Rock Health’s mid-year market report, published July 13, contains a quiet admission with loud implications: the firm stopped labeling startups “AI-enabled” a quarter ago because the label no longer distinguishes anything. The investor question has changed, in the report’s own words, from “Who has AI?” to “Who has something AI alone can’t provide?” [Rock Health, 2026]. Digital health startups raised $7.4 billion across 244 deals in the first half of 2026, and that capital is no longer buying the technology. It’s buying what happens after you sign.
The Problem
Health systems still purchase AI as if capability were the scarce thing. The evaluation stack proves it: the demo and the accuracy bake-off, scored against a feature matrix. That made sense in 2024. It doesn’t now. Seventy-five percent of US health systems run at least one AI application, up from 59% a year earlier, and half run three or more, per Eliciting Insights’ survey of executives at 120 systems [Fierce Healthcare, 2026]. Capability is everywhere, which is exactly why the people financing your vendors quit paying a premium for it.
What they pay for instead, Rock Health names precisely. Four moats kept surfacing across 2026’s financings: founder domain expertise, “owning” more of the healthcare operating layer, hands-on delivery through forward-deployed engineers, and ecosystem partnerships [Rock Health, 2026]. In a pitch deck these read as defensibility. From the buyer’s chair, three of the four are the same mechanism wearing different clothes: each one raises the cost of ever replacing the vendor.
The Insight
Start with the forward-deployed engineer, the role Commure and Qualified Health have made core to their go-to-market and that Anthropic and OpenAI are now fielding themselves [Rock Health, 2026]. Rock Health describes the job without euphemism: FDEs “work directly with customers on an ongoing basis and co-develop custom workflows from within the client’s environment.” That engineer produces real value. The custom workflow works. It is also unportable by design, because it exists only inside that vendor’s stack, and nobody in your budget cycle has priced the rebuild when renewal arrives at a higher rate. Moat construction, billed as customer success.
“Owning the operating layer” runs the same direction. The more workflow a vendor orchestrates, the more context it accumulates and the harder it becomes to carve out. Rock Health flags the buyer-side consequence in a single line most readers will skim past: as vendor roadmaps expand, customers “may increasingly find themselves evaluating redundant offerings” [Rock Health, 2026]. You will pay twice for overlapping capability, and the overlap now includes Epic, which is handing its customers do-it-yourself agent tooling on the same turf. Hospitals that deferred their AI strategy to Epic’s release cycle already know how that dependency prices itself.
Partnerships close the loop. The report is candid: “every new integration or endorsement can make a product more trusted and harder to replace, giving customers even more reason to stay within the same ecosystem.” Abridge’s latest announcement added Nvidia, AHIMA, the American Diabetes Association, the American Academy of Family Physicians, and two smart-room companies in one sweep [Rock Health, 2026]. Harder to replace is not a side effect. It is the strategy.
“In venture language it’s defensibility. On the hospital’s ledger it’s the price of leaving, and the buyer is funding its construction.”
None of this is aimed at some other industry. Rock Health founder Halle Tecco puts it flatly in the report: “if you look at the big AI companies getting funded, most of them are building for health systems.” A decade ago, she notes, few startups would sell into hospitals at all; the sales cycle was too long. Now the best-capitalized companies in digital health are organized around hospital workflows [Rock Health, 2026]. That attention is flattering. It is also the point: health systems are the terrain the moats get built on.
Here is the part I’d argue most leadership teams have backwards: this is the strongest negotiating position hospital buyers have held since the category existed. Twenty megadeals absorbed 45% of first-half capital across just 19 companies, up from 22% of funding in 2024 [Rock Health, 2026]. Vendors carrying that much capital need lighthouse customers and reference logos more than any single hospital needs any single vendor. Commodity technology plus desperate differentiation equals buyer leverage. Most systems spend it on a better subscription price instead of the terms that will matter in three years: data egress, workflow portability, performance benchmarks with consequences, decommissioning obligations.
Measurement compounds the problem. More than half of the systems able to quantify ROI on deployed AI report at least a 2x return, and the qualifier is doing all the work in that sentence: plenty of systems can’t quantify it at all [Fierce Healthcare, 2026]. The market said the quiet part in May, when Optura raised a $17.5 million Series A for a platform whose entire purpose is measuring whether healthcare AI pays off [Fierce Healthcare, 2026]. A vendor category now exists because buyers can’t price what their AI delivers. And a system that can’t price what a tool delivers can’t price what leaving it would cost either. Lock-in hides inside that blind spot.
In Practice
The four moats, translated to the buyer’s side of the table.
| The moat VCs fund | The pitch you hear | What compounds on your ledger |
|---|---|---|
| Operating-layer ownership | ”One platform for the whole workflow” | Every added task raises exit cost; roadmaps overlap with your other vendors and Epic |
| Forward-deployed engineers | ”White-glove partnership” | Custom workflows that exist only inside this vendor’s stack |
| Ecosystem partnerships | ”Integrated with everyone you trust” | Each integration is one more dependency to unwind at exit |
| Founder domain expertise | ”Built by people who’ve done your job” | The one moat that costs you nothing |
The Bottom Line
The spread between what a vendor charges and what the underlying model costs to run is increasingly a switching-cost premium, and the first half’s numbers say the market intends to widen it. Watch the M&A tape: 115 digital health acquisitions in H1, 71 of them in Q2 alone, the busiest quarter since Q3 2021 [Rock Health, 2026]. Consolidation means the vendor you signed may resurface as a line item inside a larger platform whose roadmap you never chose. The exposed party is the mid-rollout health system, the one among the 50% already running three or more applications, renewing contracts while the entanglement deepens each quarter. The beneficiaries are vendors who convert white-glove service into permanence, and the minority of buyers who negotiate exit terms while vendors still need their logo. When your ambient documentation renewal lands next year, you will know the license cost to the dollar, because the vendor wrote it down. What would leaving cost? Nobody in your building has run that number. Your vendor’s investors are betting nobody ever does.