Epic Systems
Licence plus a long Epic-staffed implementation plus recurring maintenance, sold direct on a single integrated codebase; pricing unpublished. Three facts separate it from every listed comparable: no outside capital ever (founded 1979 on a $70,000 investment); no acquisitions; and AI monetised by SUBSTITUTION rather than a marketplace toll — developer access priced near zero while Epic re-addresses the ambient and revenue-cycle line items itself, at roughly zero customer-acquisition cost.
The thesis on this name
State of AI for Healthcare
PRIVATE and permanently uninvestable — carried because omitting the sector's largest value accruer would misstate the board. Epic is reported at $6.7B of revenue in 2025 — company-stated to Becker's, 28 Apr 2026, unaudited, and unverifiable because Epic files nothing — against a reported $5.7B in 2024 and $4.9B in 2023 (a growth rate implied by two reported figures, one of them a trade-press estimate). It holds 43.7% of US acute-care hospitals and 56.9% of beds after a fifth consecutive year of gains — +77 hospitals and +18,679 beds in a year when hospital EHR purchasing decisions fell ~40% versus 2024 — and runs 325M+ patient records and 190M+ MyChart users on ~15,000 employees. It now ships the AI itself: AI Charting reached general availability on 4 Feb 2026, Penny is live at 200+ organisations in revenue cycle, and Art runs at 16M uses a month. The monetisation mechanism is SUBSTITUTION, not a toll. What Epic actually charges an AI vendor to cross the chokepoint is negligible — Vendor Services membership at ~$1,700–1,900/yr, a Connection Hub listing at ~$500/yr, no disclosed revenue share since App Orchard closed in December 2022, and free USCDI read APIs under the 21st Century Cures Act. What it takes instead is the line item: it re-addresses the ambient and revenue-cycle dollar inside a contract it already holds, at a price it does not disclose, and it acquires that customer at zero cost because it already owns the relationship, the BAA and the security review. So a hypothetical listed Epic would be a competitor to both of this book's longs, not a better way to own them. There is no equity instrument to buy: Epic has never taken outside capital, and three health-system CEOs are reported to have pledged to block an IPO. The largest single lever on US clinical-AI distribution cannot be owned, and that is the board's finding rather than a gap in it.
Earnings, margins, COGS & capex
Epic is the sector's largest value accruer and its least analysable object; the board carries it because omitting it would misstate the value chain, not because a position exists. Every financial fact is company-stated and unaudited ($6.7B of 2025 revenue, told to Becker's 28 Apr 2026), trade-press estimate, or unknowable — every margin, the cash balance, the AI pricing. Share, by contrast, is measured by a third party (see keyMetrics). The decisive fact is what Epic does NOT charge: even at 1,000 paying vendors the entire AI-vendor toll is ~$1.9M, roughly 0.03% of that $6.7B (canon CM-EPIC-02 — the 1,000-vendor step is load-bearing). Epic is not a tollbooth taxing tenants; it is a distribution owner selling the AI itself, collecting at the next contract cycle rather than at the gate.
Revenue trend
Margins
n/a. The marginal seat costs near zero, but implementation scales with headcount and native AI adds per-use inference the licence model does not re-price.
n/a. Substituting Epic's own AI into existing contracts wins the line item but buys the compute bill.
n/a. Forty-seven years without outside capital while self-funding a large owned campus implies durable positive FCF — an inference, not a measurement.
COGS structure
People, increasingly compute. A ~15,000-person salaried engineering and implementation organisation on one campus, one integrated codebase, no acquired products, supports deployments taking months to years that Epic staffs directly — why switching costs are so high and why the gross-margin question is genuinely open rather than assumed software-like. Native AI adds per-encounter inference, and transcription comes from Microsoft's Nuance/Dragon stack, so part of the marginal cost of the AI Epic sells is a payment to a competitor.
Capex
Not disclosed. The visible item is the owner-occupied Verona campus, continuously expanded from operating cash; no clinical hardware fleet. Whether Epic builds or rents inference capacity is unsettled in the public record.
Latest earnings
n/a — no consensus, no guidance, no print against which a beat or miss could be defined.
None, ever. Company-side signals only, each vendor-stated: revenue 'increased to $6.7 billion in 2025' (spokesperson to Becker's, 28 Apr 2026, unaudited); AI usage up 40x during 2025 and 85%+ of customers actively using Epic AI (HIMSS26, Mar 2026, via trade relay); Penny cutting coding-related denials 20%+ at high-usage sites. All are vendor or single-site claims — no control arm, no confidence interval — and none is used in any sizing here.
- US acute-care share
- 43.7% of hospitals in 2025, up from 42.3% — fifth consecutive year of gains; 56.9% of beds; Oracle Health 21.9%, Meditech 14.7% (per KLAS as reported; canon CM-KLAS-01)
- Net additions, 2025
- +77 hospitals / +18,679 beds — carried forward, NOT re-confirmed this run (canon CM-KLAS-02)
- Revenue, 2025
- $6.7B — company-stated to Becker's 28 Apr 2026, unaudited; no primary source, no knowable margin (canon CF-EPIC-01)
- Records / MyChart / employees
- 325M+ patients with an Epic record; 190M+ MyChart users; ~15,000 employees (reported; canon PB-040)
- Native AI footprint
- AI Charting GA 4 Feb 2026 at an UNDISCLOSED price; Penny at 200+ organisations; Art at 16M uses/month (canon PB-040 / PB-055)
- Developer fees and the whole toll
- Vendor Services ~$1,700-1,900/yr, Connection Hub ~$500/yr, free USCDI read APIs; even at 1,000 paying vendors the entire toll is ~$1.9M, ~0.03% of $6.7B of company-stated, unaudited 2025 revenue (canon CF-EPIC-04 / CM-EPIC-02)
- FDA AI authorisations
- ZERO across all 1,524 rows, vintage 2026-06-16 — a verified zero (canon REG-FDA-019)
Growth drivers
- Share compounding against a shrinking pool — 43.7% of US hospitals in 2025 versus 42.3%, +77 hospitals / +18,679 beds while purchasing decisions fell ~40% (per KLAS as reported; canon CM-KLAS-01/02)
- Displacement of the #2 — Oracle Health lost 56 hospitals and 14,676 beds in 2025 (canon CM-KLAS-03); Epic's gains and Oracle's losses are largely the same transactions
- Substituting its own AI into contracts it already holds — AI Charting GA 4 Feb 2026 at an UNDISCLOSED price, Penny at 200+ organisations, Art at 16M uses/month (canon PB-040)
- Revenue cycle, the adjacent layer that already prints software-like margins — Penny takes coding and appeals work inside the same contract
- International into a weak market — 34 hospitals / 6,580 beds signed in 2025 while global purchasing hit a five-year low (KLAS Global, 8 Jul 2026)
- Data scale into model scale — CoMET / Curiosity: 118M patients, 115B medical events, 151B tokens, 78 zero-shot tasks (arXiv 2508.12104), access from Feb 2026
Bull & bear
Epic owns the distribution layer every clinical AI must render into, and it has stopped renting that layer out — charging almost nothing at the gate while re-addressing the ambient and revenue-cycle line items inside contracts it already holds. Share is still compounding in a market where purchasing decisions fell roughly 40%, and the Cosmos corpus is not purchasable at any price. The most durable franchise in US health IT — and, by construction, the one nobody can own.
- The share position is a ratchet: 43.7% of US acute-care hospitals in 2025 against 42.3%, plus +77 hospitals / +18,679 beds while purchasing decisions fell ~40% (per KLAS as reported) — the gains came out of a competitor
- Substitution beats tolling. Even at 1,000 paying vendors the whole toll is ~$1.9M, ~0.03% of $6.7B of company-stated, unaudited 2025 revenue (canon CM-EPIC-02); the line item inside an existing contract is worth orders of magnitude more
- The acquisition asymmetry is close to unbeatable — a rival pays for a sales cycle, security review, BAA, integration and pilot per health system; Epic starts with all five done, which is how AI Charting reached GA base-wide on 4 Feb 2026
- Cosmos and CoMET are a genuine data moat: 118M patients, 115B medical events, 151B tokens, zero-shot across 78 tasks, generally matching or beating task-specific supervised models (arXiv 2508.12104)
- The capital structure is the strategy — no quarter to defend and no integration debt, so Epic can keep one codebase, give away the gate and collect at renewal
- Agent Factory extends lock-in from the record to the automation layer, moving switching cost to 'replace the chart and everything we built on it'
There is nothing to own, nothing to price and nothing to verify: no shares at any vintage, no post-money, no filing, no audited statement. Beyond un-ownability the franchise carries live risks — three surviving Sherman Act Section 2 claims in discovery, a substitution strategy paying a direct competitor for its core speech input, per-use AI costs on per-seat contracts, and a founder in her eighties whose doctrine is the only thing holding the no-IPO commitment together. Its uninvestability is the finding, not a gap in the research.
- The exposure is UNHEDGED and unhedgeable — private, no IPO path, three CEOs reported pledged to block one, so the only expression is position size. That is why the board's two longs total 11% of NAV rather than 30% (canon PB-033); being small IS the tail hedge
- Every financial figure is unverifiable by construction: $6.7B of 2025 revenue is a spokesperson statement (Becker's, 28 Apr 2026, unaudited), the $5.7B 2024 base a Fortune estimate, the 2023 figure without a named carrier
- A hypothetical listed Epic would be a COMPETITOR to the board's longs — it takes the line item those vendors sell rather than renting them access
- The 9 Sept 2025 ruling denied dismissal on all three core Section 2 claims, accepted the payer-platform market and let a tortious-interference claim over XCures proceed, dismissing only defamation and trade libel
- AI Charting's price is UNDISCLOSED and must not be asserted — a CHIO's expectation of 'comparable to Dragon Copilot' is an expectation, not a price (canon PB-055) — so nobody outside can benchmark the economics
- Key-person risk is the whole governance structure: a no-capital, no-IPO, no-acquisitions doctrine carried by a founder in her eighties on an informal pledge
What it is worth
NO VALUATION IS CARRIED, BY DECISION — and unlike a private peer the reason is non-existence, not staleness. Epic has never raised outside capital, so there has never been a round, a post-money, or a negotiated mark of any vintage to quote with a date. A valuation must not be inferred from revenue, a founder's estimated net worth, or a comparable's multiple. Even a revenue multiple is unusable: the revenue is company-stated and unaudited, the prior-year base is a trade-press estimate, and no margin is knowable — an unverified numerator over an invented denominator. The correct output is a structural finding, not a number: the largest single lever on US clinical-AI distribution cannot be owned.
Two halves pointing opposite ways. Franchise bear: the Section 2 claims survive summary judgment and a conduct remedy reaches API terms or the payer platform; native AI ships broadly but performs poorly enough that health systems re-buy specialists; Microsoft re-prices the speech supply; per-use compute outruns per-seat contract value; a succession event disrupts the discipline forty-seven years bought. Investor bear — the one that binds: none of it is observable, tradable or hedgeable. No security to short, no filing in which deterioration would appear, no mark to re-rate. The mirror-image tail is just as disruptive: Epic takes outside capital, files, or is acquired.
Epic keeps winning the record and the line item, slowly and unobservably. Share gains continue but decelerate as US acute care saturates; the mix shifts toward adjacency and abroad. Native AI displaces some third-party spend at renewal and coexists with the rest — enough to compress standalone ambient pricing and cap net revenue retention across the specialist cohort, not enough to clear the field. The Particle litigation grinds toward a settlement or a narrow ruling, governance holds, and Epic remains the most important asset in US health IT and the one with no price.
The franchise compounds and nothing external stops it. Share pushes past 43.7% as Oracle Health's losses feed the pipeline and international wins accumulate. Substitution works as designed: AI Charting, Penny, Emmie and Art take the ambient, coding and patient-engagement line items inside existing contracts, and the 2027 renewal cohort prices through with third-party ARR re-addressed. Cosmos and CoMET turn the installed base into a prediction franchise no rival can assemble; the Particle claims settle without a conduct remedy. The outside investor participates in none of it — the bull case for Epic is the bear case for owning the layer above it.
Framing only. Not a recommendation and not investment advice. Epic is carried as a high-conviction durable compounder AND as permanently uninvestable, because omitting the sector's largest value accruer would misstate the board. The exposure is UNHEDGED and unhedgeable, so the only expression is size: the two longs total 11% of NAV rather than 30%, sized so a total Epic impairment of both is an ~11% NAV event (canon PB-033); premium on a protective structure would exceed the loss it caps, so being small IS the tail hedge. A hypothetical listed Epic would be a COMPETITOR to both longs, because it takes the line item those vendors sell. WHAT WOULD MAKE IT INVESTABLE: nothing this board can underwrite. The falsifier is precise — Epic takes outside capital, files, or is acquired.
SWOT
Strengths
- Owns the chokepoint — 43.7% of US acute-care hospitals and 56.9% of beds per KLAS as reported, defining what renders in the chart and what an outside model may see
- Switching costs close to absolute — an EHR replacement is a multi-year board-level programme, so churn at the top of this market is near zero
- Distribution economics no AI vendor can match — it sells into a relationship with contract, BAA and security review already done
- Non-purchasable data scale — Cosmos and the CoMET / Curiosity models (118M patients, 115B medical events, 151B tokens; arXiv 2508.12104)
- A capital structure with no outside pressure — no capital raised since a $70,000 start in 1979, no sponsor, no quarterly print
Weaknesses
- Total financial opacity — unaudited company-stated revenue and a magazine-estimate prior year; a 40%-margin and a 15%-margin business look identical from outside
- Key-person concentration — Faulkner founded the company in 1979, still runs it, and is in her eighties; what keeps Epic private is a doctrine plus a reported pledge, not a charter
- Zero FDA AI authorisations at vintage 2026-06-16 (canon REG-FDA-019) — the AI ships as workflow software, fast today and reclassifiable tomorrow
- Three core Sherman Act Section 2 claims survived dismissal on 9 Sept 2025 and are in discovery on an accepted payer-platform market
- The AI it sells is not mostly AI it built — ambient transcription comes from Microsoft, which sells Dragon Copilot to the same buyers
Opportunities
- Revenue cycle is the richest adjacent budget and Penny is already at 200+ organisations, distributed at zero acquisition cost into 43.7% of US acute-care hospitals
- Oracle Health's decline converts directly into pipeline — 56 hospitals and 14,676 beds lost in 2025 (canon CM-KLAS-03)
- International, where the market is weak and Epic is not — 34 hospitals / 6,580 beds in 2025, on national deployments only Epic and Oracle can bid
- Agent Factory turns customers into builders inside Epic's runtime, extending switching cost from the chart to everything built on it
- CoMET and Cosmos open a prediction franchise with payer and life-science extensions Epic has monetised lightly
Threats
- Antitrust and interoperability regulation are the only forces that compress the chokepoint; a conduct remedy on payer-platform access or API terms would touch substitution directly
- Substitution can underdeliver rather than be defeated — native AI that ships broadly but performs poorly rebuilds the opportunity it just re-addressed
- Per-use inference on per-seat contracts: if consumption outruns re-pricing, Epic absorbs the delta invisibly
- Microsoft is supplier and rival — Dragon Copilot at a reported ~33% ambient share and 100,000+ clinicians daily, reported list $369-604 per provider per month
- Key-person transition is both the largest threat to Epic's discipline and the likeliest route to the capital, filing or acquisition the board names as its falsifier
Moats, dependencies & bottlenecks
Moats
arguably the strongest in health IT An EHR replacement is a multi-year board-level programme, so top-of-market churn is near zero and 43.7% of hospitals / 56.9% of beds behaves like an annuity.
and structurally different from a tollbooth High while the record is the workflow surface Epic decides what renders in the chart and which capability is a paid module — monetising that not by taxing tenants (even at 1,000 paying vendors the whole toll is ~$1.9M, ~0.03% of $6.7B of company-stated, unaudited 2025 revenue) but by shipping the AI itself.
Strong and genuinely non-replicable subject to customer consent and data governance 118M patients, 115B medical events, 151B tokens, 78 zero-shot tasks (arXiv 2508.12104) — contributed by customers, which is why no rival can buy it and the axis a consent challenge would attack.
built organically with no acquisitions No acquired products to reconcile, so a capability ships base-wide at once — AI Charting to GA 4 Feb 2026, Penny to 200+ organisations.
Strong as a competitive asset; absolute as an investor barrier High but personal, not structural No shares, no sponsor, no quarterly print, three CEOs reported pledged to block an IPO (canon CF-EPIC-06) — a doctrine rather than an inspectable structure, so the moat most exposed to succession.
Dependencies
the sole buyer class Purchasing decisions fell ~40% in 2025 and Epic grew anyway, taking share from Oracle Health rather than new demand; saturation turns growth into a pricing question.
Supplier that is also a category competitor Ambient transcription comes from Microsoft's stack while Microsoft sells Dragon Copilot to the same buyers — Epic controls the workflow, not the speech input or its price.
ONC/ASTP, TEFCA and Carequality Regulatory constraint on the chokepoint USCDI read APIs are free under the Cures Act, so the pipe cannot be the tollbooth (canon CF-EPIC-04); the open question is whether a future rule reaches payer-platform access.
Particle Health Inc. v. Epic Systems Corp., S.D.N.Y. 1:24-cv-07174 Legal / structural All three core Section 2 claims survived dismissal on 9 Sept 2025; a conduct remedy on API terms is the half of the exposure that reaches the substitution strategy.
Key person / succession High as a tail, low as a base case The founder has run the company since 1979 and is in her eighties; the three-CEO pledge is a pledge, not a charter, and succession is the likeliest path to the board's falsifier.
Advantages
- The largest single lever on US clinical-AI distribution — 43.7% of acute-care hospitals and 56.9% of beds, gaining a fifth consecutive year (per KLAS as reported)
- Zero-cost customer acquisition for every new module: contract, BAA and security review already exist
- A near-zero gate price that keeps the ecosystem attached — and no toll revenue to defend when Epic substitutes
- Cosmos and the CoMET / Curiosity models — contributed by customers, unreproducible with capital
- One integrated codebase with no acquired products, turning a capability into a base-wide release
- A closed capital structure — freedom to price the AI privately and collect at renewal
Weaknesses
- Nothing financial can be verified — a spokesperson-stated revenue, a magazine-estimate prior year, no margin, cash or burn anywhere
- No security, no mark, no vintage — un-ownable, so even a patient investor has no instrument and no price
- Three live Sherman Act Section 2 claims in discovery on an accepted payer-platform market (S.D.N.Y. 1:24-cv-07174)
- The flagship ambient product depends on transcription supplied by Microsoft, a direct competitor for the same dollar
- Zero FDA AI authorisations at vintage 2026-06-16 (canon REG-FDA-019) — clinical claims rest on workflow framing
- Founder-bound governance, with the no-IPO commitment resting on a reported pledge rather than a structure
Bottlenecks
- Saturation arithmetic — at 43.7% of hospitals and 56.9% of beds the remaining share is smaller and harder, and the decision pool shrank ~40% in 2025
- Implementation capacity, not software — Epic-staffed deployments take months to years, so trained human capacity gates new logos
- AI unit economics inside a per-seat contract — nothing external reveals whether native AI is accretive or dilutive
- Undisclosed AI pricing stops customers benchmarking — friction that eventually invites a procurement fight or a regulator
- Speech-layer dependency — Epic owns the workflow but rents the core input from the company selling the leading competing product
- Succession: the doctrine keeping the codebase integrated and the toll near zero has no published continuity plan
Top signals & trends
Top signals
Word it 'per KLAS, as reported by…' — the 2026 report is paywalled, the percentages come from three concurring trade relays (canon CM-KLAS-01).
Growth taken out of competitors, not market expansion. Carried forward and NOT re-confirmed this run (canon CM-KLAS-02).
The mirror image of Epic's gains (canon CM-KLAS-03); the #2 is an unreported line item inside an AI-infrastructure stock.
bullish for Epic, bearish for standalone ambient and coding vendors · The price must not be asserted — a CHIO's expectation of 'comparable to Dragon Copilot' is an expectation, not a price (canon PB-055). 2027 is the first cohort renewing against a native bundle.
neutral, and the most misread number on the board · The 1,000-vendor multiplier is load-bearing; dropping it is wrong by roughly a thousandfold (canon CM-EPIC-02). Read correctly it is the evidence FOR substitution.
The strongest obtainable sourcing for a company that files nothing (canon CF-EPIC-01) — and it comes with no margin, no cash and no segmentation.
neutral, and analytically important · Vendor-stated via trade relay, not independent measurement — consistent with substitution, not evidence of effect.
bearish for any investor — and the board's actual finding · Canon CF-EPIC-06 / PB-033: unhedged and unhedgeable, which is why the board's two longs total 11% of NAV rather than 30%.
Trends
Five consecutive years of Epic gains against a pool that fell ~40% in 2025 — winner-take-most, and the loser is identifiable: Oracle Health, −56 hospitals and −14,676 beds.
positive for Epic, negative for standalone AI vendors · The board's central mechanism: gate price near zero while the ambient and revenue-cycle dollars are taken inside contracts Epic already holds.
CoMET / Curiosity on Cosmos (arXiv 2508.12104) is a different asset class from a clinical LLM — event-sequence modelling over a corpus assembled by owning the record.
Surviving Section 2 claims on an accepted payer-platform market, plus a regime that already prices read access at zero — aimed at the layer Epic monetises by substitution.
Ambient, agents and patient chat are consumption costs on annual per-user pricing; Epic faces it invisibly, with no statement in which the effect could appear.
Doximity Scribe at $0 and athenaAmbient at no additional cost sit alongside Epic AI Charting at an undisclosed price (canon PB-055) — Epic cannot price far above a free substitute indefinitely.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
Transcription supply plus a major hosting option — consequential precisely because it is also the largest ambient competitor.
Chronicles has long been reported to run on InterSystems' platform (not company-confirmed) — a third-party core database beneath the most entrenched US clinical record.
Inference for AI Charting, Art, Penny, Emmie and Agent Factory, plus hosting; neither volume nor unit cost is observable.
Large integrated delivery networks and academic medical centres The core franchise; bed share (56.9%) well above hospital share (43.7%) is the tell that Epic's customers skew large.
KLAS's 2026 qualitative finding was 'Epic's Footprint Expands Among Small Health Systems' — the tier defended by Meditech and TruBridge.
190M+ MyChart users and 325M+ patients with an Epic record (canon PB-040); Emmie puts an assistant on the largest direct-to-patient surface in US healthcare.
The exact product market the court accepted in the surviving Particle Health claims — the most legally exposed surface Epic operates.
The #2 US acute EHR at 21.9% and the source of most of Epic's gains — 56 hospitals and 14,676 beds lost in 2025, a third year as largest net share loser, Millennium lowest-scoring across all three organisation sizes in 2026 Best in KLAS (canon CM-KLAS-03). Health is not a reported segment inside a $67.4B revenue base, so none of it is expressible in the equity.
Private. 14.7% of US acute-care hospitals (per KLAS as reported), concentrated in the community and rural tier KLAS flags Epic as newly expanding into.
Private, sponsor-owned by Bain Capital and Hellman & Friedman. The sharpest price attack in ambulatory: athenaAmbient ships at no additional cost inside the core athenaOne fee — the only confirmed zero-incremental-fee EHR-native ambient bundle.
Listed. 7.6% of US acute-care hospitals on the canon's carried-forward, not-independently confirmed basis (canon CM-KLAS-02); rural focus, more exposed to Epic's downmarket push than able to counter it.
Private, Harris/Constellation-owned. 2.9% on the same carried-forward basis — a maintained installed base and a source of Epic replacement opportunities rather than a threat.
Epic's transcription supplier and the largest incumbent in the category Epic is substituting into: Dragon Copilot at a reported ~33% ambient share and 100,000+ clinicians daily, reported list $369-604 per provider per month.
Private. The best-capitalised standalone ambient vendor — $5.3B post-money on a $300M Series E led by a16z and Khosla in June 2025, a fourteen-month-old negotiated mark at 7 August 2026, plus a $316M extension in April 2026 at an undisclosed price. It distributes through Epic-based health systems and carries a reported Epic equity-and-revenue-share relationship.
Listed revenue-cycle vendor and the clearest read-through on Penny: Waystar reaches its customers THROUGH Epic-based health systems, so Epic sells the same job into the same buyer at zero acquisition cost. The board's falsifier is a Penny-attributed client loss at a named health system.