Recursion Pharmaceuticals

Phase A — Understand the business

Lens 1 · Company Overview

Recursion is a clinical-stage "TechBio" decoding biology and chemistry to industrialize drug discovery. The Recursion Operating System (Recursion OS) runs iterative "atoms-and-bits" loops — wet-lab phenomics + multi-omics generate one of the world's largest proprietary biological/chemical datasets (the Recursion Data Universe), which trains ML models that propose novel mechanisms and molecules, which are then synthesized and tested, feeding the loop again. The November 2024 acquisition of Exscientia bolted a precision generative-chemistry engine onto Recursion's biology-first phenomics, making it one of the few end-to-end (target → molecule → clinic) AI-native shops.

How it makes money (today vs. eventually): Today, nothing from products. Revenue is "operating revenue" from pharma partnerships — milestone payments and data-map acceptance fees. FY2025 total revenue was just $74.7M against a $644.8M net loss. The company explicitly states it does not expect significant revenue "until the Company successfully completes significant drug development milestones or in collaboration with third parties, which the Company expects will take a number of years".

Two revenue/value engines:

1. Wholly-owned pipeline — internal oncology + rare-disease assets Recursion would develop and (eventually) commercialize or out-license. This is where the equity upside lives.
2. Partnered platform — Roche/Genentech, Sanofi, Bayer, Merck KGaA use the OS to generate maps/molecules; Recursion books upfronts + milestones + future royalties. >$500M received to date; >$20B in potential additional milestones before royalties.

Customers/partners (the "customers" of a platform pre-revenue biotech): Roche & Genentech (neuroscience maps — $213M cash inflows to date), Sanofi (oncology/immunology small molecules — $134M to date), Bayer (oncology, up to $1.5B potential ), Merck KGaA, and Tempus as a data supplier (a 5-yr, up-to-$160M oncology-data purchase, paid in cash or stock).

Contract structure: Milestone-and-royalty, not recurring/take-or-pay. Cash is lumpy and event-driven — a single map acceptance ($30M Roche neuron-map fee, 2024) or a $4M Sanofi milestone (Feb 2026) moves a quarter. Concentration is real: Roche/Genentech + Sanofi are ~$347M of the >$500M lifetime partner cash.

Lens 2 · Supply Chain (→ data + compute + CDMO)

For a TechBio the "supply chain" is data in → compute → wet-lab → CRO/CDMO → clinic → patient. Named stakeholders along the chain:

• Data inputs: Internally generated phenomics (the moat input) + purchased external data. Tempus supplies de-identified multimodal oncology data (DNA/RNA/records) under the up-to-$160M agreement; note Recursion purchased no records in Q1-2026 (vs. $27.1M in Q1-2025) — a burn-management tell that it has front-loaded the dataset and is throttling spend.
• Compute: Historically positioned around an NVIDIA-powered "top-50 supercomputer" ambition (BioHive). The NVIDIA equity relationship is over (NVIDIA fully exited the stock, Q4 2025 ), though the compute/collaboration relationship was characterized as still alive.
• Wet-lab / automation: Recursion's own robotic labs in Salt Lake City + Exscientia's Oxford/Vienna automated-synthesis sites generate the proprietary experimental data — this vertical integration is the differentiator vs. asset-light AI shops.
• Clinical execution: CROs run TUPELO (FAP), ELUCIDATE, DAHLIA, EXCELERIZE. Trials are conducted partly outside the US — a disclosed risk that FDA may not accept foreign data.
• CDMO/manufacturing: Small-molecule API — outsourced; not a near-term chokepoint at clinical scale.
• End customer: Patients via payers, years away. The nearer "buyer" is a pharma partner taking an asset forward, or FDA granting a registration path.

Chokepoint: The binding constraint is cash + clinical proof, not any physical input. The single-source dependency that matters is the platform itself — if OS-derived insights don't translate to clinical wins, no amount of data/compute helps.

Lens 3 · Competitive Advantages (moats)

The bull's moat claim: a compounding data flywheel. More proprietary phenomics → better models → better molecules → more clinical data → better models. Management leans hard on this — "when AI functions as a continuous system rather than a collection of tools, its impact compounds — and that compounding effect is the foundation of our long-term strategic advantage". The Recursion Data Universe (proprietary, internally manufactured) is a genuine asset few can replicate at scale, and the Exscientia merger gives true end-to-end coverage.

Bargaining power: Weak today. A pre-revenue biotech burning ~$370M/yr that just diluted ~100M shares via ATM in 2025 has little leverage over partners or capital markets. Pharma partners hold the cards on milestones; the market holds the cards on the next raise. The Bill & Melinda Gates Foundation even has a share-repurchase right if Recursion defaults on Exscientia's global-access commitments — a rare disclosed obligation that constrains.

Internal dogfooding as a soft moat signal: ~91% of the technology org actively uses AI coding tools and 35% of code is AI-authored — a credible "we live the thesis" data point, but it's productivity, not a customer moat.

The honest verdict: The moat is potential, not proven. The flywheel only compounds value if the output (clinical assets) is validated. As of mid-2026 there is exactly one AI-derived clinical proof point (REC-4881/FAP). Rivals are ahead on validation: Insilico has a Phase 2a readout published in Nature Medicine (rentosertib) and a $2.75B Lilly deal; Schrödinger's zasocitinib has positive Phase 3 data via Takeda. The "biggest dataset" claim is real but has not yet out-converted leaner peers.

Lens 4 · Segments

No reportable product or geographic revenue segments — revenue is collaboration milestones, not commercial product lines. The economically meaningful breakout is internal vs. partnered pipeline (Lens 1) and revenue components:

All ``. The revenue grew respectably, but it's a rounding error against opex — the point of the segment lens here is that revenue is not the value driver; the pipeline is. See Lens 5 (pipeline-by-phase).

---

Phase B — Measure performance (clinical overlay)

Lens 5 · Pipeline by Phase (swapped from "Earnings Result")

The asset table is the company. 5 clinical + 2 preclinical wholly-owned programs after the May 2025 cull:

All program/phase/readout dates ; TUPELO efficacy .

The financial print (for context, not the thesis): Q1-2026 — total revenue $6.5M (down from $14.7M YoY, a revenue miss ); net loss $117.5M (vs. $202.5M YoY — a big narrowing); EPS −$0.22 (vs. −$0.50), an "EPS beat"; total opex $135.0M, of which R&D $87.9M. The market read the narrower loss + extended runway positively (stock climbed on the print ). FY2025: revenue $74.7M, R&D $475.3M, G&A $176.6M, net loss $644.8M.

Discontinued (May 2025 cull): REC-2282 (NF2), REC-994 (CCM), REC-3964 (C. diff) — "totality of data" didn't justify continued spend. Notably, the homegrown legacy-Recursion assets were the hardest hit, while Exscientia-derived chemistry assets advanced — a quiet admission that early phenomics-only programs underperformed.

Lens 6 · Founder/Management commentary trend (transcripts absent → web)

No transcripts on disk; sentiment from the FY2025 letter + Q1-2026 call.

Tone shift (2024 → 2026): A clear pivot from vision/scale to proof/discipline. The 2025 CEO letter's spine is "translation" — "the credibility of AI in medicine will not be earned through better models alone, but through translation — from data to decisions, from platforms to pipelines, and from science to patients". New language: "rigorous capital allocation," "materially reduced projected cash burn," "transparent go/no-go decisions." This is the voice of a company that over-promised in the hype cycle and is now under-promising to rebuild credibility — captured in the trade press as Najat Khan "bringing Recursion back down to earth". Things they stopped saying: the grand "decode all of biology" maximalism of the Chris Gibson era. Recurring new phrases: "full-stack," "compounding," "go/no-go," "discipline."

Lens 7 · Catalyst Calendar + Mechanism Comps (swapped from "Comps")

Catalyst calendar (what de-risks or kills each program, and when):

Dates ; FDA-engagement framing .

Mechanism/peer comps (NOT P/E — this is pre-revenue): Comparable AI-native drug shops by validation stage and capital:

n/a for precise EV/Sales or P/E multiples across the peer set (pre-revenue or private; multiples would be fabrication). The honest comp read: Recursion has the most data and the most compute story, but is mid-pack on clinical validation and behind the private leaders on fresh capital.

Lens 8 · Stock-Price Catalysts (what the tape reacts to)

Over the trailing ~5 years the RXRX tape reacts to (1) smart-money flows, (2) clinical data, (3) partnership deals — far more than to the quarterly P&L (it has no earnings to beat). Identified >5% moves [all web]:

• NVIDIA full exit (Q4 2025 13F): stock −12% intraday — the single cleanest example of sentiment-over-fundamentals; NVIDIA's 7.71M-share stake had been a halo, and its removal gutted the "AI-blessed" premium.
• Mubadala full exit (Q4 2025): ~13M shares, reinforcing the smart-money-leaving narrative.
• REC-4881 TUPELO positive data (2025): the validating up-catalyst; "Is Recursion finally proving its model?".
• May 2025 pipeline cull + 20% layoff: stock down on the deprioritization, even though it extended runway.
• Q1-2026 print: up on narrower loss + runway extension despite revenue miss.

What it reveals: the market trades RXRX as a sentiment-and-catalyst name, not a DCF. The 52-week range ($2.77–$7.18) on a stock at ~$3.20 shows how violently narrative repricing swings it. Down ~21% over 12 months.

---

Phase C — Judge people & books (+ science & exclusivity)

Lens 9 · Management

CEO — Najat Khan, Ph.D. (since Nov 2025). Org-chemistry PhD (UPenn), computational-chemistry undergrad (Colgate). Prior: Chief Data Science Officer & SVP/Global Head of Strategy, Portfolio & Operations for R&D at J&J Innovative Medicine, where she is credited with helping "triple pipeline value," advancing ~10 potential blockbusters, and embedding AI across R&D. She joined Recursion as Chief R&D & Commercial Officer (2024), then took the CEO seat.

• Founder transition (material governance event): Co-founder Christopher Gibson stepped down as CEO and moved to an Advisory Agreement dated Nov 4, 2025; Khan's CEO offer letter is dated Nov 5, 2025. This completes a founder → professional-operator handoff — a stage-appropriate move for a company that needs disciplined capital allocation and FDA execution more than visionary storytelling.
• Track record / skin in the game: Khan's J&J record is operational and credible, but she has no public-company CEO P&L track record yet — the REC-4881 registration path is her first defining test. Insider ownership detail is incorporated by reference to the 2026 proxy (not in the 10-K); n/a here.
• Capital-allocation history (the team's, last 18 months): Acquired Exscientia ($630M, Nov 2024); then cut 3 programs + 20% of staff (~$11M severance) within ~6 months; throttled Tempus data purchases to zero in Q1-2026; funded almost entirely by dilution (~100M shares via ATM in 2025, net $491.7M). This is a fix-the-burn posture — defensible, but it's value-preservation, not value-creation. ROE/ROIC are deeply negative (pre-revenue); not meaningful.
• Archetype: Now a professional-manager shop with founder still in the wings (advisory). Implication: more discipline, less moonshot — the right archetype for the trust-rebuild phase, but it removes some of the narrative premium that made the stock a momentum darling.

Lens 10 · Forensic Red Flags

Acting as a forensic analyst. This is a clean-books, cash-incinerating biotech — the risks are dilution and going-concern adjacency, not accounting fraud.

• Going-concern adjacency: Accumulated deficit $2.2B. Management asserts cash is "sufficient for at least the next 12 months" — the standard non-going-concern language — and notes it "did not have any unconditional outstanding commitments for additional funding" and "will likely be required to raise additional capital". Not a going-concern qualification, but the disclosure is explicit that another raise is coming.
• Revenue recognition: Collaboration revenue is the classic biotech soft spot — milestone timing can flatter a quarter. The Q1-2026 revenue fell YoY (no large map fee recurred), which actually argues against aggressive recognition. The Tempus data purchase is expensed to R&D as records download, with over/under-payment parked as "Prepaid data assets" / "Accrued data liability" — a judgment area worth watching but disclosed cleanly.
• Stock-based comp flattering the loss: SBC is large — $111.2M in FY2025 (≈17% of the net loss); $22.0M in Q1-2026 alone. $159.1M of unrecognized RSU cost remains. Non-cash, but it is real dilution and inflates the "we cut cash burn" narrative — cash opex fell faster than GAAP opex partly because SBC is non-cash.
• Dilution machine: Weighted shares went 402.8M → 529.3M YoY (+31%). A fresh $300M TD Cowen ATM is armed (Feb 2026, $0 drawn at Q1) — i.e., the next dilution is pre-loaded.
• Goodwill/intangibles from M&A: Two acquisitions (Valence 2023, Exscientia 2024) created intangibles; FY2025 D&A jumped to $83.7M and asset impairment of $6.0M was taken. Impairment risk on Exscientia goodwill is live if pipeline value disappoints — a watch item.
• Auditor change: Ernst & Young (2017–2024) → PricewaterhouseCoopers (since 2024). The switch coincides with the Exscientia merger (a common reason for re-tendering), and the 10-K notes ICFR attestation by the new auditor with no error-correction/restatement flags — not a red flag on current read, but auditor changes always warrant a note.
• Dual-class control: Class B = 10 votes/share concentrates voting with insiders — governance, not accounting.

Regulatory findings (required sub-section):

• SEC Litigation Releases: None. No LR naming Recursion since 2021-06-18.
• AAERs: None in the search period.
• 10-K Item 3 (Legal Proceedings): Only ordinary-course litigation — "The Company is, and may in the future from time to time be, involved in various legal proceedings arising in the normal course of business," cross-referenced to Note 7 (Commitments & Contingencies); no specific material matter disclosed.
• Non-SEC (FTC/DOJ/FDA/CFPB) web search: No material enforcement actions, consent decrees, fines, or penalties surfaced for Recursion as of 2026-06-18.
• Conclusion: No material regulatory or legal findings — verified via SEC EDGAR EFTS (LR, AAER), web search, and 10-K Item 3 as of 2026-06-18.

Science & exclusivity (clinical-overlay add):

• Mechanism validation: REC-4881 (MEK1/2 in FAP) is the only clinically validated OS-derived insight — the MEK1/2-rescues-APC-loss hypothesis translated to a durable polyp-burden reduction. The rest of the pipeline is mechanistically rational but clinically unproven.
• IP estate: Proprietary Data Universe + Exscientia chemistry IP; specific patent-cliff/LOE timing n/a (pre-commercial; no marketed product to lose exclusivity on).
• Reimbursement path: Not near-term for any asset; FAP (orphan, no approved pharma) would carry orphan economics if REC-4881 reaches approval — a favorable payer setup if it gets there.

---

Phase D — Project & stress-test

Lens 11 · rNPV + Runway-to-Catalyst (swapped from EPS projection)

For a pre-revenue biotech the question is not "FY28 EPS" — it's (a) does cash reach the next value-inflection catalyst, and (b) what is the lead asset worth risk-adjusted?

Runway (the survival question) — the more important one:

• Cash + money-market at Q1-2026: $654.5M (excl. $10.7M restricted); $665M incl. restricted.
• FY2026 cash-burn guidance: <$390M.
• Management states runway extends into early 2028 with risk-adjusted partner inflows; FY2025 10-K MD&A frames it more conservatively as "at least the next twelve months".
• : $654.5M cash − ~$390M FY2026 burn ≈ $265M entering 2027; at a similar (or slightly lower, post-discipline) burn, base-case cash runs thin around late 2027 / early 2028 absent the $300M ATM or new milestones. The armed $300M ATM is effectively the bridge — meaning dilution is the base case, not the bear case.
• Verdict on runway: Cash does reach the two pivotal 2H26 catalysts (REC-4881 FDA path; REC-1245 data) — but probably not the full 1H27–2H27 readout cluster without a raise. The runway is calibrated to get to proof, then refinance off a (hoped-for) re-rate.

rNPV of the lead asset (REC-4881/FAP):

• FAP US prevalence ~1 in 7,000–10,000 → ~30–50k US patients; addressable treated population far smaller.
• Orphan pricing + no approved competitor → assume peak US sales $300–600M if approved.
• Phase-2-to-approval PoS for oncology/rare-disease ≈ 15–25%, nudged up by clean MoA + orphan path → use 20%.
• rNPV ≈ peak sales × 3–4x sales multiple × PoS × discount → **$200–450M** risk-adjusted for REC-4881 alone. Against a $1.67B market cap, the market is therefore paying ~$1.2–1.5B for the platform + partnerships + the rest of the pipeline — i.e., the equity is mostly platform optionality, not de-risked clinical value.
• Brier forecast (logged conceptually, not via forecast.ts per --watchlist rules): "REC-4881 receives an agreed FDA registrational path by end-2026, p ≈ 0.40" — the single highest-information binary for the thesis.

Lens 12 · Bull vs Bear

Bull case. Recursion is the most vertically integrated AI-native drug shop — owns the data generation, the chemistry (post-Exscientia), and now clinical execution (ClinTech). The flywheel is real and, if REC-4881 secures a registration path, the market re-rates the entire platform as "AI drug discovery that actually produces approvable drugs." Optionality is enormous: 5 clinical + 2 preclinical wholly-owned shots, >$20B in potential partner milestones, and a balance sheet (post-cull) that no longer screams panic. At ~$3.20 with a $1.67B cap and $665M cash, you're paying ~$1B EV for a platform that took a decade and >$2B to build. The new CEO has credible big-pharma R&D pedigree and has imposed exactly the discipline the company lacked. Secular tailwind: pharma is pouring billions into AI discovery (Lilly/Insilico $2.75B, Lilly/Isomorphic $1.75B) — validation that the category is real.

Bear case. Three things could permanently impair the equity: (1) the platform doesn't translate — one Phase-2 win in a tiny orphan indication after a decade and $2.2B of accumulated losses is thin proof that the OS produces better drugs than conventional discovery; if REC-617/1245/3565 disappoint in 2027, the flywheel thesis collapses. (2) Forced dilution at a low price — the $300M ATM is armed and runway is calibrated to a re-rate that may not come; raising at $3 destroys per-share value and the share count already grew 31% YoY. (3) Smart money has already voted — NVIDIA and Mubadala fully exited in Q4 2025; the most informed holders left, and competitors (Insilico, Schrödinger) are further along on validation with deeper pharma checks.

Pre-mortem (18 months out, thesis broke): REC-4881's FDA interaction yields a long, expensive registrational trial rather than an accelerated path; the 2027 oncology readouts are unremarkable in crowded CDK7/MALT1 fields; the company raises $300M+ via ATM at <$4; the "AI compounding" narrative is quietly dropped from the deck; RXRX trades sub-$2 as a busted platform story. Plausible.

Are multiples too high? There are no earnings multiples. On EV-to-pipeline, the market is already skeptical (down 21%, smart money gone), so the equity is arguably not over-loved — the risk is less "priced for perfection" than "priced for a coin-flip that could still lose."

Contrarian view (what the market refuses to see): The market is treating Recursion as a failed AI-hype trade (NVIDIA dumped it, so it must be over). But the category just got the biggest pharma validation in its history (multi-billion Lilly deals), and Recursion is the only public pure-play with both end-to-end vertical integration and a real clinical PoC. If REC-4881 gets a clean path, the "left-for-dead" sentiment inverts violently — the same narrative volatility that crushed it cuts both ways. This is a binary, not a slow bleed.

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

Dismantling the bull case. The structural break: Recursion's business model is "spend ~$400M/yr to generate data and hope the platform produces approvable drugs faster/cheaper than incumbents." After a decade and $2.2B, the scoreboard is one Phase-2 signal in a niche orphan disease — and the homegrown phenomics assets were the ones cut, while the acquired Exscientia chemistry assets survived (i.e., the original thesis underperformed and they bought their way to a better pipeline). Revenue concentration: ~$347M of >$500M lifetime partner cash is Roche+Sanofi; lose a flagship partner and the "platform demand" story cracks. The moat is weaker than bulls think: "biggest dataset" hasn't out-converted leaner rivals — Insilico got to a Nature Medicine Phase 2a and a $2.75B Lilly deal with far less capital; Isomorphic raised $2.1B privately without burning a public balance sheet. Most dangerous competitor bulls underestimate: Isomorphic Labs — Alphabet's compute + AlphaFold lineage + $2.1B fresh capital and no quarterly dilution pressure; it can out-spend and out-recruit Recursion indefinitely. Worst capital allocation: funding a decade of losses almost entirely through serial equity dilution (share count +31% YoY) while the stock fell — shareholders financed the science and got diluted for it. What must hold for $3.20: that REC-4881 converts to a real registration path and the platform produces a second and third validated asset before cash forces a dilutive raise. If growth/validation disappoints 20–30%: there's no earnings floor — a busted platform biotech with $2.2B of accumulated losses and an armed ATM trades on cash-per-share, ~$1.25/share, implying material downside. Single scenario that permanently impairs: REC-4881 FDA path is murky/long AND the 2027 oncology data is mediocre → the AI-discovery premium evaporates, the company becomes a serial diluter, and it's eventually a sub-$2 zombie or a fire-sale acquisition. Plausibility: moderate-to-high — this is the base-rate outcome for most clinical-stage biotech, AI-branded or not.

Lens 14 · Management Questions (15, ordered by information value)

1. After your 2H26 FDA interaction on REC-4881, what specifically would a "win" look like — accelerated approval on polyp-burden endpoint, or a multi-year outcomes trial — and what is your honest probability on each?
2. Your runway reaches early 2028 with risk-adjusted partner inflows. At what stock price and date does the $300M TD Cowen ATM get drawn, and what's the trigger to not draw it?
3. The homegrown phenomics programs (REC-994/2282/3964) were the ones cut, while Exscientia chemistry assets advanced. Does that mean the original phenomics-first thesis underperformed — and what does it imply about the platform's true edge?
4. Insilico reached a Phase 2a Nature Medicine readout and a $2.75B Lilly deal with a fraction of your capital. Why has the industry's largest proprietary dataset not produced faster clinical validation?
5. NVIDIA and Mubadala both fully exited in Q4 2025. What did they see — or stop believing — and what would bring that class of investor back?
6. You say AI authors 35% of your code and the platform "compounds." Translate that into a falsifiable metric: by what date will platform-derived programs show a measurably higher Phase-1→2 success rate than conventional discovery?
7. Of the >$20B in potential partner milestones, what is the risk-adjusted NPV you actually underwrite internally, and how much is from the top two partners?
8. What is the single most likely reason REC-1245, REC-617, or REC-3565 fails in 2027, and how correlated are those risks given they share the platform?
9. How do you defend recruiting and compute economics against an Alphabet-backed Isomorphic with $2.1B and no dilution pressure?
10. Goodwill/intangibles from Valence + Exscientia: under what pipeline scenario do you take a material impairment, and what's the trigger?
11. What is your hard go/no-go burn discipline — if two 2027 readouts disappoint, do you cut deeper or raise, and at what dilution threshold do you instead seek a sale/partnership for the whole platform?
12. SBC was $111M last year (~17% of the net loss). How should investors think about the gap between your "cash burn down 30%" message and GAAP losses?
13. With the founder now on an advisory agreement and you in the CEO seat, what specifically changes in strategy that couldn't have happened under the prior structure?
14. The Bill & Melinda Gates Foundation share-repurchase right tied to Exscientia's global-access commitments — under what circumstances could it be triggered, and what's the cash exposure?
15. Five years out, is Recursion a drug company that sells approved medicines, a platform company that licenses to pharma, or an acquisition target — and which are you building toward?

---