to resemble consulting economics.

Infrastructure companies can invert that dynamic. A network that has already connected the relevant endpoints, normalized their data, established the necessary permissions, and earned customer trust can amortize that work across many applications and transactions. Each additional node can make the network more useful; each additional workflow can deepen its position.

The result is not automatically a great business. Network density, data rights, service quality, and pricing discipline still matter. But a scaled exchange platform begins with an advantage that a newly built application cannot reproduce quickly: it has already paid the integration cost and secured a place in the flow of data.

The pattern is readily apparent in consumer-facing use cases. When OpenAI and Perplexity launched healthcare-specific consumer experiences in 2026, they did not build longitudinal record retrieval from scratch. They worked with health-data platforms to bring clinical records, biometrics, and other user context into their products leveraging their intelligence with access they had to rent.⁵ 

As AI creates more uses for healthcare data, it should increase the value of reliable access to that data. The busier the application layer becomes, the more important the exchange layer becomes beneath it.

Healthcare’s regulatory architecture is often treated solely as friction. In data infrastructure, it can also create defensibility.

TEFCA, the Trusted Exchange Framework and Common Agreement, establishes a common framework for nationwide health-information exchange through Qualified Health Information Networks. The 21st Century Cures Act and its information-blocking rules made electronic information sharing the rule with limited exceptions and consequences for actors that improperly interfere with access, exchange, or use. CMS interoperability requirements have pushed regulated payers toward standardized APIs as well.⁶

These policies do not grant any one company permanent control of the market. But they do raise the cost of participating in a credible manner. Technical conformance, security, consent, governance, trust, and reliable exchange across organizations take time to build. A competitor cannot route around those requirements with a better model or a more elegant interface.

The moat isn’t strictly regulation, but regulation combined with implementation history, customer trust, network density, and reach into difficult settings. That combination can make plumbing unusually hard to displace giving regulated infrastructure the durability of a utility and the growth profile of software.

It also explains why infrastructure assets can attract buyers from outside traditional healthcare IT. We should be explicit about our perspective: AxiaMed, MobileMD, and Kno2 have all been Health Enterprise Partners portfolio companies, and I served on the boards of AxiaMed and Kno2. These examples reflect our own experience and should be weighed accordingly.

When Bank of America acquired AxiaMed in 2021, it was buying its way deeper into provider-side payment infrastructure. When Siemens acquired MobileMD in 2011, it gained a health-information exchange connecting hospitals and physician practices across organizational and technical boundaries.⁷ In both cases, the strategic value extended beyond a discrete product feature. The acquirer gained access to infrastructure embedded in a regulated, networked market.

Kno2 occupies a related position today: an interoperability platform through which applications can reach data across care settings. AI did not create the value of that position. It increases the number of products that depend on it.

Health Enterprise Partners invests with an LP base composed largely of hospital systems and health plans. From inside those organizations, the infrastructure thesis is less of an abstraction.

The operators we speak with cite constraints largely outside of model performance including data readiness, workflow redesign, governance, and integration with the systems already in place. At the same time, vendor consolidation is accelerating. In our 2026 executive survey, 73% of payers and providers said they intended to reduce vendor count, with cuts concentrated in patient engagement, workforce management, and population-health analytics.⁸

That combination matters. When buyers consolidate application vendors, they favor products that can span more of the workflow and work with the existing exchange fabric. They are less interested in another isolated point solution, however impressive its demo. Infrastructure that makes multiple applications usable can become more important as the number of applications shrinks.

Infrastructure is not simply a cost center that better models will optimize away. It is the layer whose value can rise when AI adoption succeeds, because every successful application creates more demand for governed, normalized, permissioned data.

The strongest objection is platform capture.

Epic has the largest clinical footprint in the country and participates in nationwide exchange. Oracle and the hyperscalers have distribution, capital, and technical resources that make independent networks look small. Any infrastructure thesis must account for the possibility that these platforms absorb the economics.

There are two reasons independent networks can still matter. First, healthcare remains a multi-vendor market, and nationwide exchange is designed to operate across organizational and technical boundaries. A network does not need to displace Epic; it needs to connect the places no single system reaches. Second, some of the hardest and most valuable exchange problems sit outside the acute-care core: in post-acute, behavioral-health, and home-based settings where data is fragmented and dominant EHR footprints are thinner.

The more serious risk is commoditization. Regulation may create utility-like durability while also producing utility-like pricing. Standardization can lower barriers for customers and reduce what any network can charge for basic transport.

The best infrastructure businesses will need more than connectivity. They will need differentiated reach, normalized data, high-value workflows, strong service levels, or network effects that deepen as usage grows. The thesis is not “all plumbing is valuable,” but rather that the right infrastructure can become more valuable precisely because intelligence is becoming abundant.

If that thesis is right, three things change.

First, application-layer diligence must identify a moat independent of models. Workflow ownership, proprietary feedback loops, regulatory standing, distribution, and measurable outcomes can qualify.

Second, infrastructure deserves to be evaluated on network economics rather than dismissed as plumbing, or awarded a software multiple merely because it moves data. Investors should examine connection density, rights to use and exchange data, implementation cost, transaction growth, customer concentration, pricing power, and reach into fragmented settings.

Third, the strategic-acquirer universe is broader than healthcare IT. Financial institutions, industrial companies, and technology platforms may all need direct access to the networks on which their healthcare products depend. They are not only buying features. They are buying a position in the flow of data and transactions.

This is not bearishness on AI, but instead thinking about what follows if AI lives up to its promise.

The application layer will generate enormous value, but it will require infrastructure beneath it to scale. In healthcare, that infrastructure is regulated, embedded, and difficult to recreate. The headline from this cycle may eventually be that AI delivered. The footnote will be that its successful delivery happened on rails built and capitalized far from the hype and headlines.

For investors paying attention, the footnote is where the durable returns live.

A note from the author:

The infrastructure layer is one of several areas where Health Enterprise Partners is spending time this year. The broader market conditions shaping those choices are discussed in our annual reflections, Break Stuff: Ten Things the Healthcare Market Is Telling Us in 2026.

Ezra Mehlman is Managing Partner at Health Enterprise Partners, a healthcare IT and services growth-equity firm whose limited partners include leading hospital systems and health plans. He is also an Adjunct Professor at Columbia Business School, where he has taught the Healthcare Venture Capital and Private Equity: HCIT and Services elective for the past 12 years.  The views expressed are his own.