Investors poured $14 billion into health tech AI last year. That is 55% of all healthcare technology venture funding, up from 37% just two years ago.¹ Average venture capital round jumped 42% to $29.3 million.¹ The money is loud, confident, and everywhere.
Meanwhile, the average primary care practice — the one seeing Medicaid patients at 8 a.m. tomorrow — still juggles 12 value-based care contracts spanning Medicaid, Medicare, Medicare Advantage, D-SNP, and Fee-For-Service, tracks 50+ distinct HEDIS measures across those contracts, and burns roughly 1,000 hours of administrative work annually. By hand. Across spreadsheets, portal logins, and fax machines. The funding boom and the exam room exist in two different centuries.
Where the $14 Billion Is Going — and Where It Is Not
Follow the capital and a pattern emerges quickly. The largest health tech AI deals are flowing to enterprise platforms — large health systems, national payer analytics, drug discovery, and administrative automation at scale. These are important problems. They are also problems that exist many layers above the point of care.
Independent primary care practices, small group clinics, Federally Qualified Health Centers serving Medicaid populations — these are not the customers that a $29 million average deal size is designed for. The economics do not work. A five-provider Medicaid practice cannot absorb a six-figure annual platform license, a 12-month implementation timeline, and a dedicated IT team to manage the integration. So the innovation wave washes over them. The practices that need workflow relief the most are the last to get it.
This is not a technology gap. It is a deployment gap. The tools exist. The funding exists. What is missing is a delivery model that meets small and mid-size practices where they actually are — inside the daily routine of seeing patients, closing care gaps, and trying to make value-based care financially sustainable.
Why 80% of Healthcare AI Projects Stall Beyond the Pilot
Industry analyses consistently estimate that 80% of healthcare AI projects fail to move beyond the pilot stage.² The reasons are consistent: the technology was built for a controlled environment, not a chaotic one. It assumed clean data, standardized workflows, and dedicated staff to operate it. Real primary care has none of those things.
A practice running Medicaid patients does not have a data science team. It has a medical assistant who arrived at 7 a.m., a front desk coordinator managing no-shows and walk-ins, and a provider who will see 25 patients today across four different payer contracts. When technology requires that practice to change its behavior — to log into a new portal, interpret a new dashboard, learn a new data language — it does not get adopted. It gets abandoned. Not because the practice does not care about quality, but because the innovation was designed for the conference stage, not the exam room.
Meanwhile, the independent practice landscape keeps shrinking. The share of physicians in private practice has fallen from 60% in 2012 to just 42% in 2024, according to the American Medical Association.³ Every practice that closes or gets absorbed by a health system takes with it the kind of community-rooted, relationship-driven primary care that Medicaid populations depend on. The irony is painful: the AI investment boom is accelerating at precisely the moment the providers who need the most help are disappearing.
What Technology That Actually Reaches the Exam Room Looks Like
The question is not whether AI and advanced analytics belong in primary care. They absolutely do. The question is how they get there. And the answer has very little to do with the sophistication of the algorithm. It has everything to do with the delivery model.
Technology that reaches the exam room does not arrive as a standalone product with a login and a training manual. It arrives embedded inside a workflow the practice already uses — with human support to make the transition seamless. It does not ask a provider to become a data analyst. It puts a prioritized worklist in front of the team every morning that says: here are the patients who need to come in, here is what they need when they arrive, and here is how you get paid for doing it.
This is the principle behind CareEmpower® — a platform we did not build for conference keynotes. We built it for the medical assistant pulling up today’s schedule at 6:45 a.m. CareEmpower consolidates worklists from multiple payer contracts into a single prioritized view. Patients are sorted not alphabetically, but by clinical urgency: recent hospitalizations needing follow-up within seven days come first. Chronic care and complex needs visits follow. Preventive care aligned with Early and Periodic Screening, Diagnostic and Treatment (EPSDT) schedules fills in behind them. The practice does not need to cross-reference 12 contracts to figure out who to call. The platform has already done that work.
Before each visit, the Chart Prep Tool surfaces everything the provider needs: preventive screenings due, medication review needs, a 12-month care timeline, documentation and coding opportunities for risk adjustment — assembled and ready before the patient walks through the door. After the visit, the Manage Visit workflow lets staff mark activities as complete, update screening statuses, and submit Care Team Referrals for patients who need support beyond the practice walls — all within the same interface. No second login. No separate system. No extra clicks.
And when the practice needs backup it cannot provide alone, CareEmpower connects directly to Care Specialists and Community Health Workers who live in the communities they serve. A referral takes under a minute. The care team acts on it. The outcome feeds back into the platform. That closed loop — technology plus human support plus financial alignment — is what actually moves quality measures. Not the algorithm alone.
Adoption Over Hype — the Metric That Actually Matters
Only 14% of payers have adopted domain-specific AI tools, and just 25% report having an established AI strategy.⁴ The vast majority of provider networks are expected to figure out new tools on their own. In an environment where the average practice already dedicates 1,000 hours a year to VBC administration, that expectation is not just unrealistic — it is a design failure.
Adoption is the only honest metric in healthcare technology. Not deployment. Not licenses sold. Adoption — as in, the practice uses it voluntarily, repeatedly, as part of their daily routine. Across our network, 77% of attributed members have primary care practices where staff are regular CareEmpower users. In our most mature market, that figure reaches 83%. Roughly 1,400 unique users engage with the platform monthly across five states, with approximately 16,900 sessions over the past 12 months. Our newest market, Virginia, already has the highest engagement levels of any launch — reflecting eight years of learning about what it takes to make technology stick in a real practice.
That adoption did not happen because we built a better dashboard. It happened because we paired the platform with Practice Performance Advisors who coach teams on-site, Provider Account Managers who review performance monthly, and the Equality Care Incentive Program (ECIP) — which pays practices quarterly for the population health activities they complete through the workflow. Practices earn for wellness visits, transitions of care, chronic condition management, and preventive screenings — work they are already doing. That is not technology driving adoption. That is technology, human support, and financial alignment driving adoption together.
The Exam Room Does Not Need More Innovation. It Needs Innovation That Shows Up.
We are not against the $14 billion. Healthcare needs better technology, and the investment will eventually produce breakthroughs that reach every level of care. But the timeline matters. The practices serving Medicaid populations today cannot wait for the enterprise AI wave to trickle down to a five-provider clinic in south Phoenix or a rural FQHC in east Tennessee. They need technology that integrates with the 36+ EHR systems they already run, delivered by people who understand value-based care at the practice level, backed by financial models that pay for the work they do every quarter — not every 12 months.
The most important question in health tech right now is not “what can AI do?” It is “who is it actually reaching?” Until the answer includes the independent Medicaid PCP juggling 12 contracts across five lines of business and 50+ measures by hand, the $14 billion is solving someone else’s problem.
Sources
¹ Bessemer Venture Partners, “State of Health AI 2026.” https://www.bvp.com/atlas/state-of-health-ai-2026
² HealthTech Digital, “The AI Implementation Gap: Why 80% of Healthcare AI Projects Fail to Scale Beyond Pilot Phase.” https://www.healthtechdigital.com/the-ai-implementation-gap-why-80-of-healthcare-ai-projects-fail-to-scale-beyond-pilot-phase/
³ American Medical Association, “Physician Practice Benchmark Survey, 2024.” https://www.ama-assn.org/practice-management/private-practices/smaller-share-doctors-private-practice-ever
⁴ Menlo Ventures, “2025: The State of AI in Healthcare.” https://menlovc.com/perspective/2025-the-state-of-ai-in-healthcare/
