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.
| Date |
|---|
| Type |
|---|
| What happened |
|---|
| Source |
|---|
| 2026-08-10 | editorial note | Margin figure revised: 68.9% → 71.4%Margin moved from 68.9% (deep-dive-2026-06-18.md) to 71.4% (deep-dive-2026-08-10.md). | dossier |
| 2026-08-10 | editorial note | Revenue figure revised: 10% → $32.612MRevenue moved from 10% (deep-dive-2026-06-18.md) to $32.612M (deep-dive-2026-08-10.md). | dossier |
| 2026-08-10 | editorial note | Verdict changed: 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 sale…Before (deep-dive-2026-06-18.md): 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. After (deep-dive-2026-08-10.md): The platform thesis got materially better — eleven Embedded partners, four of them BCI, and a 51.8% software mix — and the price got worse faster: the multiple roughly doubled to ~21x EV/sales in 53 days while operating cash burn rose 60% and record revenue was harvested out of deferred revenue rather than earned in cash. | dossier |
The verdict
The platform thesis got materially better — eleven Embedded partners, four of them BCI, and a 51.8% software mix — and the price got worse faster: the multiple roughly doubled to ~21x EV/sales in 53 days while operating cash burn rose 60% and record revenue was harvested out of deferred revenue rather than earned in cash.
Primary sources
SEC filings
Earnings calls
Source documents — open to read in full
Midjourney shipped a product, and the stock went vertical the day after the anchor dossier was written. On 2026-06-18 Midjourney unveiled Midjourney Medical and The Midjourney Scanner, a full-body tomographic imaging machine whose prototype carries 40 Butterfly Ultrasound-on-Chip modules per system, with future generations expected to use "substantially more". BFLY +55.9% in one session to close $8.90. The anchor dossier's price was $5.69 as of 2026-06-17 — the re-rate began roughly the hour it was filed.
Embedded went from a sample size of one to a stated ELEVEN partners. DeVivo on the Q2 call: "we have 11 partners", after adding two new agreements in Q2 — one in neurotechnology, one in women's-health wearables — and separately: "I think it's four neurotechnology BCI companies now in the Garden". On 2026-06-25 a second Embedded partner went public: Aleph Neuro, a brain-interface lab, claiming the highest-resolution 3D images of the human brain captured from outside the skull (ultrasound through the skull, with contrast). No financial terms disclosed for Aleph.
Q2-2026 was a record beat-and-raise. Revenue $32.612M, +39.5% YoY, vs consensus $29.22M / $29.9M — sources disagree on the consensus base, surfaced not averaged. Gross margin 71.4%, an all-time high (+7.7pts YoY). Adjusted EBITDA loss $1.4M vs $6.2M, −78%. FY2026 revenue guidance raised to $119-123M from $117-121M; FY adj-EBITDA loss tightened to $19-23M.
The cash-flow inflection unwound. H1-2026 operating cash used $(30.163)M vs $(18.844)M a year earlier — burn worsened 60.1% even as the net loss narrowed. Cash $150.489M → $124.659M. The anchor dossier's headline positive — "first positive operating cash flow in company history" — no longer describes the current run-rate.
The founder started selling, and it shows in the share register. Subsequent to 2026-06-30, 5.0M Class B shares automatically converted to Class A upon sale; Class B outstanding fell 26,426,937 → 21,426,937 between the 6/30 balance sheet and the 7/24 cover page. Rothberg's Mar-2026 Rule 10b5-1 planA schedule an insider sets in advance saying when their shares will be sold, so the sales cannot be timed on private news. A sale made OUTSIDE such a plan is a live decision, which is why it carries more information. plan is executing on schedule. No new plans were adopted, modified or terminated in Q2.
Litigation exposure tripled in one quarter. The Sezonov indemnification loss contingency went $3.0M (Q1) → $10.0M (Q2), matched by an insurance-recovery asset of $10.0M. Net P&L effect in Q2 was zero ($4.0M accrual, $4.0M recovery); the H1 net drag was $2.75M. Legal-related expense inside "Other" ran $3.537M in Q2 vs $1.499M (+136%).
The structural thesis stands, and one of its two legs got materially stronger. The anchor's core claim — a real semiconductor IP moat, a contested product moat, and the whole re-rate riding on whether Embedded becomes a platform — is intact. The Embedded leg is now much better evidenced: eleven partners, two shipped public demonstrations (Midjourney's scanner, Aleph's brain images), a named developer ecosystem (Butterfly Garden, with three partner tools due commercially by year-end), and a firewalled dedicated Embedded organization so it does not cannibalise core attention.
The anchor's bear leg also got stronger, in the numbers. Its sharpest line — "strip Midjourney and you have a ~$85M, slow-growing hardware company" — is now confirmed by the filing rather than inferred: Core business revenue grew just +2.2% YoY ($21.851M vs $21.387M) while units fulfilled grew +22.5% and ASP fell 12%. Volume is up five times faster than core revenue. That is price/mix erosion arriving exactly where the anchor predicted commoditization would show up.
What broke in the anchor: two claims are now wrong and should not be carried forward.
The anchor described two engines (Core + Embedded). Management now frames three engines of growth:
Two adjacent structures matter and were not in the anchor:
Customer concentration persists: one customer >10% of total revenue for both the three and six months ended 2026-06-30 (none a year ago); notably, no customer >10% of receivables at either 6/30/26 or 12/31/25 — a change from Q1-2026, when one customer was >10% of receivables. CEO/Chairman Joseph DeVivo; CFO John Doherty — unchanged.
Unchanged in substance, see the previous dossier. Two figures moved and are recorded here only: the non-cancellable single-vendor minimum inventory purchase commitment fell $3.3M → $2.0M, and the vendor-advance asset specific to that agreement is now $0.4M net of write-downs. Total vendor advances on the balance sheet are $7.025M ($2.157M current + $4.868M non-current). Fabless model, foundry still unnamed in filings — remains n/a. One new supply-side risk was voiced on the call: "pressure from tariffs in certain markets and AI-driven supply chain shortages".
The anchor's judgement was "a real IP moat, a contested product moat" and the open question was whether Embedded converts IP into platform economics. Three pieces of new evidence, all in the last 53 days:
Against that, the product moat visibly eroded this quarter: units +22.5%, ASP −12%, core revenue +2.2%. DeVivo's own counter-claim is the P5.1 chip: "If this holds true, with better imaging in a handheld than our competitors, both big and small, why would anyone buy anything other than a CMUT?" — an unverified forward claim about an unreleased product, not evidence. Competitive set (GE Vscan Air, Philips Lumify, Clarius, Siemens) carried from the anchor; no new share data sourced this pass — n/a.
One GAAP segment; three disaggregations, all `` and now written to our figures:
| Q2-2026 | Revenue | YoY | Share |
|---|---|---|---|
| By business line | |||
| Core business | $21.851M | +2.2% | 67.0% |
| Embedded | $10.761M | +439% | 33.0% |
| By type | |||
| Product (point-in-time) | $15.720M | −5.4% | 48.2% |
| Software & other services (over time) | $16.892M | +149.8% | 51.8% |
| By geography | |||
| United States | $27.591M | +57.3% | 84.6% |
| International | $5.021M | −14.1% | 15.4% |
Three things this table says that the anchor's could not:
Sequential shape of Embedded, the number that decides everything: Q4-2025 $6.8M → Q1-2026 $5.735M → Q2-2026 $10.761M. The anchor flagged the Q4→Q1 decline as evidence of lumpiness. Q2 nearly doubled sequentially. Management's own explanation is milestone timing, not run-rate: "We made a good amount of progress against some of the milestones in the second quarter, that resulted in the Embedded number being higher than, say, it was in the first quarter". Lumpiness is confirmed in both directions — and Q3 guidance implies it swings back down (Lens 5).
Top line. Revenue $32.612M, +39.5% YoY. Beat consensus — but the consensus base is disputed: $29.22M vs $29.9M. On either base it is a 9-12% beat. Non-GAAP adjusted EPS $(0.01) vs $(0.06) modelled.
Margins. Gross profit $23.288M, gross margin 71.4% vs 63.7% — an all-time high, +7.7pts, driven by licensing mix. Product COGS as % of revenue fell to 22.6% from 28.5%; software COGS to 6.0% from 7.8%.
Profitability. Loss from operations $(13.664)M (from $(16.100)M). Net loss $(12.910)M, EPS $(0.05). Adjusted EBITDA loss $1.4M vs $6.2M, −78%. Note the GAAP-to-adjusted gap: SBC alone was $7.038M in Q2, 21.6% of revenue. Adjusted EBITDA is close to breakeven only because it adds back a fifth of revenue in stock comp.
Opex — the underappreciated line. Total opex $36.952M, +19.2% YoY, growing faster than core revenue by an order of magnitude. R&D +26.8% ($10.542M), G&A +24.4% ($11.355M), S&M flat (−0.8%). Management is explicit that this is deliberate and continues: FY guidance "includes increased investment in key areas to support continued innovation and revenue growth in our core business and our emerging Embedded business".
Balance sheet — three flags.
Cash $124.659M, no financial debt. (stockanalysis.com shows "Total Debt $19.10M" — that reconciles exactly to capitalized operating leases, $2.804M current + $16.293M non-current = $19.097M. Not borrowings. Recorded so the two sources are not read as a contradiction.)
Cash flow — the most important delta in this dossier. H1-2026 net cash used in operating activities $(30.163)M vs $(18.844)M, +60.1%. Management's bridge: a $14.5M increase in working-capital cash usage, comprising $10.1M more cash used in deferred revenue, $8.8M more in prepaid/other assets (largely the $10.0M insurance-recovery asset), and $4.7M more in receivables, partly offset by $9.3M less in payables/accruals.
The mechanism, stated plainly: the company recognised $14.3M of Q2 revenue out of the deferred-revenue balance it was already holding (vs $5.6M in the prior-year Q2); for the half, $22.2M vs $11.2M. Record revenue and record margin were substantially harvested from the balance sheet — the Midjourney $15M upfront and prepaid annual fees converting into P&L revenue with no new cash arriving. That is why the income statement improved and the cash statement deteriorated in the same quarter. It is disclosed, it is legitimate GAAP, and it is not repeatable.
A reconciliation worth naming. MD&A says "During the three months ended June 30, 2026, we utilized $13.3 million of cash and cash equivalents for ongoing operations". That $13.3M is the total cash decline in the quarter ($137.954M → $124.659M), which is net of ~$4.0M of financing inflow from option exercises and the ESPP. The GAAP operating cash used in Q2 was $(16.269)M. Management's framing understates the operating burn by ~$3M. Worth a question.
Guidance — raised. FY2026 revenue $119-123M (22-26% growth), up from $117-121M; FY adj-EBITDA loss $19-23M, improved. Q3-2026 revenue $26-30M (+30% YoY at midpoint); Q3 adj-EBITDA loss $6-9M. Read the shape: Q3 midpoint $28M is a sequential DECLINE from Q2's $32.6M, and the Q3 adj-EBITDA loss widens from $1.4M to $6-9M. H1 delivered $59.142M; the FY midpoint implies H2 of $61.9M — essentially a flat second half against a first half management just called a record.
Market reaction — sources conflict, surfaced. Simply Wall St, dated 2026-07-31: "up 8.1%" after the raise. A separate Simply Wall St summary states the "stock fell 13% on the day to US$7.11". These cannot both describe 30 July. The reconcilable facts: the stock was ~$7 in the days around the print and is $9.89 on 2026-08-10, +12.0% on the day. No catalyst for the 2026-08-10 move was found in this sweep — recorded as unexplained rather than guessed at.
The anchor had no transcripts and read sentiment from `` summaries. This pass ingested the primary Q2-2026 call to transcripts/2026-q2.md.
Trajectory across the last four calls: turnaround (mid-2025) → vindication (Q4-2025: first positive op cash flow + the Midjourney deal, +27.2%) → execution (Q1-2026: margin, FDA gestational-age clearance) → expansion (Q2-2026). The Q2 tone is the most confident of the four, and it has shifted register again: from "we are becoming a platform" to "we are a platform, here is the roadmap through 2028."
New vocabulary that was not there before: "three engines of growth", "Butterfly Garden", "Apollo", "the Garden evolving from a developer ecosystem into a commercial AI platform", "firewalled a dedicated organization" for Embedded, "second bite of the apple" (partners going commercial), "brain-computer interfaces", "CMUT".
Phrases that have disappeared: the survival-era language is long gone; notably, the Q4-2025 "positive operating cash flow" framing is not repeated — consistent with the fact that H1-2026 operating cash flow is $(30.2)M.
Where the tone runs ahead of the evidence — three places:
Where the tone is honest: DeVivo said plainly of international, "our team had a down quarter"; Doherty volunteered the Middle East conflict, tariffs and "AI-driven supply chain shortages" as live pressures; and management conceded deals slipped out of Q2 into Q3. The self-criticism is real, which raises rather than lowers the credibility of the rest.
Net: sentiment up-shift continues and is partly justified by numbers (margin, mix, guidance raise) and partly not (cash flow, core growth, backlog). That is a change from the anchor, which judged the up-shift fully justified.
Butterfly still has no profits, so P/E and ROE remain n/m. The debate is still EV/Sales, and the multiple roughly doubled in 53 days.
| Anchor (2026-06-18) | Now (2026-08-10) | Change | |
|---|---|---|---|
| Share price | $5.69 | $9.89 | +73.8% |
| Market capitalisationThe share price multiplied by the number of shares. What the market says the equity is worth. | ~$1.49B | $2.63B | +76% |
| Enterprise valueWhat it would cost to buy the whole company: its market value plus its debt, minus the cash you would get with it. Often very different from market cap. | n/a | $2.52B | — |
| EV/Sales (trailing) | ~14.3x cited | 20.03x | +40% |
| EV/Sales on FY guide | ~10-11x on $117-121M | ~20.8x on $121M midpoint | ~2x |
| Short interestHow many shares have been borrowed and sold by people betting the price falls. | 12.6% | 14.51% of float / 31.25M sh; also cited 9.57% of shares outstanding / 25.04M sh | conflicting bases, both recorded |
| 52-week price change | +159.6% | +637.59% | — |
| 52-week range | n/a | $1.34 – $10.03 | at the high |
Peer context carried from the anchor and not re-sourced this pass: US medical-equipment industry ~2.7x sales, direct peer set ~4.4x. BFLY now trades at roughly 5-8x its peer group on sales, versus 2.5-4x at the anchor. GEHC / PHG multiples remain n/a. Rothberg-adjacent comps (HYPR, QSI) remain n/a.
Strip Embedded and the number is stark. Annualising Q2 core at $21.851M gives ~$87.4M of core revenue; EV/core-sales = 28.8x. The anchor's judgement — "on the hardware business alone it is ~3-4x too expensive" — is now closer to ~7-10x too expensive.
Analyst targets — all three sourced figures sit BELOW spot, and they disagree:
Honest statement: every sourced target the sweep produced is below the traded price, and the newest of them is nine days old. The multiple is no longer defensible on a "software/IP platform re-rate" argument alone — a ~21x EV/sales multiple on a company guiding to a $19-23M adjusted EBITDA loss and burning $30M of operating cash per half is priced for the Embedded pipeline to convert to chip-volume royalties, which it has not yet begun to do (Product revenue fell 5.4%).
| Date | Event | Move |
|---|---|---|
| 2026-06-18 | Midjourney unveils Midjourney Medical + The Midjourney Scanner; 40 Butterfly modules/system, more in future generations | +55.9% to close $8.90 |
| 2026-06-18 | 2026 Annual Meeting: all 7 directors re-elected (>99% for), Deloitte ratified, say-on-pay 98.6% for | — |
| 2026-06-25 | Embedded partner Aleph Neuro publishes highest-resolution transcranial 3D brain images | positive, magnitude not sourced |
| ~2026-07-06/09 | Brazil regulatory clearance for iQ+ and iQ3 — largest LatAm medical-device market | positive, magnitude not sourced |
| ~2026-07-24 | 5.0M Class B convert to Class A upon sale (Rothberg 10b5-1) | — |
| 2026-07-30 | Q2 beat + FY guidance raise | +8.1% — *conflicting report of −13% to $7.11 also surfaced * |
| 2026-08-10 | — | +12.0% to $9.89; no catalyst identified in this sweep |
What the tape now reveals. The anchor's finding was that the market pays for (1) Embedded/licensing news and (2) cash-flow inflection, not for probe units. The last 53 days confirm the first half emphatically and falsify the second: the single largest move (+55.9%) came from a partner's product demo, while the quarter that actually reversed the cash-flow inflection was rewarded with +8.1%. Butterfly is now traded almost purely as a call option on other companies' Ultrasound-on-Chip products — Midjourney's scanner, Aleph's brain interface. The dominant price driver is Embedded-partner news flow, and BFLY holders are underwriting execution risk at companies whose results they cannot see.
Two consequences worth holding: FDA status. Midjourney's scanner is not FDA-cleared, plans an initial non-diagnostic use case (body-composition mapping), and intends to open its first location inside a San Francisco spa in 2027. Benzinga's own framing — "Climbs On Midjourney Hype Despite FDA, Reimbursement Questions" — is the sceptical read, and it is the correct one to keep on the dashboard.
DeVivo and Doherty in place; no management changes this quarter. Board unchanged and endorsed: all seven directors re-elected at the 2026-06-18 annual meeting with >99% support, Deloitte ratified, say-on-pay approved 614.1M for / 8.86M against (98.6%). There is no shareholder revolt to report — which is itself notable given a dual-class structure and an executing founder sale.
Founder selling moved from filing to fact. The anchor recorded Rothberg's 10b5-1 plan (up to 2.8M Class A + 5.0M Class B over Jul-2026→Jul-2027) as a watch item. It is now executing: 5.0M Class B converted to Class A upon sale after 6/30. Voting arithmetic:
| Class A | Class B (20 votes) | Total votes | Class B share | |
|---|---|---|---|---|
| 2026-06-30 | 237,995,479 | 26,426,937 → 528.5M votes | 766.5M | 68.9% |
| 2026-07-24 | 244,858,246 | 21,426,937 → 428.5M votes | 673.4M | 63.6% |
Control remains comfortably with Class B, but it erodes roughly 5 points per 5M Class B shares sold. The plan runs to Jul-2027. No new 10b5-1 arrangements were adopted, modified or terminated in Q2 — this is the existing plan running, not fresh selling appetite. The neutral read: programmatic diversification by a founder whose stock is up 638% in a year. The uncharitable read: the anchor's pre-mortem line — "the founder's 10b5-1 selling, in hindsight, marked the sentiment top" — is now a live test rather than a hypothetical.
Capital allocation. Still no debt, no buyback ("We did not repurchase any of our equity securities during the three months ended June 30, 2026"), no raise. Financing inflow in H1 was $6.257M, entirely employee option exercises and ESPP. Discipline holds. One overlooked item: all warrants expired unexercised on 2026-02-12 at an $11.50 strike, removing 20.65M shares of potential DilutionIssuing new shares, so each existing share owns a smaller slice of the same company.. At today's $9.89 those would be near the money — the company forfeited a potential ~$237M of exercise proceeds by roughly four months of share-price timing. Not a management error; worth knowing when assessing the "7-year runway" story, because that optional capital is gone.
Cleared since the anchor:
New or escalated — five:
Revenue is being recognised out of the balance sheet. $14.3M of Q2 revenue (44% of the quarter) came from the opening deferred-revenue balance, vs $5.6M a year ago; $22.2M vs $11.2M for the half. Deferred revenue fell $10.6M and Remaining performance obligationsRevenue a company has already signed contracts for but has not yet delivered or booked. fell $13.8M in six months. Record revenue coincided with backlog shrinking — the definition of harvesting rather than building. Not improper; decisive for how the print should be read.
Operating cash burn up 60% against a narrowing loss. $(30.163)M vs $(18.844)M H1. The anchor praised the opposite pattern ("cash is the weaker number here… healthy sign") when Cash burnHow much more cash goes out than comes in, per period. The clock on a company with no profits. slightly exceeded net loss. Now cash burn is 1.18x net loss and rising. The Q4-2025 "first positive operating cash flow" milestone has not held.
Litigation accrual tripled, offset entirely by an estimated insurance receivable. Sezonov indemnification liability $3.0M → $10.0M; insurance recovery asset $6.0M → $10.0M, described as "deemed probable and estimable" by management. The offset is a management estimate, not a signed settlement. If the insurer disputes coverage, a fully-reserved non-event becomes a $10M cash hit. This is the single most estimate-dependent item on the balance sheet and it grew 233% in one quarter. The Rose class action carries a separate $0.3M accrual; Koenig and Bhavsar remain unestimable.
The fastest-growing revenue line is still the most management-estimated one. Embedded is recognised "over time using input methods to determine a measure of progress" against milestones, and management stated outright that Q2's step-up was milestone-progress timing. 33% of revenue is now measured by management's own assessment of its own progress. Discretion has grown with the line.
Receivables stretching + a distributor credit loss. AR +$7.8M; DSO ~86 → ~96 days; allowance for credit losses $1.389M → $1.785M with a $0.702M Q2 provision (vs $0.121M a year ago), including "a $0.4 million provision for credit losses… related to a prior sale to one international distributor". Small in dollars; the direction is the signal, in the same segment that shrank 14%.
Dilution. SBC $7.038M in Q2 (21.6% of revenue) and $12.580M in H1. Weighted shares 262.1M vs 248.4M, +5.5% YoY. Class A outstanding 227.3M → 238.0M in six months. RSUs outstanding 20.2M (8.0M granted in H1), options down to 2.2M. Real, persistent, ~4-5%/yr dilution — and the entire adjusted-EBITDA-near-breakeven story is built on adding it back.
Regulatory findings (required sub-section).
total_sec_findings: 0.Model integrity first — the workbook cannot be cited. our model ran and wrote model.xlsx, but our model returns "No computed values. The workbook has 51 formulas and no cached results." Every valuation output is blank. It also flagged three seed gaps: only 2 quarters of financials (this pass wrote 2026-Q1 and 2026-Q2; the CSV was empty before), so LTM figures are partial; fewer than 8 quarters, so revenue growth was seeded at a placeholder 8%; and the opening balance sheet is only partly sourced (8/10 lines) — missing other-current-assets and other-current-liabilities. No number below comes from the model. Everything is `` with arithmetic shown, off filing and guidance inputs.
Anchors. FY2026 revenue guide $119-123M, FY adj-EBITDA loss $19-23M. H1-2026 actual: revenue $59.142M, gross profit $41.574M (70.3%), opex $69.110M, operating loss $(27.536)M, net loss $(25.587)M, EPS $(0.10). Q3 guide $26-30M revenue, $6-9M adj-EBITDA loss.
FY2026. H2 implied revenue $61.9M at the guide midpoint — flat on H1. Embedded mix falls (Q3 midpoint is a sequential decline), so H2 GM ~68-69%. Opex steps up on management's own stated investment: H2 ~$74M vs H1 $69.1M.
This is worse than the anchor's FY2026 base of $(0.17), and the reason is not revenue — the guide went up. It is opex: management is spending the gross-margin windfall on Apollo, P5.1, iQ Station and the Embedded organisation rather than dropping it to the bottom line. That is a defensible strategic choice and a real deferral of profitability.
FY2027. Core ~$88M growing 5-8% → ~$93M. Embedded FY2026 ≈ $32M; FY2027 $35-50M on the Apollo development programme, the $10M/yr Midjourney annual fee, up to $9M of milestones, and the first commercial conversions among the other ten partners.
FY2028. The first year Apollo could produce chip volume (delivery to Midjourney end-2027/early-2028) and iQ Station addresses the cart market.
GAAP breakeven moves out. The anchor put it at FY2028-2029; on these inputs the base case reaches breakeven beyond FY2029, with only the bull case getting close in FY2028. Adjusted-EBITDA breakeven, by contrast, is nearly here — Q2's adjusted loss was $1.4M — but FY guidance deliberately widens it again ($19-23M), so even that is a 2027-28 event on management's own numbers.
Runway — the anchor's "7+ years" is wrong and must be retired. Cash $124.659M, no financial debt.
Brier forecast: NOT logged (unattended run; our model create is out of scope per the dispatch rules). The natural one to log on an attended pass: "BFLY FY2026 revenue ≥ $121M (guide midpoint), p ≈ 0.70, resolves 2026-12-31" — H1 delivered $59.1M and the implied H2 is flat, which is an unusually low bar for a company that has beaten twice running.
Bull case (materially stronger than at the anchor). The platform is no longer a hypothesis with one customer — it is eleven partners, four of them in brain-computer interfaces, with two shipped public demonstrations in eight days: Midjourney's whole-body tomographic scanner (40 Butterfly modules today, "substantially more" in future generations) and Aleph Neuro's transcranial 3D brain images. Software crossed 51.8% of revenue and gross margin hit an all-time 71.4%; adjusted EBITDA loss shrank 78% to $1.4M. Management raised FY guidance twice-over (revenue up, EBITDA loss down), has no debt, and owns a semiconductor roadmap — Apollo, feeding high-speed ultrasound data directly into GPUs — that positions Ultrasound-on-Chip as an AI data-acquisition layer, not a probe. The Garden ecosystem converts partners into recurring software, three tools go commercial by year-end, Compass AI's pipeline is 5x larger than a year ago, the company sits in 80% of US medical schools, Brazil just opened LatAm, and FedRAMP opens the VA. Every one of the eleven partners is, in DeVivo's phrase, a "second bite of the apple" when it commercialises — chip sales and revenue share on top of licence fees.
Bear case (also stronger — the mechanisms are now visible in the filing).
Pre-mortem (18 months out — thesis broke). It is now mid-2028. Apollo shipped to Midjourney on time, and that is exactly the problem: the development-revenue programme that produced $10.8M quarters ended, and the chip-purchase/revenue-share economics that were supposed to replace it never scaled — because Midjourney's scanner is still not FDA-cleared, opened in a single San Francisco spa doing body-composition scans, and no payer reimburses it. Of the eleven Embedded partners, seven quietly lapsed and the neurotech ones remain research programmes. Core revenue is $95M growing 3% with ASPs down another 15% against a subscription-free GE Vscan. Cash is under $50M and the company raises at a fraction of the 2026 high. In hindsight the tell was Q2-2026 itself: record revenue, record margin, and 60% more cash burned — the balance sheet paying for the income statement, in plain sight, while the founder sold into it.
Are the multiples too high? On the current revenue base, unambiguously yes — ~21x sales for a business guiding to a $19-23M adjusted EBITDA loss, whose largest growth line is milestone-estimated and whose core grows 2%. On a 2029 platform where a meaningful share of eleven partners ship products carrying 40+ Butterfly chips each, no. The multiple is a bet not on Butterfly's execution but on its customers' execution — Midjourney's FDA path and Aleph's science.
Contrarian view (what the market refuses to see). Consensus now treats the eleven-partner count as proof the platform compounds. The contrarian read: partner count is an input metric, and Butterfly is being valued as though it were revenue. Ten of the eleven have no disclosed economics. The one that does generates development fees that are, mechanically, the amortisation of a $15M cheque and a $10M annual fee — and every quarter of "record Embedded revenue" spends down the deferred balance that produced it. The market re-rated 74% in 53 days on two demos: a scanner in a spa and a research lab's brain images. Neither has FDA clearance, a payer, or a shipped unit.
Dismantling the bull case with this quarter's own filing.
The bull's headline is 39% revenue growth. Strip the licence amortisation and the operating business grew 2.2%. Core was $21.851M against $21.387M. Everything else — the record revenue, the record 71.4% margin, the 78% adjusted-EBITDA improvement, the guidance raise — traces to $10.761M of Embedded revenue that is, by the company's own disclosure, milestone-progress recognition against a contract signed in November 2025. The company recognised $14.3M of the quarter out of deferred revenue it was already sitting on, drew total deferred revenue down to $25.7M, watched RPO fall to $85.8M from $99.6M, and burned 60% more cash than the year before. That is not an inflection. That is a company converting a prepayment into an income statement while telling you the margin is structural.
The unit economics are deteriorating underneath. 22.5% more probes shipped, 12% lower ASP, 2.2% more core revenue. Management attributes ASP decline to iQ+ mix and "strategic price subsidies for the VCOM agreement" — a $10M-over-four-years one-to-one medical-student probe programme. They are subsidising probes into medical schools to buy an installed base while GE Vscan Air ships with no subscription at all. International revenue fell 14% and came with a bad-debt provision against a distributor.
The most estimate-dependent line on the balance sheet grew 233% in a quarter. The Sezonov indemnification accrual went $3.0M → $10.0M, offset dollar-for-dollar by an insurance receivable management "deemed probable and estimable." Q2's P&L effect was exactly zero because the two moved in lockstep — a symmetry that is convenient, disclosed, and entirely within management's judgement. Legal spend inside "Other" more than doubled. If coverage is contested, $10M lands in cash.
Worst governance marks. A dual-class structure giving Class B twenty votes each; a founder executing a plan to sell up to 7.8M shares into a 638% one-year run, with 5.0M Class B already converted-on-sale by 24 July; and 21.6%-of-revenue stock comp that the entire "near-breakeven adjusted EBITDA" narrative depends on adding back.
Assumptions that must hold for today's $9.89 / $2.63B. (a) Embedded converts from development fees to chip volume + revenue share — the leg currently contributing negative growth, since prior-year chip sales did not recur; (b) Midjourney clears FDA and finds reimbursement for a scanner whose first venue is a spa; (c) several of the ten unquantified partners produce disclosed economics; (d) core stops decaying on price; (e) the market keeps paying ~21x sales through a guided flat second half.
If growth disappoints 20-30% — FY2027 landing at $110-120M rather than $140M as the Midjourney development programme completes without a successor — the multiple compresses toward even a generous 8-10x software-platform level: $0.9-1.2B enterprise value, roughly $3.50-4.75 a share, 52-65% below spot. At the peer-set 4.4x it is under $2.50.
The single scenario that permanently impairs: Apollo ships, the Midjourney development revenue stops, no partner reaches commercial chip volume, and core POCUS is a no-subscription commodity by 2028 — leaving a ~$100M business burning $50M a year with brilliant, stranded IP and a diluted share count. Plausibility: moderate, and higher than at the anchor — not because the technology got worse, but because the price now requires the best branch of an eleven-way option to pay off, and the company just told you the second half will be flat.
Unchanged in structure, see the previous dossier (fifteen questions, Embedded-pipeline first). Five are now answered or superseded and five should replace them:
Answered: Q1 (pipeline) — eleven partners, two added in Q2. Q5 (E&O repeat) — no repeat, charges insignificant. Q10 (customer concentration) — still one >10% of revenue, but no longer >10% of receivables. Q14 (FDA/AI roadmap) — Brazil clearance, VA provisional, FedRAMP expected Q3.
New questions this quarter raises:
Every dossier we have written on Butterfly Network, newest first.
The platform thesis got materially better
A genuine semiconductor moat wrapped around a sub-scale, still-unprofitable medtech
| Industry | Neurotech & BCI |
| Size | Public Company |
Where Butterfly Network sits against the other names we cover on this beat. Each line is that company’s verdict, not a summary of it.
World's first commercial invasive BCI and the first BCI company to file a public listing
World-first non-invasive SCI neuromod with real FDA labels and a science moat
The open-source pioneer that defined research-grade DIY neurotech
The first BCI company with a *shipping product and real efficacy*
China's only fully-invasive (intracortical) BCI in the clinic and the best-capitalized BCI name in the country