The investable universe for artificial intelligence in healthcare is not one market but several distinct categories with different regulatory paths, revenue models, and risk profiles. A useful starting point is to separate companies whose AI is embedded in an FDA-authorized medical device from those selling data platforms, drug discovery tools, or administrative software. The FDA maintains a searchable list of AI-enabled medical devices authorized for marketing in the United States, which provides a primary documentation trail for device-focused due diligence [2][7]. The FDA states that it has authorized over 1,600 AI-enabled medical devices for marketing in the United States as of September 2026 [2]. For platform and pharmaceutical names, the evidence is more dispersed across earnings disclosures, partnership announcements, and venture surveys. This guide explains those category distinctions, the concrete workflows they support, and a hypothetical evaluation example that can be adapted without inventing clinical outcomes or investment recommendations.

Regulated Devices as a Distinct Category

The FDA's Center for Devices and Radiological Health regulates AI-enabled medical devices under the Federal Food, Drug, and Cosmetic Act using a risk-based approach tied to intended use and technological characteristics [2]. A marketing submission may include one or more AI-enabled device software functions and may include a Predetermined Change Control Plan, which describes how a manufacturer will manage anticipated modifications over the product life cycle [2]. This regulatory structure matters for investors because it creates a documented authorization event that can be checked against the FDA's public list.

The FDA's AI-enabled device list is not comprehensive. It includes devices identified primarily through AI-related terms in marketing authorization document summaries or device classifications, and it is updated periodically [7]. Devices authorized but lacking published decision summaries within a data collection period appear in later updates [7]. The list therefore functions as a screening resource rather than a complete census. For example, a June 2026 entry shows Agada Medical's Auto-Seg spine software under the radiology panel with submission number K253628 and primary product code QIH, but the list alone does not establish commercial adoption or revenue contribution [8].

Data Platforms and Diagnostics Workflows

A second category includes companies that sell AI-powered insights to physicians or researchers rather than operating as regulated device manufacturers in every product line. U.S. News describes Tempus AI as a healthcare data specialist using artificial intelligence to advance precision medicine, with AI-powered insights intended to help physicians find details they may miss [1]. The source reports a share price around $70 and a year-to-date gain of 28.5% as of mid-November 2026, but these figures are point-in-time market data, not a forecast [1].

The distinction between a data platform and a regulated device is not always clean. A company may hold FDA authorizations for some software functions while selling other analytics services outside that pathway. Investors should check the FDA list for specific product names rather than assuming a company's entire portfolio is regulated. The FDA notes that some AI-enabled devices may adapt over time as new data becomes available, which introduces total product life cycle considerations from development through monitoring and modification [2]. The FDA also states that it does not regulate AI as such; it regulates medical devices, including AI-enabled medical devices, and certain software functions are excluded from the device definition under section 520(o) of the FD&C Act [2].

Building a Source Checklist for One Candidate

For a single candidate, assemble three documents before forming a view. First, locate the FDA list entry if the company claims a regulated device, recording the submission number, panel, and product code [8]. Second, pull the most recent earnings or investor disclosure that separates AI-related revenue from legacy business, if such a breakdown exists. Third, review the Menlo Ventures survey or equivalent demand-side research to see whether the company's target workflow is rising or falling in buyer priority [3]. The FDA list is updated periodically, so a missing entry may reflect a publication lag rather than absence of authorization [7]. A company may also hold authorizations not yet reflected in the public list, which is why the list should be treated as a screening tool rather than a definitive registry.

Large Operators and the Adoption Cycle

Large pharmaceutical and technology companies represent a third category where AI is deployed across existing franchises rather than sold as a standalone product. Alphabet's Google for Health platform provides generative AI-driven health information, while its AlphaFold model supports scientists researching proteins for vaccines, tuberculosis treatment, and drug discovery [1]. Eli Lilly, Pfizer, Stryker, and GE HealthCare appear in U.S. News coverage as larger companies deploying AI, with implied upside figures based on analyst price targets as of mid-November 2026 [1]. BlackRock's May 2026 commentary describes major pharmaceutical companies partnering with specialized diagnostics firms that hold genomic and clinical datasets, with the stated goal of improving how cancer is detected, monitored, and treated [4].

Fidelity's healthcare commentary frames adoption in phases, starting with administrative efficiency and moving toward earlier disease detection and treatment discovery as AI computing becomes more available [6]. The source cites administrative spending of roughly $1 trillion annually, about 20% of U.S. healthcare spending, as a potential efficiency target [6]. This phase framing is an analytical lens, not a verified timeline. The same source notes that the timing of breakthroughs remains uncertain and depends on infrastructure build-out and computing cost declines [6]. BlackRock reports that the FDA cleared six AI devices in 2015 compared with 331 in 2025, a cumulative total of 1,394 authorized AI and machine learning devices since 2015, with radiology and diagnostics accounting for 77% of clearances [4].

Payer Pressure and Procurement Signals

Menlo Ventures' September 2026 survey of 550 executives across providers, payers, and life sciences organizations offers a demand-side view of AI adoption [3]. The survey reports that 43% of payers, and 57% of large payers, say provider documentation and coding AI raised their medical cost trend in 2026, up from 23% in 2025 [3]. Prior authorization became payers' top AI priority at 82%, up from 59%, in response to CMS interoperability and prior authorization rule deadlines [3].

These figures describe payer perceptions and priorities, not verified cost causation or vendor performance. The survey methodology covers executives who participated in procuring an AI solution, which means the sample is skewed toward organizations already active in AI purchasing [3]. The same survey reports that utilization management rose to 41% as a top-five payer AI priority from 10%, and care management rose to 61% from 46% [3]. For investors, the value is in identifying which workflows are attracting budget: utilization management, care management, and administrative automation. The survey does not establish which public companies will capture that spending.

A Hypothetical Evaluation Workflow

A disciplined evaluation might proceed in four steps. First, identify the category: regulated device, data platform, or large operator. Second, for any device claim, search the FDA's AI-enabled device list by company and product name, noting the submission number and date of final decision [7][8]. Third, separate regulatory authorization from commercial traction by checking earnings disclosures or partnership announcements. Fourth, test the revenue model against the adoption cycle described in the sources, asking whether the product addresses administrative efficiency, clinical decision support, or drug discovery [6].

Consider a hypothetical analyst comparing a radiology AI vendor with a payer-focused documentation vendor. The radiology vendor's product appears on the FDA list with a 510(k) submission number, which confirms a regulatory event but not hospital purchasing volume [8]. The payer vendor's product is not a regulated device, so the analyst instead checks the Menlo Ventures survey for evidence that payer documentation AI is raising cost trends and driving utilization management priorities [3]. The two evaluations use different evidence because the products operate under different regulatory and commercial frameworks. Neither check settles whether either vendor's product improves clinical outcomes; that question requires evidence the cited sources do not provide.