GE HealthCare
Capital equipment plus an annuity: a scanner is sold or leased, then the installed base yields multi-year service, consumables, upgrades and increasingly subscription AI. Three reportable segments as of Q2 2026 — Advanced Imaging Solutions (AIS, the Imaging and Advanced Visualization Solutions businesses combined), Pharmaceutical Diagnostics (PDx, sold per dose) and Patient Care Solutions (PCS, monitoring and life support). AI reaches the customer as a feature of a machine or a paid upgrade to one, never as a separately reimbursed product with its own payment code — the structural fact governing every conclusion below.
The thesis on this name
State of AI for Healthcare
The direct test of the clearance-stock leg, and it fails on GE's own filings. GEHC is the largest single holder of FDA AI authorisations on earth — 130 of the 1,524 on the list (8.5%), being 107 filed under a GE name plus 23 that arrived with acquired filers — and it earns corporate economics, not software economics: $21.27B TTM revenue (+6.5%), $1.99B TTM net income (−11.4%), 15.7x trailing earnings at $68.02 / $30.72B (31 Jul 2026), with the Advanced Imaging Solutions (AIS) segment at a 13.9% EBIT margin in Q2 2026 on $3,771M of segment revenue. Two facts kill the count as a value proxy. First, the stock is mostly not software: 33 of the 130 sit under a software product code and the other 97 are registrations of scanner — a split derived from an openFDA product-code mapping this board has not independently re-verified, so treat it as derived-unverified rather than primarys that embed an algorithm. Priced at the FY2026 standard 510(k) user fee of $26,067, the entire 130-authorisation stack cost $3.39M nominal — 0.016% of one year's revenue. A filing fee, not a rent. Second, GE's own capital allocation says distribution is the asset: it paid $2,293M in cash, net of cash acquired, for Intelerad — an outpatient imaging-workflow company holding a verified ZERO entries on the FDA AI list — closing 18 Mar 2026, against $357M for the two AI-clearance acquisitions priced in its current 10-K and 10-Q on that same net-of-cash basis (MIM Software $259M, icometrix ~$98M), which carry 12 of the 23 authorisations it acquired; the other 11 carry nothing in that figure. No multiple is drawn between those two amounts: eleven of the 23 acquired authorisations — Caption Health 6, Bay Labs 2, BK Medical 2, Spectronic 1 — carry nothing in that figure, and BK Medical's own price is known: $1.45B cash, completed 21 Dec 2021 under pre-spin GE, roughly 4x the entire $357M. Meanwhile the four largest non-OEM holders of genuine AI-software authorisations — Aidoc 33, RapidAI 17, Viz.ai 12, Qure.ai 9 — hold 71 software-code authorisations between them, more than GE (33) and Siemens (32) combined, and every one of them is private. Own GEHC as an imaging OEM if you want one. It is not an AI-in-healthcare expression.
State of AI for Healthcare
The name that disproves the clearance-stock thesis with its own chequebook: GE paid $2,293M in cash, net of cash acquired, for Intelerad — which holds zero FDA AI authorisations — against $357M for the two AI-clearance acquisitions priced on that same net-of-cash basis in its current 10-K and 10-Q (MIM Software $259M, icometrix ~$98M), which carry 12 of the 23 authorisations GE acquired; the other 11 carry nothing in that figure.
Earnings, margins, COGS & capex
A $21.27B imaging OEM the market prices as an imaging OEM. The board's interest is that it settles an argument, not that it offers a return. The argument is whether a stock of FDA AI authorisations is an economic moat, and GEHC is the maximal case: 130 authorisations, 8.5% of the entire FDA AI-Enabled Medical Device List — ahead of Siemens Healthineers' 95, Philips' 45 and Canon Medical's 43 — earning 13.9% segment EBIT in the segment that holds them on a 39.5% consolidated gross margin. Three facts explain the non-conversion. It is cheap: 130 at the FY2026 standard 510(k) fee of $26,067 is $3,388,710 nominal, 0.016% of one year's revenue, and 96.2% of the list is 510(k) clearance rather than PMA approval — substantial equivalence, not demonstrated clinical benefit. It is mostly not software: on a derived-unverified product-code mapping 33 of the 130 are software-code and 97 are registrations of scanners that embed an algorithm; across the eight imaging OEMs only 105 of 403 (26%) are software-code. And GE's revealed preference prices the two assets: $2,293M in cash, net of cash acquired, for Intelerad — an outpatient imaging-workflow business with a verified zero entries on the FDA AI list — against $357M for the two AI-clearance acquisitions priced on that same basis, carrying 12 of the 23 authorisations GE acquired; the other 11 carry nothing in that figure. What is improving is orthogonal to AI (record orders, PDx at 14.6% organic on 29.6% margins, margin guided up); what is broken is also orthogonal (PCS down 13.5% organically into negative EBIT, China guided down, a Q2 cash figure positive only on tariff refunds).
Income statement — where each revenue dollar goes
% of revenueOf every $1 of revenue, ~60¢ is cost of goods and ~27¢ operating expense, leaving ~13¢ of operating profit (~11¢ net).
Revenue trend
Margins
+90bp YoY on volume and price; quarterly, segment-level, EBIT — the only margin that belongs beside the 130 count
AIS below the company average, PDx far above at 29.6%, PCS negative at (3.8)%
10–40bp of guided expansion; full-year, company-wide, adjusted — a different scope from the 13.9%
Slightly down; a hardware gross margin
Down ~54bp; the gap to the adjusted guide is acquired-intangible amortisation, widening with each acquisition
Up YoY, with $129M of recognised tariff refunds inside it
~$1.6B guided; Q2 delivered $68M and needed $107M of refunds
COGS structure
Overwhelmingly physical: CT gantries, MR magnets and cryogenics, X-ray tubes and detectors, transducers, semiconductors, and the field-service organisation. Deep-learning reconstruction (AIR Recon DL; True Definition DL2, 510(k) cleared 2 Apr 2026) is engineering cost amortised into the machine price with inference on the scanner's own compute — capitalised R&D, not per-study COGS. Acquired software carries real amortisation: MIM Software's allocation put $48M into developed technology, 17.0% of the $283M total purchase price or 18.5% of the $259M net-of-cash consideration, and Intelerad added $845M of other intangibles beside $1,629M of goodwill — exactly what the adjusted-EBIT bridge excludes. PDx is chemistry and logistics, priced by isotope half-life and delivery radius. Tariffs are a live input: $129M of refunds recognised in Q2 2026 net income, against USMCA certification and local-for-local manufacturing as mitigation.
Capex
$522M TTM, ~2.5% of revenue (FY2025 $482M): plant, the PDx radiopharmacy network, internal-use software. The meaningful capital is M&A and buybacks — $2,293M of cash for Intelerad, $200M of repurchases (3.3M shares) in Q2 with $500M remaining, and a token $0.035 quarterly dividend. Against ~$8.0B of net debt and ~$1.6B of guided free cash flow the acquisition budget is constrained unless leverage rises, which matters because the board's falsifier runs through whether Intelerad converts into segment margin, not whether GE buys another algorithm.
Latest earnings
Beat on the headline and on orders; the quality of the beat is the argument. $129M of refunds ran through net income, free cash flow was positive only because $107M landed in cash, and PCS swung to negative segment EBIT on self-inflicted component shortages. Guidance was reaffirmed rather than raised — management did not treat the beat as a run-rate.
FY2026, reaffirmed 29 Jul 2026: organic revenue +3.0–4.0%; adjusted EBIT margin 15.4–15.7% (10–40bp above FY2025's 15.3%); adjusted EPS $4.80–5.00; free cash flow ~$1.6B. Refunds on 2025 tariff amounts are excluded from adjusted figures and the guide assumes continued China decline. At $68.02 that is 13.6–14.2x guided adjusted EPS, so the board's 13.15x forward multiple — a consensus estimate implying ~$5.17 — reaches past the guided year.
- FDA AI authorisations
- 130 — 107 filed under a GE name plus 23 filed by companies GE has since bought; 8.5% of the 1,524-row FDA list (REG-FDA-010/-012, list vintage 2026-06-16)
- Software-code share of that stock
- 33 of 130 (25%); the other 97 (74.6%) are scanner registrations — REG-FDA-015, DERIVED-UNVERIFIED
- Nominal 510(k) fee cost of the whole stock
- $3,388,710 — 130 at the FY2026 standard fee of $26,067 = 0.016% of TTM revenue (CF-GEHC-19)
- Intelerad
- $2,293M in cash, net of cash acquired, closed 18 Mar 2026; goodwill $1,629M, other intangibles $845M; a verified ZERO FDA AI authorisations (CF-GEHC-10, REG-FDA-019)
- AI-clearance acquisitions priced on that same basis
- $357M — MIM Software $259M and icometrix ~$98M, both net of cash — carrying 12 of the 23 authorisations GE acquired; the other 11 carry nothing in that figure (CF-GEHC-15)
- The 23 acquired authorisations
- MIM Software 7 · Caption Health 6 · Bay Labs 2 · icometrix 5 · BK Medical 2 · Spectronic Medical 1 = 17.7% of the 130 (REG-FDA-014)
- Outside the 23
- Intelligent Ultrasound's 2 sit in the fully-enumerated family of 132, not the 130 headline; its consideration is a derived ~$54M residual appearing in no filing (REG-FDA-013, CF-GEHC-13)
- Orders and backlog
- $23.9B backlog, organic orders +11.1%, book-to-bill 1.15x — all post-spin highs
- Capital returned and tariffs
- 3.3M shares repurchased for $200M with $500M remaining; $0.035 quarterly dividend; $129M of tariff refunds in net income, $107M in free cash flow
Growth drivers
- Record orders and backlog — +11.1% organic orders, 1.15x book-to-bill, $23.9B backlog — post-spin highs, and a hospital capital-cycle signal rather than an AI one.
- PDx and the Flyrcado ramp — +14.6% organic at 29.6% segment EBIT, 545 doses reported in the week ending 24 Jul 2026 (~40% above April) against a stated target of $500M or more in annual revenue by 2028.
- Premium imaging refresh with AI-enabled new products — Feb 2026 510(k) clearances of SIGNA Sprint with Freelium (a 1.5T sealed magnet cutting helium dependence), SIGNA Bolt (3T) and SIGNA One, plus True Definition DL2 for CT. Management attributes AIS margin expansion to 'higher-margin AI-enabled NPIs'.
- Intelerad and the reading pipeline — $2,293M buys the outpatient worklist, PACS and reporting layer where AI output is consumed and recurring revenue lives; already contributing to service revenue growth. A workflow bet, not an AI bet.
- Subscription attach — management cites 'a subscription model, which brings recurring revenue' but discloses no software or recurring percentage, so the driver cannot be sized — the absence is itself the finding.
- China stabilisation, currently a headwind — ~10% of FY2025 revenue, guided to decline again in 2026 as volume-based procurement broadens. Any inflection is upside and outside GE's control.
Reported financials — SEC EDGAR
Audited GAAP figures pulled from SEC filings · latest filing 2026-02-04. The audited primary-source spine — not financial advice.
Revenue — annual (GAAP)
Margins & balance sheet — FY’25
Bull & bear
Not that GE HealthCare is an AI company — that it is a cheap, improving industrial with the best order book since the spin, its highest-quality growth hiding in radiopharmaceuticals, and an underpriced option on owning the imaging reading pipeline. At $68.02 it trades at 13.6–14.2x its own guided FY2026 adjusted EPS, ~24% below the 52-week high, with margin guided up and $23.9B of backlog to convert. Everything the bear says about the authorisation count can be true and the stock still works, because nobody is paying for the authorisation count.
- The order book leads revenue and is the strongest signal in the business: +11.1% organic orders, 1.15x book-to-bill, $23.9B backlog, no reported pullback in US imaging capital spend. Reported 3.5% organic understates what customers have already committed to buy.
- PDx is better than the market credits: +14.6% organic at 29.6% segment EBIT, Flyrcado at 545 doses in the week ending 24 Jul 2026 against a $500M-plus 2028 target. A per-dose diagnostic with its own reimbursement is the payment identity AI-in-a-scanner lacks, and GE already owns it.
- Margin is guided up and mix helps: FY2026 adjusted EBIT margin 15.4–15.7% against FY2025's 15.3%, with AIS already +90bp YoY. Two of three segments are expanding; the third's problem is component supply, which is fixable.
- Intelerad could re-rate the name and it is early — it puts GE where studies are read and recurring software revenue lives, and already contributes to service revenue growth. AIS segment operating margin sustained above 20% with Intelerad named as the driver is the board's own falsifier.
- Valuation and returns give a floor: 15.67x trailing, 13.15x forward on a consensus estimate, 24% below the 52-week high, $200M of buybacks with $500M remaining, and a workable ~$8.0B net debt load against ~$1.6B of guided free cash flow.
- GE is the likeliest consolidator of the assets that DO have software economics: Aidoc, RapidAI, Viz.ai and Qure.ai are all private and there is no US-listed vehicle for radiology AI software, so an OEM willing to write a $2,293M cheque buys at private clearance prices rather than public software multiples.
GEHC is the sector's best proof that FDA AI authorisations are a filing fee rather than a franchise, and the company itself knows it. It holds more AI authorisations than anyone on earth — 130, 8.5% of the list — and earns 13.9% segment EBIT in the segment that holds them, materially below its closest peer's imaging margin. The whole stock cost $3.39M in nominal user fees, 0.016% of a year's revenue, and roughly three-quarters of it is scanner registrations. When GE spent real money it spent $2,293M in cash on a workflow company with zero AI authorisations, against $357M for the two AI-clearance acquisitions it prices on the same net-of-cash basis. Owning GEHC is owning an imaging OEM with a China problem, a tariff-flattered cash quarter and a broken monitoring segment — a defensible trade on a different board, not health-AI economics. This board holds it at 0% of NAV with high conviction.
- The moat is a filing fee: 130 authorisations at $26,067 is $3,388,710 nominal, 0.016% of one year's revenue, and 96.2% of the entire FDA AI list is 510(k) clearance rather than De Novo or PMA. A barrier a competitor clears with a filing budget is a barrier to entry, not a margin pool.
- The count is mostly hardware: 33 of GE's 130 are software-code and 97 are scanner registrations; across the eight OEMs, 105 of 403 (26%) — derived-unverified, but internally consistent everywhere it is used. On that same basis the four largest private independents hold 71, more than GE (33) and Siemens (32) combined. The leaderboard reverses the moment you ask which authorisations are software.
- GE's chequebook prices the two assets: $2,293M in cash, net of cash acquired, for Intelerad — verified zero entries on the FDA AI list, closed 18 Mar 2026 — against $357M for the two AI-clearance acquisitions priced on that basis in its current 10-K and 10-Q, MIM Software $259M and icometrix ~$98M, carrying 12 of the 23 authorisations GE acquired; the other 11 carry nothing in that figure. No programme-wide ratio can be honestly computed — BK Medical's $1.45B was disclosed by pre-spin GE and appears in no GEHC filing, and Caption Health, Bay Labs and Spectronic Medical were never priced — but the two clean amounts say what the acquirer believes.
- Even the software GE buys is mostly not software: MIM's allocation put only $48M into developed technology — 17.0% of the $283M total purchase price, or 18.5% of the $259M net-of-cash consideration — against $189M of goodwill and $52M of customer-related intangibles.
- The quarter was propped up in three places at once: $129M of tariff refunds in net income, $107M of them inside a $68M free cash flow figure, and a segment down 13.5% organically into negative EBIT. Guidance was reaffirmed, not raised. Strip the refunds and it is a 3.5% organic grower with a cash-negative core.
- Structural exposures AI does not offset: China at ~10% of revenue and guided down as VBP broadens and United Imaging scales at 40 authorisations; ~$8.0B of net debt after Intelerad; a 39.5% hardware gross margin capping how software-like this P&L can look; and a 13.15x 'forward' multiple reaching past the company's own $4.80–5.00 guidance to a consensus estimate of ~$5.17.
What it is worth
Market marks at the board's vintage plus a guidance-anchored earnings check — no DCF and no target price, because the board holds this name at 0% of NAV and a price target would imply an intention to own it. At the 31 Jul 2026 close of $68.02 the market capitalisation is $30.72B, the trailing multiple 15.67x and the forward multiple 13.15x on an unnamed third-party consensus estimate (CF-GEHC-02 requires that estimate be labelled as such wherever it appears). The cross-check that matters: 13.15x implies ~$5.17 of EPS, ABOVE the company's own reaffirmed FY2026 guidance of $4.80–5.00, so the forward multiple reaches into FY2027; on the guided year the same price is 13.6–14.2x. Scenario weights are 25% bull / 55% base / 20% bear, with no price targets attached. NOTE, a canon ruling rather than an editorial choice: no multiple is stated anywhere here for GE's distribution-versus-clearance acquisition comparison. That denominator cannot be assembled on a single basis while BK Medical's $1.45B sits outside every GEHC filing and Caption Health, Bay Labs and Spectronic Medical were never priced. The two clean amounts — $2,293M and $357M, both net of cash acquired, both verbatim in GEHC filings — carry the argument without arithmetic.
The $2,293M Intelerad price is not recovered in segment margin and the clearance portfolio proves to be what the evidence says it is — a barrier to entry, not a margin pool. AIS stays in the low teens through FY2027 while $845M of acquired intangibles amortise against it; tariff refunds do not repeat and the year-over-year impact turns negative; China declines faster than guided as VBP broadens and United Imaging takes share; the PCS supply failure persists; and free cash flow misses the $1.6B guide. Roughly $8.0B of net debt then constrains the buyback and the acquisition strategy simultaneously. The 52-week low of $58.75 is the visible reference on that path. 20% probability, no price target.
Advanced Imaging Solutions stays roughly where it is — 13.9% segment EBIT on $3,771M of quarterly revenue, +90bp YoY, against 14.2% for the total company — while consolidated adjusted EBIT margin grinds to the guided 15.4–15.7% on price, mix and tariff mitigation. Organic growth lands in the 3–4% band, free cash flow reaches ~$1.6B on a back-half conversion, and the authorisation count keeps growing without changing the margin structure. An imaging OEM, priced as one, at 13.6–14.2x its own guided earnings. 55% probability, no price target.
Intelerad converts: AIS segment operating margin sustains above 20% with management naming software attach or Intelerad as the driver — the board's stated falsifier, which would make GEHC an ownable workflow-software story rather than an OEM. Alongside it, the $23.9B backlog converts at the guided 3–4% organic with adjusted EBIT margin at or above the top of the 15.4–15.7% range, PDx compounds toward the $500M-plus Flyrcado target, the PCS component problem resolves and China stops declining. A move back toward the 52-week high of $89.77 is arithmetically available on those facts. 25% probability, no price target — a workflow and industrial-cycle outcome rather than a validation of the AI thesis.
Framing, not a recommendation, and not investment advice. The board's position is unchanged: avoid, 0% of NAV, high conviction, with the sizing note that owning GEHC is owning an imaging OEM at roughly 13x forward earnings — a defensible trade on a different board, not health-AI economics. The exclusion IS the finding: a 0% weight is the price the board pays to say an FDA AI authorisation count does not produce software margin. There is no US-listed vehicle for radiology AI software — Aidoc 33, RapidAI 17, Viz.ai 12 and Qure.ai 9 hold 71 software-code authorisations between them, more than GE (33) and Siemens (32) combined, and all four are private — and GEHC is not a substitute for one. The catalyst worth watching is the FY2026 10-K, expected around February 2027: AIS segment margin after a full year of Intelerad, a workflow-integration read rather than an AI read.
SWOT
Strengths
- The largest installed base and regulatory footprint in imaging — 130 authorisations, 8.5% of the list, ahead of Siemens' 95 and Philips' 45. Not chargeable, but not quickly assembled, and every algorithm ships into machines already in the room.
- A record order book that leads revenue — +11.1% organic orders, 1.15x book-to-bill, $23.9B backlog, with no reported pullback in US imaging capital spend.
- PDx is a genuinely superior business inside an imaging OEM — +14.6% organic at 29.6% segment EBIT, Flyrcado at 545 doses in the week ending 24 Jul 2026 against a $500M-plus 2028 target. Per-dose radiopharmaceuticals carry the payment identity GE's AI lacks.
- Real AI engineering monetised as a hardware premium — deep-learning reconstruction improves throughput and image quality on the installed base, and management names AI-enabled new products as a driver of AIS margin expansion.
- Capital-allocation optionality — ~$8.0B of net debt against ~$1.6B of guided free cash flow still absorbed a $2,293M acquisition and $200M of buybacks, making GE the likeliest acquirer of the private radiology-AI assets the board says have no listed vehicle.
Weaknesses
- The AI leadership is a hardware fact dressed as a software one — 33 of 130 software-code against 97 scanner registrations, 105 of 403 (26%) across the eight OEMs, and 96.2% of the whole list being 510(k) clearance rather than PMA.
- Segment margin is low for the category. AIS earned 13.9% — Siemens reported an Imaging margin of 21.6% before tariff refunds in its Q3 FY2026 quarter (26.5% including a 370bp refund contribution) — a different basis and fiscal calendar, so not like-for-like, but the gap's direction has persisted since the spin.
- PCS is broken and self-inflicted — −13.5% organic into a negative (3.8)% segment EBIT because critical components were short, roughly $100M of revenue against the prior-year quarter.
- Free cash flow is thin, seasonal and flattered — $68M in Q2 with $107M of refunds inside it — and the gap between 12.86% GAAP operating margin and the adjusted guide widens with each intangible-heavy deal.
- No disclosed software, recurring or AI revenue line, so the mechanism the whole bull case rests on cannot be sized from the filings.
Opportunities
- Convert Intelerad into the reading pipeline — the one route to software economics. The board's falsifier is AIS segment operating margin sustained above 20% with Intelerad named as the driver.
- Buy the private radiology-AI stock at what the market has shown it is worth: Aidoc 33, RapidAI 17, Viz.ai 12 and Qure.ai 9 hold 71 software-code authorisations between them — more than GE (33) and Siemens (32) combined — and every one is private.
- Scale PDx into the profit engine — a $500M-plus 2028 Flyrcado target at 29.6% segment margins is capital GE can deploy without bidding against software valuations.
- Monetise the installed base with upgrades rather than new boxes — a deep-learning reconstruction sold into an existing scanner is high-incremental-margin revenue that does not wait on a capital cycle.
- Tariff normalisation — USMCA certification of eligible imports and local-for-local manufacturing cut a headwind worth hundreds of millions, and a lower net 2026 impact would flow almost entirely to margin.
Threats
- The AI layer decouples from the scanner. Aidoc — 34 authorisations, 33 software-code, across nearly 2,000 hospitals — lets a health system buy AI once and run it over a mixed fleet, at which point GE's 130 is a checklist rather than distribution.
- China: ~10% of FY2025 revenue (~$2.03B, per an aggregator breakdown reconciling to the primary total), guided down again in 2026, with VBP broadening and United Imaging already at 40 authorisations competing on price in GE's core modalities.
- Tariffs: the year-over-year impact is neutral only when $129M of refunds are counted, and refunds are non-recurring and policy-dependent on a 39.5% gross margin.
- Hospital capital budgets — a $23.9B backlog is only as good as customers' ability to fund it, and a Medicare rate cycle or coverage shock shows up in orders first.
- Paying up for workflow and not earning it back — $2,293M plus $845M of intangibles and $1,629M of goodwill now sit against a segment earning 13.9%, competing with Sectra and Pro Medicus, who run software margins there with zero FDA AI authorisations.
Moats, dependencies & bottlenecks
Moats
7–12 year replacement cycles The genuine moat, and it has nothing to do with AI: a hospital on GE CT and MR buys GE service, GE upgrades and by default GE's next machine. This is what the $23.9B backlog actually measures.
107 GE-named plus 23 acquired filers, 8.5% of the list) Real as a barrier to entry, weak as an economic moat High as a count, low as a rent The board's central finding. $3,388,710 of nominal FY2026 510(k) fees bought the whole thing, 96.2% of the list is 510(k) clearance, and on a derived-unverified mapping only 33 of the 130 are software-code. A new entrant must clear it; the incumbent cannot charge for it.
Moderate to strong The expense is not the $26,067 fee but the QMS, post-market surveillance, change-control plans and multi-jurisdiction registration needed to maintain 130 authorisations — a fixed-cost advantage the startup bears as cost, not one GE exercises as pricing power.
MIM, icometrix) purchased, not earned execution-dependent The only moat here that could produce software economics. GE paid $2,293M in cash to enter a surface where Sectra and Pro Medicus already run software margins with zero FDA AI authorisations. Verdict pends the FY2026 10-K's AIS margin, expected around Feb 2027.
manufacturing, isotope logistics and its own payment identity A radiopharmaceutical has a payment code, a half-life and a delivery radius; all three are barriers and none is a software barrier. GE's most attractive moat is the one with no AI attached.
GPO contracting, fleet agreements and tender qualification favour the incumbent and are the most exposed to price competition — China's volume-based procurement is this moat being dismantled by a monopsony buyer.
Dependencies
Every dollar of AIS revenue is a discretionary capital purchase. The $23.9B backlog is a claim on those budgets, not a guarantee, and it moves with payer mix and coverage policy before technology.
Geographic concentration ~10% of FY2025 revenue (~$2.03B on a ~$20.63B base, per an aggregator breakdown reconciling to the primary total), guided down again in 2026 as VBP broadens. United Imaging — 40 authorisations, Shanghai-listed and outside this board's US-first mandate — competes on price in CT and MR.
$129M of refunds in Q2 2026 net income, $107M in free cash flow, $23M in adjusted EBIT; management calls the year-over-year impact neutral only when refunds are included. USMCA certification and local-for-local manufacturing are partial mitigation.
and currently realised PCS fell 13.5% organically into a negative (3.8)% segment EBIT explicitly because critical components were short — the clearest evidence that the binding constraint is physical rather than algorithmic.
130 authorisations on a list that is 96.2% 510(k), with a strategy that ships model updates. A tightening raises GE's maintenance cost — and a startup's entry cost by more, the one way this cuts in GE's favour.
The best-margin segment depends on cyclotron capacity and a delivery radius set by half-life; the Flyrcado ramp fails on logistics before it fails on demand.
Strategic / capital allocation $2,293M of cash, $1,629M of goodwill and $845M of other intangibles now sit inside a segment earning 13.9%, with no synergy targets, milestones or software-revenue disclosure given. The re-rating case and the board's falsifier both run through this.
Advantages
- The largest FDA AI authorisation stock in the world at 130 — 8.5% of the entire list — which no competitor assembles quickly even if it cannot be charged for.
- A global installed base and field-service network that turns every new algorithm into an upgrade sold into machines already in the room.
- A record $23.9B backlog at 1.15x book-to-bill, the strongest forward demand signal since the 2023 spin.
- PDx at 29.6% segment EBIT on 14.6% organic growth, with a ramping product that has its own reimbursement.
- Balance-sheet capacity to buy the private AI assets the public market cannot access — a $2,293M cheque in 2026 alongside $200M of buybacks.
- Differentiated deep-learning reconstruction (AIR Recon DL, True Definition DL2, SIGNA One) monetised as a premium on new products.
Weaknesses
- 13.9% AIS segment EBIT — the segment holding the authorisations is structurally the weakest and sits below the peer's imaging margin.
- 39.5% consolidated gross margin caps how software-like this P&L can look regardless of how many algorithms ship inside the boxes.
- No disclosed software, recurring or AI revenue line, so the bull mechanism cannot be underwritten from the filings.
- Patient Care Solutions in outright decline with negative segment EBIT on a self-inflicted supply failure.
- Roughly $8.0B of net debt after Intelerad against thin, seasonal free cash flow.
- China at ~10% of revenue and guided down, with VBP broadening and a well-funded domestic competitor scaling in GE's core modalities.
Bottlenecks
- Component and semiconductor supply — already binding, and the direct cause of a 13.5% organic decline and negative segment EBIT in Patient Care Solutions.
- Hospital capital-approval cycles, sitting between 11.1% order growth and recognised revenue and stretching $23.9B of backlog over years.
- Installation and site readiness — an MR suite is a construction project, and revenue waits on the building, not the algorithm.
- Radiopharmacy and cyclotron capacity, capping how fast Flyrcado reaches the $500M-plus 2028 target regardless of demand.
- Helium and cryogenics for superconducting magnets — why a sealed-magnet product like SIGNA Sprint with Freelium is a supply-chain answer as much as a clinical one.
- Integration bandwidth: Intelerad, MIM Software and icometrix absorbed into a hardware company at once.
- Free cash flow seasonality — $68M in Q2 including $107M of refunds against a ~$1.6B annual guide.
Top signals & trends
Top signals
The most informative capital-allocation fact on this board: the acquirer's own pricing of distribution versus clearance stock, on one basis, both legs verbatim in GEHC filings. No programme-wide ratio is asserted — BK Medical's $1.45B was disclosed by pre-spin GE and appears in no GEHC filing, and Caption Health, Bay Labs and Spectronic Medical were never priced (CF-GEHC-10/11/12/15/18).
The strongest genuine datum in the Q2 print, and a hospital capital-cycle signal rather than an AI one. It leads revenue by several quarters and is why guidance was reaffirmable despite a soft cash quarter.
The underlying quarter consumed cash; with ~$1.6B guided, the whole conversion is back-half weighted on an unwind not yet demonstrated.
Management called it operational fulfilment failure, not demand — the more worrying attribution for a manufacturer.
The best business in the portfolio has no AI attached, and it is the only GE line with its own payment identity — the feature the board's thesis says determines whether margin exists.
DERIVED-UNVERIFIED: the openFDA product-code-to-21-CFR mapping is unpublished and could be neither confirmed nor falsified in audit, though internally consistent at every site carrying it (REG-FDA-015/-021). The 130 count is primary; the 25%/74.6% split is not.
Organic +3.0–4.0%, adjusted EBIT margin 15.4–15.7%, adjusted EPS $4.80–5.00, free cash flow ~$1.6B. Management treated neither $129M of refunds nor an 11.1% order quarter as a run-rate.
Evidence the authorisation machine keeps running and that new products carry AI premiums — and equally evidence of what these are: 510(k) clearances of scanners and reconstruction software, not PMA approvals with an outcome claim, and not products with their own payment code.
A modest but real return of capital supporting the valuation floor; the token dividend and the debt load together say the balance sheet is held for acquisitions rather than distributions.
An unnamed third-party consensus estimate reaching into FY2027 (CF-GEHC-02 requires it be labelled an estimate wherever it appears). On guided numbers the same $68.02 is 13.6–14.2x.
Trends
neutral for GE, negative for the AI-as-a-moat thesis · GE monetises AI as a premium on new products and as paid upgrades to installed machines — a real, durable mechanism priced at hardware margins. It is why 130 authorisations coexist with a 13.9% segment margin.
Aidoc — 34 authorisations, 33 software-code, across nearly 2,000 hospitals analysing more than 60 million patient cases a year — lets a system run one AI layer over a mixed fleet. If procurement moves to the layer, the OEM's count stops being a distribution advantage.
GE is the leading example: MIM Software (2024), Intelligent Ultrasound's clinical-AI business (2024), icometrix (Nov 2025), Intelerad (Mar 2026). Clearance stock trades as M&A currency rather than rent, and the pattern favours whoever owns the pipeline.
positive if executed · Sectra and Pro Medicus run software economics on the exact surface GE paid $2,293M to enter, and hold zero FDA AI authorisations between them. The prize is real; so is the incumbency GE is buying into.
~10% of FY2025 revenue, guided down again as VBP broadens, with United Imaging at 40 authorisations competing on price at home and increasingly abroad — a monopsony buyer dismantling an incumbency moat.
PDx at 14.6% organic and 29.6% segment EBIT with Flyrcado ramping toward $500M-plus by 2028 — a per-dose product with its own payment code, structurally better than a bundled algorithm.
negative, partially mitigated · $129M of refunds make the year-over-year impact neutral on management's own framing, which means the gross impact is not. USMCA certification and reduced US–China shipment reliance lower but do not remove the exposure.
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.
detector and X-ray tube suppliers The binding physical input: component shortages drove PCS to a 13.5% organic decline and negative segment EBIT, and long qualification cycles make substitution slow.
Superconducting MR magnets need liquid helium in a market with recurring shocks; GE's answer is the sealed-magnet SIGNA Sprint with Freelium, 510(k) Feb 2026.
Accelerated-compute and cloud infrastructure vendors Deep-learning reconstruction and SIGNA One run on accelerated silicon at the scanner edge and cloud for enterprise deployments; GE discloses no compute cost line.
cyclotron operators and contrast-media feedstock suppliers PDx is chemistry and logistics — isotope availability and half-life set the delivery radius, the real gate on the Flyrcado ramp.
Contract manufacturers and the international plant footprint The lever behind the tariff response: USMCA certification of eligible imports and local-for-local manufacturing.
US hospital systems and integrated delivery networks The core buyer of premium CT, MR and interventional systems and the counterparty behind the $23.9B backlog; no pullback in US imaging capital spend reported.
Outpatient imaging centres and radiology chains The fastest-shifting site of service and the constituency Intelerad serves — and where the customer becomes the competitor, since RadNet operates centres and holds 29 authorisations through DeepHealth.
The MIM Software franchise — contouring, dosimetry, planning — extended by the MIM Contour ProtégéAI+ 2.0 clearance in June 2026. Seven of the 23 acquired authorisations came with MIM, the largest single block.
Public tender and national procurement authorities, including China's VBP Price-setting buyers with monopsony power; China is ~10% of FY2025 revenue and guided down again.
Cardiology practices and nuclear-medicine departments The per-dose buyer of contrast media and radiopharmaceuticals including Flyrcado — the only GE customer relationship transacted against a payment code rather than a capital budget.
The closest peer and sharpest benchmark: 95 FDA AI authorisations on the board's one-basis roll-up (90 Siemens-named plus Varian's 5), second only to GE. Coverage of its Q3 FY2026 quarter reported Imaging revenue of €3.01B (+2.3%) at a 21.6% margin before tariff refunds versus 23.7% a year earlier — a different basis, segment definition and fiscal calendar, so not like-for-like, but the persistence of the imaging-margin gap is the most uncomfortable single fact for a GEHC bull.
45 FDA AI authorisations. Competes across imaging, monitoring and enterprise informatics — the same three-legged shape as GE, and the direct rival in the Patient Care Solutions segment GE just shrank 13.5% organically.
43 FDA AI authorisations. Credible CT and MR competitor with its own deep-learning reconstruction, funded by a parent that does not need imaging to carry a margin target, and hardest on the premium CT refresh driving AIS growth.
40 FDA AI authorisations — fifth on the entire list and ahead of every independent AI vendor except Aidoc. Price-led challenger in CT, MR and molecular imaging, dominant in the Chinese tender market that is ~10% of GE's revenue and guided down. Shanghai STAR-listed and outside this board's US-first mandate: a competitive fact, not an instrument.
Absorbed Hitachi's diagnostic imaging business; competes in CT, MR, endoscopy and enterprise imaging with its own AI platform, taking share in value-tier imaging and in the PACS layer GE just bought into.
21 FDA AI authorisations, strongest in ultrasound and mobile CT — ultrasound being where GE's own acquired AI (Caption Health, Bay Labs, Intelligent Ultrasound) is concentrated.
The strategic threat that matters most, and not an OEM. 34 FDA authorisations of which 33 sit under software product codes — all 510(k)-class clearances, so no claim of demonstrated clinical efficacy attaches to the count — across nearly 2,000 hospitals analysing more than 60 million patient cases a year. A $150M Series E was announced 29 April 2026 led by Growth Equity at Goldman Sachs Alternatives with General Catalyst, SoftBank Vision Fund 2 and NVentures; more than $500M raised in total and NO valuation disclosed — do not underwrite one and do not infer one from the raise.
29 FDA AI authorisations across the DeepHealth family (DeepHealth, Quantib, iCAD, Gleamer, See-Mode, Aidence). A provider roll-up that reads the studies, buys the scanners and builds its own AI — GE's customer, its AI competitor and a rival bidder for the private assets GE would want.
The enterprise-imaging and PACS incumbents GE bought Intelerad to compete with — and the cleanest refutation of the clearance-stock thesis, since both hold a verified ZERO entries on the FDA AI list while running software growth and software margins on the same reading pipeline.
The principal price competitor in patient monitoring and life support — the segment that just fell 13.5% organically into negative EBIT — and increasingly in ultrasound. Shenzhen-listed and outside this board's US-first mandate: a competitive fact, not an instrument.