Healthcare AI: Where We Stand — September 2026

Healthcare AI: Where We Stand

Healthcare AI is becoming part of the machinery of healthcare. It is entering the way we discover treatments, document visits, interpret information and help people find care. Money is flowing into new businesses, while established companies are deciding which capabilities to build, buy or bring into their existing operations.

My assessment is that we are at a major change moment. Understanding that change requires looking at the investment alongside the work the technology actually performs.

I’m Jivesh Sharma, and this is HealthIT.com. In this September 2026 briefing, we’ll examine where healthcare AI is taking hold, who is backing it, and what deserves a closer look.

The healthcare AI landscape

Healthcare artificial intelligence spans five broad areas.

In research, AI supports work such as identifying promising drug candidates and analyzing biological data. The commercial ambition is to improve the productivity of discovery. A promising computational result still has to survive laboratory and clinical testing.

In clinical care, applications include image analysis, information retrieval and documentation. These involve very different responsibilities. Drafting a note and helping interpret a scan need different kinds of evaluation.

In administration, companies are applying AI to coding, claims and other document-heavy processes. Much of the business case rests on whether these tools reduce the cost of getting work done.

For patients, AI offers another way to ask health questions, understand benefits and navigate services. Accuracy matters, but so does whether a person can reach a human when the situation requires one.

Underneath these applications sits the infrastructure: data connections, computing capacity and controls over access and use. This less visible work helps determine whether an application can operate reliably inside a healthcare organization.

Use is spreading through different routes

Healthcare AI adoption is happening through both professional and personal routes.

In the American Medical Association’s 2026 survey, eighty-one percent of responding physicians reported using AI professionally. Their uses included research summaries, documentation and patient communications. That broad measure tells us that AI has entered professional activity. It does not tell us that every physician uses it daily or that each application improves care.

Rock Health’s December 2025 consumer survey found that thirty-two percent of respondents had used an AI chatbot for health information. The surveys ask different questions, so we should not compare the percentages as if they measured the same behavior.

Together, they illustrate an important operational issue. Healthcare organizations need to understand both the tools they formally deploy and the information patients bring into a conversation.

Capital is growing and concentrating

The financing picture for healthcare technology deserves precision. Rock Health reports that United States digital health startups raised seven-point-four billion dollars in the first half of 2026, compared with six-point-four billion a year earlier. This is a digital health total, which is broader than healthcare AI.

The number of financing deals was almost unchanged. More money therefore does not establish that more companies were created. And forty-five percent of the capital went into rounds of at least one hundred million dollars.

My interpretation is that investors are placing substantial bets while concentrating resources in selected businesses. For buyers, that makes the commercial questions more important. A well-funded vendor still needs a useful product, a workable implementation and a reason for customers to keep paying.

Established companies are committing resources

Established healthcare companies are participating in several ways.

Lilly and NVIDIA announced up to one billion dollars in combined investment over five years for an AI research collaboration. That commitment covers talent, infrastructure and computing. It is an example of building capacity for drug discovery, with the scientific outcomes still to be established.

GE HealthCare completed its acquisition of Intelerad in March 2026 for a base cash price of two-point-three billion dollars. This is a broader imaging software acquisition that supports its AI strategy. The purchase also illustrates the importance of the systems through which new capabilities reach clinicians.

UnitedHealth Group describes applying AI to member navigation and clinical documentation. Here, the mechanism is deployment across existing operations. The company’s description establishes what it says it is doing, while performance claims need their own assessment.

These examples involve different financial categories and different routes to value. They show why we need to look beyond venture funding to understand the scale of the transition.

New companies need a route into care

Emerging healthcare AI companies face a broad and competitive market. Mayo Clinic’s accelerator announced a June 2026 cohort of twenty startups, working across areas that include clinical reasoning, research workflows and care coordination. That is a useful illustration of breadth, although participation is neither a clinical endorsement nor a count of newly founded businesses.

At the same time, established software vendors such as Epic are adding AI to their products. My view is that this raises the bar for startups. They need to show why their particular capability is worth purchasing, integrating and maintaining when a customer’s existing systems are also evolving.

For a healthcare buyer, the question becomes whether a new product solves enough of a problem to justify another operational dependency.

Results depend on the product and setting

A randomized study of ambient AI scribes illustrates why healthcare AI needs evidence about specific products. Published in the New England Journal of Medicine Group’s AI journal in 2025, the trial enrolled two hundred thirty-eight physicians at one academic institution.

Nabla reduced time spent in notes by about nine-and-a-half percent relative to usual care. The DAX Copilot group did not show a statistically significant reduction on that measure.

The study evaluated particular product versions in a particular setting. It does not settle how today’s versions perform elsewhere, and these comparisons do not establish that one product is superior to the other.

The useful lesson is that an attractive category can contain different results. We need to examine the actual task, the measure of improvement and the conditions under which the tool was tested.

Value depends on who benefits

The definition of success for healthcare AI changes depending on whose experience we measure.

Patients may value faster access and clearer explanations. Clinicians may value less clerical work and more time for judgment. Organizations need a sustainable cost structure. Payers may focus on expenditure and administrative efficiency.

Those interests can align, but alignment requires deliberate choices. If AI saves documentation time, an organization can use that capacity in different ways. It might reduce work after hours, make room for more appointments or allow longer conversations. The software alone does not decide who receives the benefit.

That is why evaluation should include the patient-provider relationship, along with financial and technical performance.

What deserves attention next

Over the next year, I would watch whether improvements persist after the initial rollout of a healthcare AI system. I would examine the full cost, including integration, supervision and correction. And I would look closely at accountability when a system makes a mistake or cannot complete its task.

Healthcare AI is a major change moment because it reaches into both the delivery of care and the economics around it. The opportunity is substantial. The work now is to identify which applications earn a lasting place through useful results.

Continue the research at HealthIT.com

HealthIT.com brings together AI product research, Clinical AI Briefs and practical evaluation guidance. It is part of the Nexgen Precision portfolio, and we make sources and evidence gaps visible.

Reading and downloads remain open without a subscription.

I’m Jivesh Sharma. Thank you for joining me.

This episode uses an AI-generated version of my own voice.