The AI Pivot

The AI Pivot

Doximity's AI push is, so far, a cost the income statement already carries and a revenue line the disclosures do not yet show. Research spend jumped 40% to $131 million in fiscal 2026 — a fifth of revenue — and the fiscal 2027 guide sets adjusted EBITDA below the prior year in absolute dollars even as revenue grows. The clinical AI tools physicians use are free by design; the money is meant to come later, from a just-launched advertising product, in a market where an AI-native rival is compounding faster.

The cost is already in the numbers

For three years Doximity's research-and-development line drifted down as a share of revenue — from 19.1% in fiscal 2023 to 16.3% in fiscal 2025 — the picture of a business getting more efficient as it scaled. Fiscal 2026 reversed that in one step: R and D rose 40% to $130.7 million, or 20.3% of revenue [1], while revenue grew 13%. Management labelled fiscal 2026 "our year of AI investments," pointing to spending on functionality, content licensing, and compute usage [2].

Source: FY2026 Annual Report (Form 10-K), Management's Discussion and Analysis [3]; FY2027 adjusted-EBITDA-margin guidance from the Q4 FY2026 call [4].

Source: derived from reported financials, FY2022–FY2026 10-Ks; FY2026 figures per the Consolidated Statements of Operations [5].

The pressure shows up faster in the cash-earnings line the company steers by. Adjusted EBITDA margin held at 55% for the full year, but the fourth quarter tells the trajectory: 45%, down from 50% a year earlier, with management naming "increased investment in AI compute driven by a steep ramp in AI usage which is outgrowing overall workflow engagement" as the driver [6]. The fiscal 2027 guide carries that forward for a full year: adjusted EBITDA of $323–335 million at a 49% margin [7]. At the midpoint that is roughly $329 million against $358 million in fiscal 2026 — cash earnings falling about 8% while revenue rises 4%.

Source: Q4 FY2026 earnings call — FY2025/FY2026 actuals and FY2027 guidance [8]. FY2027 is the guidance midpoint.

Stock-based compensation runs the same way: about 18.9% of revenue in fiscal 2026, guided to the "low 20s" as a percent of revenue in fiscal 2027 before trending back down, driven by the Pathway acquisition and performance grants for the growing AI team [9]. The buyback more than offsets the dilution (Network and Slowdown), but the AI build widens the gap between reported and cash earnings before it narrows it.

Two different bets under one label

"AI" at Doximity is two initiatives with opposite economics, and they have to be underwritten separately.

Source: FY2026 10-K and Q1–Q4 FY2026 earnings calls, per the citations below.

The clinical suite is a defensive spend, not a product line. Doximity GPT (which composes letters), Scribe (physician-controlled ambient notes), and Pathway's Corpus (a fact-checked clinical and drug reference) are bundled into one free assistant [10]. Pathway Medical, acquired July 29, 2025, was a six-person Montreal startup whose model scored 96% on the U.S. medical licensing exam; Doximity is dropping the roughly $300-a-year subscription Pathway charged and giving the tool away inside Doximity GPT [11]. The purchase was small: $36.3 million of consideration including $26.7 million of cash, plus $23.9 million of restricted stock vesting over five years [12] — a rounding error against $749 million of cash and securities. The point of the free suite is engagement: 300,000 physicians used the AI tools in the first full quarter after the deal [13], and 20% of health systems were using Doximity's AI within a quarter of launch [14]. Management's own framing is that the tools deepen the moat rather than earn a margin — with clinical AI still "in its nascent stages," more than half of 1,800 surveyed physicians had used no clinical AI at all [15].

The revenue case rests on the second bet: commercial AI Search. Launched in the fourth quarter of fiscal 2026, it is described as "already generating strong early interest" [16]. It reaches for an advertising budget Doximity has not sold into before — paid search. Management cites an eMarketer figure that 55% of healthcare digital-marketing spend goes to search [17], and frames the product as a way to "tap into innovation upsell budgets and search budgets" [18]. The company likens the moment to the early days of HCP programmatic advertising three years ago and guides to exiting the calendar year as a double-digit grower once search budgets and delayed pharma funds are released [19].

That is the honest boundary of what can be underwritten today. The suite that physicians touch is free and not sold; the product meant to pay for the whole pivot is one quarter old with no disclosed revenue, and the clinical-reference asset behind it — Pathway — is explicitly not being monetized yet: "we are not trying to actively sell or monetize in the marketplace yet," management said, adding that $91 million of quarterly operating cash flow lets it "invest for the longer term" [20]. The TAM is real and adjacent; the revenue is a plan with early interest attached.

The rival that sets the bar

The reason the clinical suite is defensive is visible in Doximity's own filings: the fiscal 2026 10-K lists OpenEvidence — a company barely three years old — alongside WebMD's Medscape as a competitor for its Marketing Solutions customers [21]. OpenEvidence is the fastest-adopting clinical-AI tool among U.S. physicians, and by public reporting it raised capital in January 2026 at a $12 billion valuation — up from about $1 billion a year earlier and $3.5 billion in mid-2025 — with roughly 45% of U.S. physicians using it and about 18 million clinical consultations in a single month, several times its volume a year before. Those figures come from funding announcements and press, not company filings, and cannot be reconciled to a source document here; they are directional press estimates rather than precise counts.

The two companies are also in court. On June 20, 2025, OpenEvidence sued Doximity, its chief technology officer, and its director of AI products in the U.S. District Court for Massachusetts, alleging unauthorized access to its AI platform; Doximity filed counterclaims in September 2025 for false advertising and defamation, and the plaintiff amended its complaint in October 2025 to add Pathway Medical and computer-fraud claims [22]. The litigation matters less for its damages exposure than for what it signals: the two are fighting over the same physician attention in clinical reference, and Doximity is the incumbent defending, not the challenger.

This rivalry is where the structural risk to the network sits. Doximity's read — stated repeatedly — is that in clinical AI "the tech here is increasingly a commodity" and that trust, relationships, and the platform decide who wins, an argument its 85%-plus verified network is built to make (The Moat) [23]. The counter-case is that a well-funded, AI-native product is winning physician usage in exactly the reference workflow Doximity is entering with a free tool — and that giving Pathway away is itself evidence the company cannot charge for clinical AI while OpenEvidence is free and ahead. Which of those holds is not yet in the numbers.

What would turn cost into revenue

The pivot is falsifiable, and the lines that would confirm or deny it are named. My read: the AI build is a rational, self-funded defense of the network today, and a credible but unproven growth bet for tomorrow — worth carrying because Doximity can afford it, not yet worth crediting as revenue. What would move that read, in order of decision value:

Sources: Q3 FY2026 call — double-digit-exit and search-budget commentary [24]; Q4 FY2026 call — margin and stock-comp trajectory [25].

The market has already taken a side on the near term: the shares fell roughly a quarter on the fiscal 2027 guide, reading the combination of a 4% top line and a stepped-up AI bill as reason to worry rather than to wait. That reaction is about timing. Whether the pivot works is about the three lines above — and none of them will resolve before the fiscal 2027 back half.