Insitro

Phase A — Understand the business

Lens 1 · Company Overview

Insitro is a machine-learning-native drug discovery and development company founded in 2018 by Daphne Koller (Stanford CS professor of 18 years, Coursera co-founder, former Chief Computing Officer of Alphabet's Calico). The thesis is unusually clean: most drug failures are biology failures — we pick the wrong target — and the way to fix that is not better chemistry but better data plus better models of human disease. Insitro generates its own large, purpose-built, multi-modal biological datasets (high-content cell imaging, induced pluripotent stem-cell / iPSC disease models, genomics, histopathology, molecular measurements) specifically to train ML models, then uses those models to de-convolute disease into patient subsets and nominate targets.

What it actually sells today. Insitro has no approved product and no clinical-stage proprietary asset of its own (more in Lens 5). Its revenue is pharma collaboration / platform-licensing income — upfronts, target-nomination milestones, and option fees from large pharma partners who pay to point Insitro's platform at their targets. Reported cumulative collaboration revenue from BMS, Lilly and Gilead is ~$150M. A third-party tracker pegs a ~$69M ARR estimate as of June 2024.

Platforms (the product).

• Virtual Human™ — the integrated data-generation + ML system that maps genotype→cellular phenotype→disease and nominates targets. This is the platform BMS's ALS targets came out of.
• ChemML — small-molecule design stack: ML + physics-based in-silico screening + affinity ML models trained on proprietary DNA-encoded libraries + an active-learning medicinal-chemistry engine.
• POSH (Pooled Optical Screening in Human cells) / CellPaint-POSH — pooled CRISPR screening + high-content imaging + ML to map gene function at scale without predefined biomarkers; validated in Nature Communications, Dec 2025. This is the credibility-with-the-science-community lens (Lens C).

Customers / partners (these are the revenue): Gilead (NASH, 2019), Bristol Myers Squibb (neuro/ALS, expanded through 2026), Eli Lilly (metabolic + ADMET/TuneLab, 2024-25). See Lens 5 for terms. Suppliers / inputs: compute (GPU), automated "robo-lab" wet-lab hardware, iPSC lines, reagents, DNA-encoded chemical libraries, and increasingly partner data (the Lilly TuneLab deal is literally Insitro renting 40 years of Lilly's chemistry data). Competitors: Recursion (post-Exscientia), Schrödinger, Isomorphic Labs, Xaira, Insilico, Eikon, Generate:Biomedicines, Genesis, Iambic, Chai (see Lens 7).

Lens 2 · Supply Chain (CDMO / manufacturing axis — +clinical swap)

Map the value chain — named nodes:

Upstream inputs →

• Compute: GPU capacity (vendor not publicly disclosed; the sector standard is NVIDIA silicon, often via a hyperscaler). n/a — specific provider not disclosed.
• Wet-lab automation ("robo-lab"): high-throughput automated imaging + cell-handling hardware in-house in South San Francisco.
• Biological raw material: iPSC lines, CRISPR reagents, Cell Painting dyes, and DNA-encoded libraries (DELs) feeding ChemML's affinity models.
• Partner data as an input: Lilly's TuneLab preclinical/ADMET dataset (40 years of Lilly chemistry), accessed via a federated-learning architecture hosted by a third party so data stays siloed.

→ Insitro (the platform) → generates data, trains models, nominates targets / designs molecules.

→ Downstream / end customer. Here is the structural point: Insitro's "customer" is large pharma, not the patient. It hands targets/candidates to:

• Gilead — takes nominated NASH targets through chemistry, development, commercialization.
• Bristol Myers Squibb — nominates ALS targets (ALS-1/-2/-3) discovered on Virtual Human; BMS runs development.
• Eli Lilly — metabolic siRNA/antibody programs (Insitro retains global rights, Lilly gets milestones+royalties on some); plus Lilly is the channel for ChemML ADMET models to reach other biotechs via TuneLab.
• For its own wholly-owned pipeline, the downstream chain would require a CDMO for GMP manufacture of siRNA/antibody/small-molecule clinical supply — not disclosed; no in-house GMP manufacturing. n/a — CDMO partners not disclosed.

Chokepoints / single-source dependencies:

1. Revenue concentration on ~3 pharma partners. Effectively all income is BMS + Lilly + Gilead. Lose or stall any one and the P&L moves materially (Lens 13).
2. Federated dependence on Lilly's data for the headline ChemML/ADMET differentiation — Insitro doesn't own that data; Lilly can change terms.
3. Compute + the founder. Koller is the platform's gravitational center (Lens 9). Single-person key-man risk is real for a founder-defined techbio.

Lens 3 · Competitive Advantages (moats — IP/data/platform axis)

The moat thesis (bull framing): Insitro's durable advantage is proprietary, purpose-built, multi-modal data generated by an automated lab, paired with models trained on it — a data-generation flywheel rather than a model that anyone can re-train on public data. Recursion and Insitro are the two names consistently placed in the "biology-data / phenomics lane," distinct from the structure-foundation-model lane (Isomorphic, Chai) and the generative-chem lane (Iambic, Genesis). Three candidate moats:

1. Data + lab capital (scale / process moat). You cannot Google Insitro's iPSC-disease-model imaging corpus; it costs years and capital to reproduce. POSH/CellPaint-POSH (Nature Comms, Dec 2025) is evidence the data → biology pipeline produces novel, publishable signal.
2. Pharma validation as a moat-by-proxy (bargaining power). Three of the most discerning buyers in biology — Gilead, BMS, Lilly — have repeatedly re-upped (BMS expanded twice; Lilly signed two separate deals). That is a strong third-party signal that the platform clears pharma's internal bar. The March-2026 BMS expansion to ALS-2/-3 + a $10M milestone is the cleanest "it works enough to pay again" datapoint.
3. Founder / talent density (IP-people moat). Koller plus a cross-disciplinary bench (ML scientists, automation engineers, stem-cell biologists, drug hunters) is genuinely hard to assemble.

Bargaining power — honest read: weak vis-à-vis pharma. Insitro needs Gilead/BMS/Lilly's capital and development muscle more than they need any single platform — there are now ~10 well-funded techbio platforms competing for the same deals (Lens 7). The Lilly-TuneLab structure, where Insitro becomes a supplier inside Lilly's platform, shows pharma capturing the channel. Switching costs for pharma are low at the target-nomination stage.

Verdict on moat: Real but unproven where it matters. A data moat only compounds into enterprise value if the targets it nominates become approved drugs. Until an Insitro-originated molecule clears Phase 2, the moat is "best-in-class R&D services with optionality," not "defensible product economics."

Lens 4 · Segments

n/a — segment financials not disclosed (private, unaudited). Qualitatively, two implicit "segments":

1. Partnered platform revenue — milestone/option income from Gilead, BMS, Lilly (~$150M cumulative received ). This is ~all of today's cash inflow.
2. Proprietary pipeline — wholly-owned metabolic & neuro programs (pre-clinical; $0 revenue). This is where future equity value would come from. Geography: HQ South San Francisco; partners are US large-cap pharma. No geographic revenue split disclosed. `` qualitative only.

---

Phase B — Measure performance

Lens 5 · Pipeline & Funding trajectory (+clinical Lens-5 swap, fused with +private)

Pipeline by phase — the asset table IS the company. Insitro is entirely pre-clinical; there is no Insitro-originated molecule in human trials as of mid-2026.

. PoS per program: not independently estimable pre-IND — n/a (a pre-clinical small-molecule's industry-base PoS-from-here to approval is in the low-to-mid single-digit percent, but program-specific data is absent).

Management's own milestone: the May-2025 restructuring explicitly aimed to "maintain readiness for clinical trials in 2026" — so the single most important near-term catalyst is whether Insitro (or a partner) files a first IND / doses a first patient in 2026.

Funding & valuation trajectory (+private Lens-5).

Last private valuation: ~$2.4B (post-money, Series C, 2021). No new priced round since March 2021 — a ~5-year gap, itself a signal (Lens 11/13).

Burn / runway signals (the real news):

• May 2025: 22% workforce reduction (~65 people), cutting to ~230 staff, "to extend operational runway into 2027," Koller citing the "current tumultuous market environment" — she called it one of the hardest days of her career.
• Headcount 321 (Sep 2024) → 262 (Nov 2025), a ~18% decline on the trailing tracker.
• Read-through: with ~$643M raised, ~$150M collaboration cash in, no new equity since 2021, and a deliberate runway-extension cut, Insitro is managing scarcity, not growth. Runway "into 2027" means the financing/IPO clock is real — a raise or IPO is likely needed in the 2027 window.

Lens 6 · Founder / leadership signal trend (+private Lens-6 swap — interviews not earnings calls)

No earnings calls exist. Proxy = Koller's public posture and company messaging over time:

• 2019-2021 (capital-raising peak): evangelical "paradigm shift" framing — ML will reinvent drug discovery; raised at escalating marks into a $2.4B Series C.
• 2024 (commercial proof): tone shifts to validation — "$150M from partners," Lilly + BMS re-ups, ChemML positioned as "core engine for the industry's most difficult optimization challenges".
• 2025 (discipline / contraction): the layoff statement is sober and survival-framed — "tumultuous market," "runway into 2027," "readiness for clinical trials in 2026," pipeline focus narrowed to metabolic + neuroscience.
• 2026 (re-acceleration of narrative): POSH Nature-Comms validation (Dec 2025) + BMS ALS-2/-3 expansion (Mar 2026) — management re-establishing scientific + commercial momentum after the cut.

Sentiment arc: evangelism → validation → retrenchment → cautious re-acceleration. The "things they stopped saying": grand timelines and rapid-scaling language; the new vocabulary is runway, focus, readiness, capital discipline. That is the correct tone for the market — but it's the tone of a company husbanding cash to a catalyst, not one with the wind at its back.

Lens 7 · Catalyst calendar + Cap table & comps (+clinical + +private swaps, fused)

(a) Catalyst calendar — what de-risks or kills the story, and when. No PDUFA dates (nothing in registrational trials). The de-risking events:

(b) Cap table & secondary marks (+private). Syndicate quality is A-grade and IPO-relevant: tier-1 VC (a16z, ARCH, Third Rock, Foresite, GV), crossover funds T. Rowe Price and BlackRock (a classic IPO-proximity tell), sovereigns Temasek + SoftBank, and CPP Investments leading the C. But: that syndicate priced the company five years ago at $2.4B; no public markdown is disclosed, but the sector's repriced hard since (Lens 8), so the carrying mark is almost certainly stale-high. Secondary marks: n/a — not disclosed.

(c) Comps — public AI/techbio peers (provenance-critical). Insitro is private (no multiple). Peers, all ``:

**Private techbio comps (valuation marks, ``.

Lens 8 · Stock-Price Catalysts (sector proxy — +private)

No stock. The instructive "tape" is the public AI-discovery cohort, which tells you exactly what the market will reward/punish when Insitro eventually lists:

• Punished hard: clinical disappointment + revenue misses. Recursion −~60% off highs; RXRX Q1'26 revenue $6.5M vs expectations. Exscientia sold into Recursion at a fraction of peak after clinical setbacks; BenevolentAI / Exscientia / Recursion all de-rated post early-clinical results. Schrödinger −37% YoY.
• Rewarded: real, growing revenue. Tempus (a data/AI-diagnostics business with $1.6B revenue, +36%) is worth ~$8.3B — ~4× the rest combined — because it has an actual P&L.
• The lesson for Insitro: the market has stopped paying for "AI platform optionality" and started paying for either clinical de-risking or hard recurring revenue. Insitro has neither at scale yet. An IPO into this tape without a clinical asset would likely price below the 2021 mark.

---

Phase C — Judge people & books

Lens 9 · Management

Daphne Koller (Founder & CEO). Archetype: scientific-founder / domain-evangelist, the strongest possible profile for a platform-stage techbio.

1. Track record: built Coursera (co-founder, scaled MOOC platform); 18 years Stanford CS (Rajeev Motwani Professor); was Calico's Chief Computing Officer — i.e. she has already run computational biology at Alphabet scale before founding Insitro. Genuinely top-decile credibility at the ML-×-biology intersection.
2. Tenure & skin in the game: founder, ~8 years in seat. Insider ownership n/a — private, not disclosed, but founder-CEO of an unlisted company → presumed material equity + control.
3. Capital-allocation history: raised ~$643M; the defining decision is the May-2025 22% cut to protect runway into 2027 — that is disciplined, shareholder-aligned capital allocation in a brutal market, not empire-building. Signing the Lilly-TuneLab deal (turning ChemML into a revenue channel) and re-upping BMS are sensible monetizations of the platform while the pipeline matures.
4. Red flags: none of the classic ones (no related-party/promotional pattern surfaced). The honest concern is key-man risk — the platform's scientific identity is inseparable from Koller — and the 5-year gap since a priced round, which raises the question of whether the company can raise at/above the 2021 mark.
5. Founder vs professional manager: founder, and at this stage that's correct — a platform betting on a scientific paradigm needs the paradigm's author in charge.

Verdict: as good a founder-operator as exists in this category. Management is not the risk here; the science-in-the-clinic and the financing clock are.

Lens 10 · Forensic Red Flags

Standard income-statement / balance-sheet forensics are n/a — no audited financials are public (private, no 10-K). What can be flagged:

• Revenue quality / recognition: the ~$150M "revenue" is milestone + upfront + option income, which is lumpy, non-recurring, and partner-controlled — not product revenue and not a run-rate. The "$69M ARR" tracker figure is an outside estimate, explicitly marked "estimated" — do not treat as disclosed. Treat all Insitro revenue as soft, episodic, and concentrated.
• Going-concern / runway (the real forensic item for a pre-revenue private): deliberate layoffs to reach "into 2027" = cash is the binding constraint. A pre-clinical biotech with a 5-year-old last round and a stated runway ceiling is, by definition, financing-dependent within ~12-24 months.
• Valuation mark integrity: the $2.4B carrying value is a 2021 print held flat through a 35-60% sector de-rating — i.e. the headline mark is the most "aggressive" number in the file. Any new round re-marks it.
• Dilution: SBC/option dilution n/a — not disclosed; a future down-round would be the dilution event to watch.

Regulatory findings (required sub-section). Per regulatory/regulatory-findings.md (generated 2026-06-30): Insitro has no CIK and no SEC EDGAR enforcement history is searchable (LR + AAER return zero — it is private and not an SEC filer). Non-SEC web search — "Insitro" (FTC OR DOJ OR FDA OR CFPB OR "consent decree" OR settlement OR fine OR penalty) enforcement — returned no material enforcement actions, fines, consent decrees, or litigation against Insitro. No 10-K Item 3 exists (private). Conclusion: No material regulatory or legal findings — verified via SEC EDGAR EFTS (LR, AAER, both zero — non-filer), web search, and the absence of a public Item 3, as of 2026-06-30. (Unaudited per public sources.)

---

Phase D — Project & stress-test

Lens 11 · rNPV, runway-to-catalyst & IPO-readiness (+clinical + +private swap)

(a) The question that actually matters — does cash reach the next value-inflection? Management states runway "into 2027" post the 2025 cut. The next equity-value inflection (a proprietary clinical readout) is 2028+. Therefore current cash does NOT obviously reach the value-inflection catalyst — it reaches the first IND (2026) and the financing event, not a readout. This is the central tension: Insitro must raise again (or IPO) in the 2026-2027 window, almost certainly before it has the clinical data that would justify the 2021 mark. ``

(b) rNPV of the lead asset(s). n/a — not estimable. There is no disclosed lead asset in the clinic, no peak-sales guidance, no per-program PoS. Building an rNPV here would be fabrication. Honest placeholder: enterprise value today is option value on the platform, not discounted product cash flows. The defensible bound is the marked-to-market comp range in Lens 7 (~$1.0-1.6B ), well below the $2.4B headline.

(c) IPO-readiness & path-to-tradeable (the be-early payoff lens). No private-watch.json entry exists; from web evidence I assess:

• Stage: late-stage private, platform-validated, pre-clinical (no IND yet).
• IPO-readiness: LOW-to-MEDIUM, and gated. The 2025-26 AI-biotech IPO window is described as effectively shut — "IPOs are no longer a guaranteed, or even near-term, viable exit path". Peers are stuck private or down (only Insilico forced a Hong Kong listing). Insitro will not IPO at a premium to its 2021 mark without (i) a first IND/clinical asset and (ii) a friendlier biotech tape.
• Catalyst that unlocks an S-1: a first proprietary IND + a clean financing (flat-or-up round demonstrating the mark holds), plausibly a 2027-2028 window. A down-round or another cut would push it right.
• Write-back: cannot update private-watch.json — Insitro has no entry there (and research/watchlist.json edits are out-of-scope this wave). Open item: create the private-watch.json entry with stage: late-stage-private, ipo_readiness: low-med (gated on first IND + non-down round), catalyst: first IND 2026 / financing 2026-27 / IPO window 2027-28, and dossier: companies/insitro/deep-dive-2026-06-30.md.

Brier forecast (the binary that matters): the trackable call is "Insitro files a first IND (proprietary or partnered, attributable to its platform) by 2026-12-31" — I'd put p ≈ 0.45. Not logging via forecast.ts per the --watchlist no-side-effects rule; recorded here for a future /thesis pass.

Lens 12 · Bull vs Bear

Bull case. Insitro is the quality name in the biology-data lane of techbio, run by the field's most credible founder. The flywheel — proprietary multi-modal data → ML → targets — is the version of "AI for drug discovery" most likely to actually work, because it attacks target selection (where drugs really fail) rather than just speeding up chemistry. Three of the world's best pharma R&D buyers (Gilead, BMS, Lilly) keep paying and re-upping, BMS twice; the Lilly-TuneLab deal turns the platform into a revenue product; POSH's Nature-Comms validation shows the science is real. Management has shown rare capital discipline. If a first IND lands in 2026 and a proprietary asset shows even one clean human signal by 2028, Insitro re-rates from "R&D-services optionality" to "platform + pipeline" and the 2021 mark looks cheap — and it becomes the marquee AI-biotech IPO when the window reopens.

Bear case (2-3 permanent-impairment risks).

1. The whole category hasn't delivered in the clinic. BenevolentAI, Exscientia, Recursion — every public AI-discovery flagship that reached the clinic has disappointed, and the market has stopped paying for the promise (RXRX −60%, SDGR −37%). If "better targets" doesn't translate to better Phase-2 success, Insitro's core thesis is wrong, not just early — permanent impairment.
2. It's a cash-burning private at a stale 2021 mark with a 2027 runway ceiling. A down-round or a forced strategic sale (the article literally lists "being acquired outright while still private" as the realistic AI-biotech exit) re-prices equity below $2.4B. The financing clock is the nearest-term killer.
3. Pharma captures the value / commoditization. With ~10 well-funded platforms competing and pharma building its own platforms (Lilly TuneLab — Insitro is now a supplier inside it), the techbio platform may be a low-pricing-power services layer, not a product owner.

Pre-mortem (18 months out, thesis broke): It's end-2027. The 2026 IND slipped or a partnered program was quietly deprioritized; the company raised a down-round (or did a structured deal / acqui-hire) at well under $2.4B; another headcount cut followed; the AI-biotech-IPO window stayed shut. Narrative flips from "best techbio" to "expensive R&D shop that never got its own drug into humans."

Are multiples too high? There are no multiples (private). The mark (a flat 2021 $2.4B) is too high relative to marked-to-market peers — yes.

Contrarian view (what the market refuses to see): Both directions. The bear consensus ("AI drug discovery is hype") is over-applied to the one approach — proprietary-data + target-biology — that is genuinely differentiated from the discredited generative-chem hype; Insitro + the pharma re-ups may be the exception the tape is too jaded to price. But the bull consensus also refuses to see that being best-in-class doesn't matter if you're a pre-clinical private that has to raise into a closed window — quality doesn't beat the cash clock.

Lens 13 · Devil's Advocate (short-seller)

If this were public, here's how I'd dismantle the long:

• Revenue is a mirage. The "$150M" is episodic milestone/option money from three partners — it is not a business, not recurring, and partner-controlled. The "$69M ARR" is an outside estimate marked "estimated." Strip the narrative and you have a pre-revenue company that has burned most of $643M and just fired 22% of staff.
• Concentration is total. Lose/stall BMS or Lilly or Gilead and both the cash and the validation story crack. Gilead's NASH effort (the original 2019 deal) has produced no disclosed clinical Insitro-origin asset in ~7 years — what exactly came of it?
• The moat may be the weakest part. Lilly-TuneLab shows pharma building the platform and renting Insitro in — the value migrates to the data owner (Lilly) and the channel, not the algorithm vendor. Switching costs at target-nomination are near-zero.
• Most dangerous competitor bulls underrate: not another startup — Lilly/Big Pharma themselves (and Isomorphic, with Alphabet's balance sheet and $600M fresh). If pharma in-houses ML target-ID, the independent platform's reason to exist narrows.
• The capital-allocation tell: a 5-year-old last priced round + deliberate runway extension = the company cannot or will not test the market for a fresh up-round. That silence is the loudest number in the file.
• What breaks it permanently: a first clinical readout (theirs or a partner's flagship) fails → confirms AI-targets ≠ better drugs → platform de-rates to services multiple → forced down-round/sale. Plausibility: moderate-to-high given the entire cohort's clinical record.
• If growth disappoints 20-30%: for a private with no revenue base, the relevant shock isn't a revenue cut — it's a financing markdown; a 35-60% mark cut to ~$1.0-1.6B is the realistic stress outcome.

Lens 14 · Management Questions (ordered by information value)

1. What is the firm, dated plan for your first proprietary IND — is it still 2026, is it partnered or wholly-owned, and what single program is it? (The thesis hinges here.)
2. When do you next need to raise, and will you raise at, above, or below the 2021 $2.4B mark — what would a flat-or-down round mean for the strategy?
3. Of the ~$643M raised, how much cash is left today, and what is the quarterly net burn after the 2025 cut?
4. The 2019 Gilead NASH deal is ~7 years old — what clinical-stage asset, if any, came out of it, and is that collaboration still active?
5. In the Lilly-TuneLab structure, are you a supplier inside Lilly's platform — and how do you stop pharma from capturing the value your models create?
6. What is the single best piece of evidence that an Insitro-nominated target produces a better clinical outcome — not just a faster pre-clinical one?
7. Given Recursion/Exscientia's clinical disappointments, why is your data-first / target-biology approach structurally different from theirs, in a way that shows up in Phase 2?
8. How concentrated is revenue across BMS/Lilly/Gilead, and what is the plan to add a fourth top-5-pharma partner?
9. What is your wholly-owned vs partnered split of pipeline economics — where does equity value actually accrue to Insitro shareholders?
10. Is an IPO realistic before a clinical asset, or is a strategic sale / structured deal the more likely 2027 outcome?
11. What is your GMP manufacturing / CDMO strategy for proprietary clinical supply?
12. How much of the ChemML edge depends on Lilly's TuneLab data you don't own — what happens if Lilly changes terms?
13. What is the key-man succession plan, and how is the platform's scientific identity decoupled from you personally?
14. Which two programs did the 2025 reprioritization kill, and why those?
15. What metric should outside observers track quarterly to know the platform is working — and what would falsify the thesis?

---