Insilico Medicine

Phase A — Understand the business

Lens 1 · Company Overview

Insilico Medicine is a "dual-engine" AI-driven drug-discovery company: a software platform (Pharma.AI) that it licenses to pharma, sitting on top of an in-house therapeutic pipeline that it discovers, advances through the clinic, and out-licenses. Founded 2014 by Alex Zhavoronkov; Cayman TopCo, Hong-Kong-listed since Dec 2025, with major operations in the US (New York / Boston), Hong Kong, mainland China (incl. a robotics-driven automated lab), and the UAE.

How it actually makes money — three revenue lines:

1. Drug-discovery services / software (Pharma.AI licensing) — fee-for-access to Chemistry42, Biology42 (PandaOmics/TargetPro), Science42. FY2025 "drug discovery fees" jumped to US$24.95M (from US$3.14M in 2024) — the high-margin, recurring base.
2. Pipeline out-licensing & collaboration income — upfronts + milestones from partnering its own assets. FY2025 "pipeline development income" fell 68.8% to US$23.89M — lumpy by nature (it spikes on a deal, collapses without one).
3. Milestones/royalties from prior deals.

Total FY2025 revenue US$56.24M, −34.5% YoY — the decline is the tell that this is not a smooth-compounding SaaS company; it is a deal-driven biopharma whose top line whipsaws with licensing timing.

Customers / partners: software has served 13 of the top-20 global pharma and 40+ companies in total; named structural partners include Sanofi (up to $1.2B biobucks, 2022), Servier ($888M, $32M upfront, Jan 2026, oncology), Exelixis (USP1 license, $80M upfront, 2023), Menarini ($550M+, $20M upfront), Fosun Pharma (multi-target + China co-dev of rentosertib), and Qilu Pharmaceutical ($120M cardiometabolic, Jan 2026). Cumulative announced collaboration value ~$4.6B; >10 new deals worth $1.3B signed in 2025.

Contract structure: classic biopharma licensing economics — modest upfronts (tens of millions), large contingent milestone "biobucks" (mostly never paid in full), single-digit-to-tiered royalties. The headline $4.6B "cumulative value" is risk-unadjusted — treat it as marketing, not bookings.

Lens 2 · Supply Chain (→ platform & manufacturing dependencies)

For an AI-bio name the "supply chain" is compute + data + CRO/CDMO + chemistry, mapped end-to-end with named stakeholders:

• Upstream compute / models: GPU infrastructure for generative models (transformer/GAN/diffusion + reinforcement learning). Insilico runs Pharma.AI on cloud + its own infra; Tencent is both a cornerstone IPO investor and a cloud-ecosystem fit. Single-source risk is moderate — frontier compute is a commodity but a scarce one.
• Data: proprietary multi-omics + literature + chemical libraries feed PandaOmics (targets) and Chemistry42 (molecules). The data moat is the real input — see Lens 3.
• The company (midstream): target ID → generative chemistry → an automated robotics lab in Suzhou for synthesis/assay (vertical integration that most AI-bio peers lack).
• CRO / CDMO (downstream of design): Insilico does not run its own large-scale clinical trials end-to-end — it leans on CROs (FY2025 R&D fell specifically because "CRO fees" dropped after rentosertib finished Phase 2a ). WuXi AppTec is a pre-IPO shareholder and the obvious CRDMO partner of record — a China-bio dependency that is also a US-policy exposure (BIOSECURE-style risk).
• End customer: ultimately patients/payers via a big-pharma licensee (Sanofi, Servier, Exelixis, Fosun) who runs late-stage trials and commercializes. Insilico rarely touches the patient directly — it is an upstream IP factory, not an integrated pharma.

Chokepoints: (1) CRO/CDMO capacity & the China-CRO geopolitical overhang (WuXi); (2) frontier compute; (3) the scarce human capital — computational chemists + the founder. Names present; this is not a generic map.

Lens 3 · Competitive Advantages (moats)

The contrarian read: Insilico's moat is evidence, not algorithms. Generative-chemistry software is increasingly commoditizing (open models, Recursion, Schrödinger, Iambic, Chai, plus pharma's own in-house stacks). What Insilico has that almost no one else does is a clinical proof-point: rentosertib is the first molecule with both an AI-discovered novel target (TNIK) and an AI-designed molecule to post a positive randomized Phase 2a in a top-tier journal (Nature Medicine, Jun 2025). That is a genuine, hard-to-replicate credential — for now.

Durable-moat scorecard:

• Data/process moat (real, decaying): the closed-loop "target → molecule → wet-lab → clinic → feed results back" flywheel with a robotics lab is hard to copy and improves with every program. But the half-life is short — rivals are building the same loop.
• Switching costs (weak): pharma partners can and do multi-source AI vendors. The Sanofi/Servier deals are target-specific, not platform lock-in.
• Brand / first-mover (strong, temporary): "the company that got the first AI drug into Phase III" is a marketing asset that wins the next BD conversation. It is not a patent.
• IP (real on assets, thin on platform): the molecules (rentosertib, ISM3091, ISM5411, ISM6331, etc.) carry composition-of-matter patents — that is where defensibility actually lives. The platform is trade-secret + know-how, not a fortress.
• Bargaining power: asymmetric against Insilico vs. big pharma (Sanofi/Servier need Insilico less than Insilico needs the validation + cash). Stronger vs. Chinese partners (Fosun, Qilu) where Insilico brings the scarce AI capability.

Verdict on the moat: a 2–3 year evidence lead and a vertically-integrated flywheel — real, but narrowing. The bet is whether Insilico converts the lead into approved drugs and royalties before the platform edge erodes.

Lens 4 · Segments (revenue by line & geography)

segments.csv is empty — all ``, from the FY2025 results:

Read: the quality of revenue improved even as the level fell — software (recurring, scalable) overtook one-off licensing for the first time. The gross-margin compression to 81.5% reflects more services/wet-lab mix vs. pure software. Geography: not cleanly disclosed publicly; the footprint is US + China + UAE, with China-partner deals (Fosun, Qilu) accelerating — a deliberate pivot toward the China-bio funding pool. The decline year masks a genuine mix-shift toward the better business.

---

Phase B — Measure performance (clinical-stage variant)

Lens 5 → Pipeline by phase (swapped per +clinical: the asset table is the company)

The pipeline is the equity story. Lead + key clinical assets:

Pleiotropic/format note (per +clinical for metabolic): the cardiometabolic basket spans the full incretin axis (GLP-1R/GIPR/amylin) plus APJ (cardio) and Lp(a) (the genuinely under-served, differentiated target). Insilico is years behind Lilly/Novo on incretins but the Lp(a) and combination angles are the credible AI-differentiation play, de-risked by the Qilu partnership covering downstream development.

Concentration: ~70%+ of the narrative value sits on rentosertib in IPF. Everything else is early Phase I optionality or partnered. This is a one-asset re-rate story dressed as a platform.

Lens 6 → Founder/management commentary trend (no earnings calls — first reporting period)

Insilico has only one reporting period as a public company (FY2025, reported H1 2026) — no multi-call sentiment series exists yet. Tone proxy from the FY2025 release + founder interviews:

• Consistent message: "dual-engine," "value delivery," "pharmaceutical superintelligence," Phase III on track H2 2026. Management is promotional by nature — Zhavoronkov is a prolific publisher and conference presence (also runs Deep Longevity, a longevity-biomarker spin-off, and the Biogerontology Research Foundation).
• What they emphasize: the evidence (Nature Medicine), cumulative biobucks ($4.6B), top-20-pharma logos.
• What to watch them stop saying: if "Phase III in H2 2026" slips, or if the $4.6B cumulative number stops being repeated (would signal milestone slippage). First real sentiment delta lands at FY2026/interim 2026 results.

Lens 7 → Catalyst calendar + mechanism comps (swapped per +clinical)

Catalyst calendar (what de-risks or kills the equity, and when):

• H2 2026 — Rentosertib oral Phase III initiation (IPF). The single biggest binary on the calendar. Starting it de-risks regulatory alignment; the data (2027–2028) is the real event.
• 2026 — ISM3091 (USP1) Phase I dosing (Exelixis-run) — partner validation read-through.
• 2026 — ISM5411 (PHD1/2) Phase IIa IBD data — second clinical proof-point for the platform.
• 2026 — new BD deals — the lumpy revenue engine; each ~$100M+ deal is a tape catalyst.
• Rentosertib inhaled Phase I data — incremental.

Mechanism comps (by approach/target, not P/E) — public AI-drug-discovery peers:

Read: Insilico trades roughly in line with Recursion despite being smaller on revenue — the premium is the clinical-validation scarcity (and the China/Asia-IPO bid). On a pipeline-credibility basis that premium is defensible; on a revenue/quality basis it is rich.

Lens 8 → Stock-price catalysts (since IPO — only ~6 months of tape)

Listed 30 Dec 2025; short history, but the moves are instructive:

• 30 Dec 2025: IPO at HKD 24.05, +25% debut (touched ~HKD 37), 1,427× retail oversubscription — a sentiment/scarcity event (AI + biotech + first-of-kind + Lilly/Tencent cornerstones), not a fundamentals event.
• Jan 2026: Servier ($888M) and Qilu ($120M) deals within days of listing — BD momentum fed the run.
• 20 Feb 2026: ATH HKD 80.90 — a >3× from IPO in seven weeks. Pure momentum/retail melt-up; no commensurate fundamental catalyst.
• Feb–Jun 2026: round-trip down to ~HKD 40 (Jun 30) — roughly −50% from peak, ~+67% vs IPO. The de-rate is the market exhaling the IPO euphoria; no negative clinical event drove it.

What the tape reveals: this name trades on narrative and flows (AI-bio sentiment, China-IPO liquidity, BD headlines), with a thin float and high retail ownership → high beta to AI-hype cycles and to any rentosertib headline. Fundamentals (revenue) are almost irrelevant to the short-term tape; the next fundamental that will matter is Phase III progression.

---

Phase C — Judge people & books

Lens 9 · Management

• Alex Zhavoronkov (Founder & CEO): unusual hybrid — PhD Physics (Moscow State), MS Biotech (Johns Hopkins), CS + Finance (Queen's), early career at ATI Technologies (GPUs, pre-AMD). Clarivate Highly Cited Researcher (pharmacology/toxicology, 2022 & 2025); prolific publisher.
• Track record: built Insilico from 2014 to the first AI-designed-drug Phase 2a + the largest 2025 HK biotech IPO — genuinely impressive company-building. Also founded Deep Longevity (longevity biomarkers) and chairs the Biogerontology Research Foundation.
• Skin in the game: founder-led, significant retained stake (exact post-IPO % n/a — not cleanly sourced; pre-IPO he and early backers held the bulk; Qiming ~7%).
• Capital allocation: disciplined on cash (debt-free, $393M post-IPO, adjusted burn only ~$44M/yr); R&D cut 11% in 2025 as a program graduated — capital-light by AI-bio standards.
• Founder archetype: scientist-promoter-founder. The promotional energy (longevity media, "superintelligence" framing, cumulative-biobucks marketing) is a double-edged trait — it raises capital and signs deals, but it is also exactly the profile a short-seller targets. Watch for over-claiming.
• Red flags (governance): (1) founder-distraction risk — multiple ventures (Deep Longevity, the foundation); (2) promotional cadence — the $4.6B "cumulative value" headline is risk-unadjusted; (3) China/Cayman/HK structure — VIE/related-party complexity is standard for the structure but under-disclosed in English sources (n/a — verify in HK prospectus); (4) WuXi AppTec as both shareholder and CRDMO supplier = related-party + geopolitical exposure.

Lens 10 · Forensic Red Flags (+ Science & exclusivity per +clinical)

Income-statement & balance-sheet forensics, all `` from FY2025 results:

• The US$352M net loss is mostly an accounting artefact, not cash: US$296.7M of it is a non-cash fair-value loss on the conversion of preferred shares at listing (the standard IFRS "convertible-preferred remeasurement" hit that nearly every HK/China biotech IPO books). Adjusted (non-IFRS) loss was US$43.83M. Do not read the $352M headline as operating distress — but do note that a $296.7M paper loss flatters how cheap the equity now looks vs. retained earnings.
• Revenue recognition: licensing income is lumpy and upfront-loaded — the −68.8% pipeline-income drop is a recognition-timing artefact, not necessarily demand loss. Watch how much "milestone" revenue is recognized up-front vs. deferred.
• Receivables/inventory: not flagged; software + licensing model is asset-light. n/a — detailed working-capital lines not sourced.
• SBC: an AI-talent company will carry meaningful stock-based comp inflating non-IFRS adjustments — quantum n/a — not separately sourced; scrutinize at next results.
• Cash vs. earnings: operating cash burn (~$44M adj.) is well below the headline loss — cash flow is healthier than GAAP earnings, the opposite of the usual red flag.
• Going concern: clean. $393.34M cash, debt-free, 0% gearing → ~9 years of runway at the current adjusted burn (will compress materially once Phase III spending ramps — see Lens 11).

Science & exclusivity (per +clinical):

• Mechanism validation: TNIK is a novel, AI-nominated target — that is the upside (genuine novelty) and the risk (no prior validated drug against it; Phase 2a is the only human evidence). Biological de-risking is incomplete — AI sped up discovery, not biology.
• IP/exclusivity: composition-of-matter patents on the molecules are the real exclusivity; platform is trade-secret. Patent-life on rentosertib n/a (verify in prospectus).
• Reimbursement: IPF is a high-value orphan-ish indication with established pricing (Ofev/Esbriet) — payer path is well-trodden if it gets approved.

Regulatory findings (read from regulatory/regulatory-findings.md, Step 0):

• SEC (EDGAR LR + AAER): none possible — Insilico has no CIK and is not an SEC registrant.
• Non-SEC web search ("Insilico Medicine" (FTC/DOJ/FDA/CFPB/consent decree/settlement/fine/penalty) enforcement): no material enforcement actions surfaced as of 2026-06-30. FDA contact is ordinary IND clearances (rentosertib inhaled, ISM3091), not enforcement.
• Item 3 / Legal Proceedings: no 10-K exists (foreign filer); HK-prospectus legal-proceedings section not retrieved — flag to verify.
• Net: No material regulatory or legal findings — verified via the no-CIK SEC limitation, web search, and the absence of any enforcement coverage as of 2026-06-30. The unverified item is the HK prospectus's own litigation/related-party disclosure.

---

Phase D — Project & stress-test

Lens 11 → rNPV + runway-to-catalyst (swapped per +clinical)

This is a pipeline-NPV equity, not an EPS equity — projecting EPS would be false precision (it will lose money for years). Frame it as rNPV of the lead asset + runway:

Valuation anchor `]:

• Market cap HKD 21.88B ÷ ~7.8 HKD/USD = ~US$2.81B
• Less net cash US$0.39B (debt-free) → EV ≈ US$2.41B
• EV / FY2025 sales (US$56.24M) ≈ 43×; P/S ≈ 50× — a pipeline multiple, not an operating one.

rNPV sketch of rentosertib (IPF), the value driver:

• IPF drug market ~US$3.0B (2025) → ~US$6.7B by 2035; a differentiated antifibrotic could reasonably capture ~US$1.0–1.5B peak sales.
• Probability of success from Phase III-start: ~50–55% for Phase III→approval in a de-risked-MoA antifibrotic, but TNIK is un-precedented → haircut to ~35–45%.
• Royalty/own-economics, discounted: a rough rNPV of US$0.8–1.5B for rentosertib alone is defensible. Add Servier/Exelixis/Sanofi milestone streams + platform/software value (~$25M recurring, growing, at a SaaS-ish multiple) and the ~$2.4B EV is roughly fair-to-full — it is already pricing a credible Phase III path, leaving limited margin of safety if Phase III data disappoints.

Runway-to-catalyst (the question that matters): cash US$393M, adjusted burn ~$44M/yr today → headroom is ample now, but a global Phase III in IPF runs US$100–200M+. Even so, $393M + ongoing BD upfronts (Servier $32M, Qilu, etc.) comfortably funds the next value-inflection (Phase III data) without a forced raise — a genuine strength vs. cash-strapped US AI-bio peers. Runway reaches the catalyst: YES.

No Brier forecast logged (unattended --watchlist rule). If promoted, the scoreable forecast would be: "Rentosertib IPF Phase III primary endpoint met, p≈0.40, resolves ~2028."

Lens 12 · Bull vs Bear

Bull case. Insilico is the only AI-drug-discovery company that has crossed from promise to clinical proof (Nature Medicine Phase 2a), it has a vertically-integrated discovery flywheel (target→molecule→robotics-lab→clinic) that compounds, a diversified pipeline of ~8 clinical assets that gives multiple shots-on-goal, a recurring software base that grew ~7× in 2025, a blue-chip partner roster (Sanofi, Servier, Exelixis, Menarini, Fosun, Qilu) throwing off non-dilutive upfronts, $393M cash + zero debt funding it to Phase III data without a raise, and a cornerstone register (Lilly, Tencent, Temasek) signaling cross-industry conviction. If rentosertib reads out positive in Phase III, this re-rates as the validated AI-pharma platform and the multiple is justified.

Bear case (3 ways it permanently impairs). (1) Rentosertib Phase III fails or underwhelms — a +98mL FVC delta in n=71 over 12 weeks is encouraging but modest (Ofev itself only slows decline); the entire premium evaporates and the platform's "validation" unwinds. The IPF graveyard is real — Pliant halted bexotegrast in 2026 on a safety imbalance. (2) Platform commoditization — generative chemistry is racing toward open-source/table-stakes; pharma's in-house AI + Recursion/Schrödinger/Iambic erode the licensing edge and the BD pricing power. (3) Lumpy, deal-dependent revenue + promotional management — the −34.5% revenue year shows how fragile the top line is; a quiet BD year + the $4.6B-biobucks marketing make it a prime short target.

Pre-mortem (18 months out, thesis broke). Most likely: rentosertib Phase III is delayed (slips past H2 2026) or the readout is equivocal; meanwhile a BD-light 2026 prints another revenue decline; the AI-bio sentiment cycle turns; the thin float amplifies a 50–70% drawdown from here as IPO-lock-ups unwind. The biology — not the AI — is what breaks it.

Are multiples too high? At ~43× EV/sales the equity is priced for a Phase III win. That is not absurd for a validated platform, but it is the opposite of a margin of safety. Cheap it is not.

Contrarian view (what the market refuses to see): Both bulls and bears over-index on rentosertib/IPF as the "AI validation." The durable value may actually be the software + China-partnership engine (Qilu, Fosun) — a capital-light royalty-on-AI model that compounds regardless of any single trial. The market is pricing the lottery ticket (IPF) and under-pricing the toll-booth (Pharma.AI). Equally contrarian on the downside: "first AI drug" is a marketing moat with a 24-month shelf life, not a durable one.

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

Dismantling the bull case:

• Revenue concentration & lumpiness: FY2025 fell 34.5%; two-thirds of "value" is contingent biobucks that mostly never pay. A single dry BD quarter craters the print. Where's the recurring base? ~$25M of software — tiny against a $2.4B EV.
• The moat is thinner than bulls think: "AI-discovered" is a process claim, not a patent. Pharma is internalizing the exact stack; the "first AI drug" credential decays the moment a competitor's AI molecule reads out — or fails (Pliant already cracked the AI/integrin-IPF halo).
• Most dangerous competitor bulls underestimate: not Recursion — it's big pharma's own in-house AI + the open-model wave (Chai/Iambic/Boltz/AlphaFold-class) that turns Chemistry42 into a feature, not a company.
• Capital-allocation / governance: founder runs multiple ventures (Deep Longevity, foundation) — distraction risk; WuXi is shareholder and supplier (related-party + BIOSECURE geopolitical exposure); the $4.6B cumulative-biobucks headline is promotional; HK/Cayman/VIE structure under-disclosed in English.
• What must hold for today's price: rentosertib Phase III starts on time and reads out positive, BD stays at ~$1B+/yr, software keeps compounding, and AI-bio sentiment doesn't turn. That's four things, all uncertain.
• Growth disappoints 20–30%: revenue is already volatile ±35% — a soft year is the base case, not a tail. At 43× sales, a re-rate to even 20× EV/sales (still rich) is a ~50% downside before any clinical setback.
• Single scenario that permanently impairs: rentosertib Phase III fails (plausible: ~50–65% by base rates, higher given a novel target and a modest Phase 2a). That single event would re-rate the entire "validated platform" thesis to a "promising-but-unproven" multiple — and take the China-retail bid with it. Plausibility: moderate-to-high.

Lens 14 · Management Questions (ordered by information value)

1. What is the exact statistical powering, primary endpoint, sample size, and duration of the rentosertib Phase III — and does it benchmark against placebo or against background standard-of-care (Ofev/Esbriet)?
2. Given the +98mL Phase 2a delta over 12 weeks in n=71, what effect size are you powering Phase III to detect, and what's your candid read of replication risk for a first-in-class TNIK mechanism?
3. What does 2026 BD look like — how many deals/upfront dollars are realistically in the pipeline, and how do you stop revenue swinging ±35% year to year?
4. How much of the $4.6B "cumulative collaboration value" has actually been collected in cash to date, and what's the realistic risk-adjusted figure?
5. What is the post-IPO insider/founder ownership and the lock-up schedule — when do pre-IPO holders (Warburg, Qiming, Hillhouse, WuXi) become free to sell?
6. Walk through the WuXi AppTec relationship — shareholder and CRDMO supplier: how do you manage the related-party conflict and the US-policy (BIOSECURE-type) exposure?
7. As open-source and pharma-internal generative chemistry commoditize, what is the durable, defensible edge of Pharma.AI in three years — beyond "we were first"?
8. What is the fully-loaded cost and funding plan for Phase III(s), and at what cash level would you raise equity rather than burn the balance sheet?
9. How do you intend to monetize the software engine — is it a true recurring-SaaS line, or just project fees rebadged? What's the path to $100M+ recurring?
10. Which second pipeline asset (ISM3091 USP1, ISM5411 IBD, ISM6331 TEAD) is the next value-inflection, and what de-risks it?
11. On cardiometabolic — you're years behind Lilly/Novo on incretins; is the real bet Lp(a)/combinations, and what's genuinely AI-differentiated there?
12. How do you allocate capital across owning assets (full economics, full risk) vs. out-licensing early (cash now, capped upside) — what's the framework?
13. What's your China strategy — Fosun/Qilu suggest a pivot to the China-bio funding pool; how do you balance that against US/EU partnerships amid geopolitical risk?
14. The CEO runs Deep Longevity and the Biogerontology Foundation — how is your time allocated, and what governance prevents founder-distraction?
15. What single result in the next 18 months would most change your own view of the company — bull or bear?

---