Neurotech & BCI
A genuine semiconductor moat wrapped around a sub-scale, still-unprofitable medtech — the Midjourney/Embedded licensing pivot is real and re-rates the story, but at ~10x forward sales with episodic licensing revenue and a founder selling, the price already imputes the platform it has not yet proven.
Research
The verdict
A genuine semiconductor moat wrapped around a sub-scale, still-unprofitable medtech — the Midjourney/Embedded licensing pivot is real and re-rates the story, but at ~10x forward sales with episodic licensing revenue and a founder selling, the price already imputes the platform it has not yet proven.
Butterfly Network is a digital-health company that turned medical ultrasound from a piezoelectric-crystal cart into a semiconductor. Its core invention — the Ultrasound-on-Chip™ — replaces the hand-cut crystal arrays inside a conventional probe with a CMOS chip, so a single handheld probe (the Butterfly iQ, now iQ3) can image the whole body and plug into a phone or tablet. The pitch is access: a ~$2,000-class pocket device versus six-figure cart systems, paired with a cloud SaaS layer and AI guidance.
The business has two engines, and the 10-Q now discloses revenue by both:
Revenue model & key terms: hardware is one-and-done at point of sale; SaaS/warranty is ratable; Embedded is a mix of upfront license fees, annual fees, milestone payments, chip-purchase revenue and revenue-share — recognized over time against project milestones (input method). Customer concentration is now flagged for the first time: in Q1-2026 one customer accounted for >10% of total revenue and >10% of receivables, versus none a year ago — almost certainly the Midjourney Embedded contract (see Lens 4). Subsidiaries operate in the US, Australia, Germany, Netherlands, Taiwan and the UK. CEO/Chairman: Joseph DeVivo; CFO: John Doherty. The "bci" beat tag in the index is a loose fit — Butterfly is medical-imaging semiconductors, not brain-computer interface; treat it as a frontier-hardware/edge-sensing name.
Upstream → Butterfly → end customer, named where disclosed:
n/a in filings.Chokepoint summary: a fabless semiconductor model means Butterfly is exposed at two single points — foundry capacity and the one contract-manufacturing vendor it has a binding minimum with. Neither is diversified in the disclosures.
The real moat is the chip. Butterfly is the only player that delivers whole-body handheld imaging from a single semiconductor transducer rather than separate piezoelectric probes — the original iQ won the broadest single-transducer FDA 510(k) clearance (13 clinical applications) ever granted. That is a genuine, IP-protected process advantage that scales with semiconductor economics (Moore's-law cost/performance) rather than crystal-cutting craft. The Embedded program turns that moat into a platform — other companies licensing the chip to build new devices — which, if it compounds, is the most durable version of the business (royalty/IP model, asset-light, ~zero marginal cost).
But on the product axis the moat is thinner than the chip story implies. In handheld point-of-care ultrasound (POCUS) Butterfly is not the share leader:
Bargaining power: weak over customers in core hardware (commoditizing handhelds, entrenched GE/Philips hospital relationships, subscription fatigue), stronger in Embedded (Butterfly owns unique IP a licensee cannot easily replicate). Switching costs exist only where a health system has standardized on Butterfly's cloud/AI workflow (Compass AI, fleet management) — that's the part of the moat management is trying to deepen. Net: a real IP moat, a contested product moat. The investment question is whether Embedded converts the IP edge into platform economics before the hardware commoditizes.
Butterfly reports one GAAP segment (the CODM reviews consolidated net loss), so there is no segment EBITDA. But it disaggregates revenue three ways, all ``:
By type (FY2025): Product $63.4M, Software & other services $34.2M; total $97.6M (+19.0% YoY from $82.1M; FY2023 $65.9M). Three-year revenue CAGR ≈ 22%.
By geography (FY2025): US $77.4M, International $20.2M. The US is ~79% of revenue — a domestically concentrated franchise; international is the smaller, distributor-led growth edge.
By business line — the key trend (Q1-2026): Core $20.8M vs Embedded $5.7M. A year earlier Embedded was $2.3M, so Embedded ~tripled YoY and drove ~91% of total revenue growth ($4.8M of the $5.3M increase came from Software & services, "primarily driven by Embedded partnerships"). Software mix hit 44.8% of revenue, +11.5pts YoY — the highest software mix in the company's history and the entire bull thesis in one number.
Caution on the trend's quality: Embedded is lumpy/episodic. The Midjourney deal booked a $15M one-time fee in FY2025 (recognized over time) plus a $10M/yr annual fee; Embedded contributed $6.8M in Q4-2025 but only $5.7M in Q1-2026 — i.e. the segment can shrink quarter-on-quarter. Units fulfilled (the core-hardware health metric) grew only +9.1% in FY2025 and +4.9% in Q1-2026 — the hardware base is growing slowly; the re-rate is coming from licensing, not from selling more probes.
Top line: Total revenue $26.5M, +25.0% YoY ($21.2M). Beat consensus (~$25.7M-$26.1M). Product +3.5% to $14.7M; Software & services +68.2% to $11.9M — the mix shift is the story. Margins: Gross profit $18.3M, gross margin 68.9% vs 63.0% a year ago — a 590bp expansion, driven by the higher-margin software/Embedded mix and the absence of the prior inventory drag. Profitability: Loss from operations $(13.9)M (improved from $(18.5)M); net loss $(12.7)M, EPS $(0.05); adjusted EBITDA loss ~$6.1M, −32% YoY. Net loss would have been ~$10M better but for a one-off: $3.3M of new loss-contingency accruals (partly offset by a $6.0M insurance-recovery asset) netted to a $2.75M drag on the quarter. Balance sheet: Cash $138.0M (down from $150.5M at YE2025); no debt; inventory drawn down to $59.3M from $61.4M; deferred revenue $32.3M; remaining performance obligations (RPO/backlog) $91.4M, 55% expected within 12 months. Operating cash burn was $(13.9)M in the quarter — note this includes annual bonus payments, so it overstates the run-rate burn. Guidance: FY2026 revenue $117M-$121M (~20-24% growth). Market reaction: Despite the beat, the stock slipped ~1.8% premarket — a "good-but-priced-in" reaction consistent with a name up ~160% in 12 months. Vs its own history: The quarter is a clean recovery print after a brutal FY2025. Q3-2025 had a negative gross profit of $(3.8)M and a $(34.0)M net loss because of a $17.4M excess-&-obsolete (E&O) inventory write-down for components made obsolete by product-portfolio change. Q1-2026 shows that was a discrete event, not a margin-structure break.
No transcripts on the shelf (transcripts/ empty); sourced ``.
Management's narrative has shifted decisively from "survival/turnaround" to "platform" across the last ~4 calls:
Provenance-critical lens. Butterfly has no profits, so P/E and ROE are not meaningful (net margin −79%, ROE negative). The live debate is EV/Sales, and the sources disagree on the exact multiple — surfaced rather than averaged:
| Company | Ticker | Mkt cap | EV/Sales (fwd) | P/E | Div yield | 5-yr avg ROE |
|---|---|---|---|---|---|---|
| Butterfly Network | BFLY | ~$1.49B | ~10-11x fwd on $117-121M guide; some cite 14.3x trailing | n/m (loss) | 0% | negative |
| US Medical Equipment industry | — | — | ~2.7x | — | — | — |
| Direct peer set (cited) | — | — | ~4.4x ("fair ratio ~3.4x") | — | — | — |
| GE HealthCare | GEHC | n/a | n/a | n/a | — | — |
| Philips | PHG | n/a | n/a | n/a | — | — |
Read: BFLY trades at a 2.5-4x premium to its medical-device peer group on sales, justified (by bulls) as a software/IP-platform multiple rather than a hardware multiple. Listed POCUS rivals (GE, Philips, Siemens) are giant diversified caps for whom handheld ultrasound is a rounding error, so a clean pure-play comp does not exist — the closest analog is other Rothberg-incubated, pre-profit, chip-platform names (Hyperfine HYPR, Quantum-Si QSI), whose multiples I did not source here (n/a). The honest statement: the multiple only makes sense if Embedded becomes a recurring royalty stream; on the hardware business alone it is ~3-4x too expensive.
Mostly ``; pattern matters more than each tick.
CEO — Joseph DeVivo (since Apr 2023; also President & Chairman). A career medtech turnaround/exit operator: 35 years in the industry; previously CEO of InTouch Health (grew revenue ~30% CAGR over four years, sold to Teladoc for $1.1B), then President of Hospitals & Health Systems at Teladoc; Executive Chairman of Caption Health (AI-guided ultrasound, sold to GE HealthCare). By his own account he has integrated 8 acquisitions, run 4 turnarounds and delivered 5 exits. He inherited Butterfly with the stock under $1 and ~12 months of cash — and has since reached first positive operating cash flow and a +160%/12-month stock. This is a strong, on-point operator-CEO with a quantified track record, and his Caption Health/GE background is precisely why the AI + ultrasound + licensing strategy is credible.
Forensic lens — every figure labeled.
Regulatory findings (required sub-section).
Built bottom-up from the FY2026 guide ($117-121M) and the Q1-2026 run-rate. All outputs ``; inputs labeled.
EPS (basic, ~258M+ shares, rising ~3-4%/yr on SBC dilution):
--watchlist rules, not logged (no forecast.ts create in the breadth loop). A natural one to log on a future pass: "BFLY FY2026 revenue ≥ $119M, p≈0.60, resolves 2026-12-31."Bull case. A semiconductor IP moat is being converted into a platform. The Midjourney deal proves third parties will pay real money ($74M/5yr) to build on Ultrasound-on-Chip — and Midjourney is the first, not the last. Embedded is asset-light, ~zero-marginal-cost, high-margin licensing that re-rates the whole company from a hardware multiple to a software/IP multiple. Underneath it, the core franchise just hit 68.9% gross margin and first positive operating cash flow, AI tools (Compass, Gestational Age) deepen the SaaS moat and win 7-figure enterprise contracts, and a proven exit-CEO (DeVivo, ex-Caption Health→GE) is running a no-debt, 7-year-runway balance sheet. Secular tailwind: ultrasound is the cheapest, most-portable imaging modality and AI is making it usable by non-experts — a genuinely large TAM. Earnings surprise lever: each new Embedded deal is a step-function to revenue and sentiment. Bear case (permanent-impairment risks). (1) Embedded is one customer and episodic — strip Midjourney and you have a ~$85M, slow-growing (units +5-9%), still-loss-making hardware company commoditizing against GE and Philips; the segment shrank QoQ (Q4 $6.8M→Q1 $5.7M). (2) The hardware moat is contested — GE Vscan and Clarius undercut Butterfly's subscription model, and incumbents own the hospital relationships. (3) Valuation leaves no room — ~10-11x forward sales (vs ~2.7x industry) with no consensus path to profit for 3 years and 12.6% short interest; any Embedded air-pocket or guidance trim re-rates hard. Pre-mortem (18 months out, thesis broke): No second Embedded deal materialized; Midjourney revenue proved front-loaded and decelerated; core units stayed mid-single-digit; the market re-applied a ~4x hardware multiple → stock back toward $2-3 (the GuruFocus GF-Value/Alpha-Spread "fair value" zone of ~$2.2). The founder's 10b5-1 selling, in hindsight, marked the sentiment top. Are multiples too high? On the hardware business alone, yes (3-4x too rich). On a credible platform, no. The multiple is a bet on Embedded #2-#5. Contrarian view (what the market refuses to see): Bulls treat the Midjourney deal as a template and extrapolate a royalty machine; the contrarian read is that Ultrasound-on-Chip is a brilliant but narrow IP whose licensable, non-competitive use-cases are few — one marquee AI-imaging partner does not prove a pipeline, and the most valuable use-cases compete with Butterfly's own products (so it won't license them). The bear's edge is that the re-rate already happened (+160%) on a single contract.
Dismantling the bull case. The structural break is simple: Embedded is the entire re-rate, and Embedded is one lumpy contract. Revenue is concentrated — one customer >10% of total — and that customer's contribution already fell sequentially ($6.8M→$5.7M). The "platform" is, today, a sample size of one. Strip it and the durable business is a sub-$90M hardware franchise growing units 5-9%, with negative ROIC, a −79% net margin, and a 1/10 profitability rank, facing GE HealthCare and Philips — both of whom give away the functionality Butterfly charges a subscription for, and one of whom (GE) already bought DeVivo's last AI-ultrasound company (Caption Health) and can fold that capability into Vscan. The most dangerous competitor bulls underestimate is GE HealthCare: scale, hospital lock-in, no-subscription model, and the Caption AI assets. Worst capital-allocation / governance marks: a $17.4M inventory write-down in FY2025 (built obsolete stock), a dual-class structure that disenfranchises public holders, and a founder filing to sell ~7.8M shares into the rally. Revenue quality is the soft spot: the fastest-growing line is management-estimated, milestone-input-method licensing revenue — maximum discretion, minimum cash-certainty. Assumptions that must hold for today's ~$5.70 / ~$1.5B: (a) ≥1-2 more Midjourney-scale Embedded deals within ~18 months; (b) core gross margin stays ~68%+; (c) no inventory-write-down repeat on the next product cycle; (d) the market keeps paying a software multiple on a still-mostly-hardware revenue base. If growth disappoints 20-30% (e.g. FY2026 lands ~$95-105M with no new Embedded deal), the multiple compresses toward peers and the stock has ~50%+ downside to the $2-3 "fair value" cluster. Single scenario that permanently impairs: GE/Philips drive handheld POCUS to a no-subscription commodity and no Embedded pipeline emerges — Butterfly becomes a perpetual cash-burning niche probe-maker with a stranded chip IP. Plausibility: moderate — the IP moat is real, the cash runway is long, so "impairment" is more likely a de-rate than a zero.
The safest, most surgically scalable path into the brain and the only BCI Apple made a native input — but it bet the company on "good enough" 16-electrode bandwidth, and its own third-gen "whole-brain" pivot is a confession that the moat it is famous for may be the ceiling it has to escape.
The only FDA-cleared, commercially-shipping cortical BCI — but it is selling a 30-day surgical-monitoring tool, not the chronic implant the $500M valuation is priced on; Medtronic is the real tell, IP overhang from Rapoport's Neuralink past is the real risk.
The highest-bandwidth intracortical BCI just put its wireless device in a human brain (17 Jun 2026) — but it is two patients into an EFS, ~3-5 years from revenue, and out-capitalized ~3-4:1 by Neuralink, so the bet is "bandwidth wins the speech-prosthesis category" against a far better-funded field.