Recursion Pharmaceuticals
Dual engine: (1) platform partnerships with large pharma (Roche/Genentech, Sanofi, Bayer) that pay upfront + milestone + royalty economics for AI-discovered targets/molecules; (2) internal pipeline of wholly-owned clinical candidates. Pre-product-revenue; funded by cash + partner milestones, not drug sales.
Earnings, margins, COGS & capex
Recursion is a pre-commercial platform company: it burns ~$390M/yr to run an AI drug-discovery engine and a growing clinical pipeline, offset partially by big-pharma milestone revenue. FY2025 revenue rose to $74.7M but net loss widened to $644.8M as R&D (incl. data purchases + Exscientia integration) climbed to $475.3M. The 2026 story is cost discipline (30% YoY cash-opex reduction) plus 2H26 clinical readouts, with a balance sheet that funds operations into early 2028.
Revenue trend
Margins
improving on cost cuts but structurally pre-revenue
widened from -$463.7M FY2024 (Exscientia integration + data purchases)
narrowed from -$202.5M Q1'25; EPS beat consensus (~-$0.28 est)
improving - deliberate burn reduction
COGS structure
No meaningful product COGS (no marketed drugs). Cost base is R&D-dominated: FY2025 R&D $475.3M (up from $314.4M FY2024; incl. data-generation, compute, and Exscientia integration); Q1 FY26 R&D fell to $87.9M from $129.6M, G&A to $34.6M from $54.7M as synergies landed.
Capex
Capitalized spend on automated laboratory automation and the BioHive-2 GPU supercomputer (built with NVIDIA); ongoing but not the dominant cash driver - burn is R&D/opex-led, not capex-led.
Latest earnings
Mixed: EPS beat (-$0.22 vs ~-$0.28 est) on cost discipline; revenue big miss ($6.5M vs ~$16.1M est) on milestone timing
Reiterated FY2026 operational cash burn of <$390M; cash runway into early 2028
- Cash & equivalents (end Q1'26)
- $665.2M
- FY26 op cash burn guide
- <$390M
- Cash opex YoY
- -30%
- REC-4881 TUPELO polyp reduction
- Wk13: 5/6 evaluable pts (83%) reduced, 43% median; Wk25 (12wk off-therapy): 9/11 (82%) durable, 53% median
Growth drivers
- Big-pharma partnership milestones — >$500M cumulative upfront + progress payments to date; Roche/Genentech ($213M received), Sanofi ($134M received), Bayer
- Clinical pipeline de-risking — REC-4881 (FAP) Phase 1b/2 TUPELO signals and a potential FDA registrational path
- Exscientia acquisition (closed Nov 20, 2024, ~$688M all-stock) adding generative-chemistry + precision-medicine capabilities and its own partner book (Merck KGaA, BMS, others)
- Recursion OS platform scale — proprietary phenomics dataset + BioHive-2 compute as a data/compute moat that can seed new partnerships
Reported financials — SEC EDGAR
Audited GAAP figures pulled from SEC filings · latest filing 2026-02-25. The audited primary-source spine — not financial advice.
Revenue — annual (GAAP)
Margins & balance sheet — FY’25
Bull & bear
Recursion is the best-capitalized, most-industrialized platform in AI drug discovery, with a real partner book, a fortress balance sheet into 2028, sharpened clinical leadership, and multiple 2H26 catalysts that could finally convert the platform narrative into clinical proof - at a market cap (~$2.0B) that is largely cash + optionality.
- ~$665M cash funds the company into early 2028 - it does not need the capital markets to reach its key 2H26 readouts
- REC-4881 in FAP is showing genuine Phase 1b/2 efficacy (82-83% of evaluable patients with polyp-burden reduction; durable at 25 weeks) with FDA engagement on a registrational path - a credible first shot at a wholly-owned approval
- Cost discipline is real: 30% YoY cash-opex cut and a narrower loss show the new CEO is running it like a drug company, not a science project
- >$500M of non-dilutive partner milestones received validates that Roche, Sanofi and Bayer will pay for the platform's output
- At ~$2.0B market cap against ~$665M cash and a Top500 supercomputer + one of the largest phenomics datasets, the platform + pipeline optionality is cheap if any program de-risks
- Category leadership: post-Exscientia, Recursion is the scaled public pure-play on the 'first AI-designed drug' thesis
Recursion has burned over $1.1B in net losses in two years to produce lumpy milestone revenue and zero approved drugs; the AI-discovery thesis remains unproven in the clinic, marquee backer NVIDIA has fully exited, and the stock sits near 52-week lows because the market is discounting years more dilution before any payoff.
- FY2025 net loss of $644.8M on $74.7M of milestone revenue - the burn-to-revenue ratio is unsustainable without repeated dilution
- The core promise (AI finds better drugs faster) has no marketed-drug proof point after a decade; pipeline is still early Phase 1/2
- NVIDIA divested its entire ~7.71M-share stake (Feb 2026), removing the validator that partly drove the 2023-24 re-rating and sending the stock down ~12% on the news
- Revenue is milestone-lumpy and non-recurring ($6.5M in Q1'26 vs a ~$16.1M consensus) - it cannot be modeled as a growing base
- Clinical binary risk: a single failed 2H26 readout (REC-4881 or REC-1245) removes the catalyst and re-opens the 'overhyped platform' narrative
- Crowded, well-funded competition (Alphabet's Isomorphic Labs, Insitro, Xaira, Schrodinger) means even success may not translate to durable pricing power
What it is worth
Sum-of-parts optionality / reverse read on cash + platform + pipeline, not an earnings multiple (no earnings, milestone-lumpy revenue). EV ~$1.35B after netting ~$665M cash against ~$2.0B market cap.
$2-3 / ~$1.0-1.6B (toward cash value): a failed or ambiguous 2H26 readout and/or partner attrition removes the catalyst, the AI-discovery narrative de-rates, and the equity drifts toward net-cash-per-share with dilution overhang.
~$3.5-5 / ~$1.8-2.6B
incremental pipeline and partnership progress with continued burn discipline; stock range-bound as the market waits for the 2H26 binary readouts and prices mostly cash + modest optionality.
$7-10+ / ~$3.5-5B+ mkt cap
REC-4881 secures a clear FDA registrational path AND REC-1245 shows early efficacy, plus a new marquee partnership - platform thesis re-rates and category leadership commands a premium.
At ~$3.79/share the market values Recursion at roughly cash plus ~$1.35B of enterprise value for the platform, partner book, and clinical pipeline combined. With no product revenue and lumpy milestones, the price is essentially an option on (a) 2H26 clinical de-risking and (b) new/expanded pharma partnerships - it cannot be supported on a near-term revenue or EBITDA multiple. The reverse-DCF sanity: the current EV implies the market credits the platform with eventually producing at least one commercially meaningful asset or a materially larger partnership annuity; a clean 2H26 readout would validate that, a miss would compress it toward cash.
SWOT
Strengths
- One of the largest proprietary phenomics datasets in the industry + owned wet-lab automation generating internally-consistent biological data at scale
- NVIDIA-built BioHive-2 supercomputer (Top500-ranked, ~#76 in 2025; #35 at launch in May 2024) - rare in-house compute for a biotech
- Validated partner book — Roche/Genentech, Sanofi, Bayer, plus Exscientia's legacy partners - >$500M received to date
- Strong balance sheet (~$665M cash) with runway into early 2028 - can survive the biotech funding winter
- Sharpened leadership — CEO Najat Khan (ex-J&J Innovative Medicine chief data science officer; joined Recursion 2024 as Chief R&D + Chief Commercial Officer) took the helm Jan 1, 2026 as co-founder Chris Gibson moved to Chairman, signaling a sharper clinical-execution focus
Weaknesses
- No marketed products and no product revenue — revenue is lumpy milestone recognition ($6.5M one quarter, $35.5M another)
- Massive cash burn ($644.8M FY2025 net loss) - profitability is years away and dilution risk is real
- Platform-to-clinic translation still unproven at scale — the core bull thesis (AI finds better drugs faster) lacks a marketed-drug proof point
- Loss of the NVIDIA equity stake (fully divested, ~7.71M shares, Feb 2026) removed a marquee validator and pressured the stock
- Integration risk / overlap from the large Exscientia acquisition — ongoing headcount and program rationalization
Opportunities
- 2H26 clinical catalysts: REC-4881 FDA registrational-path clarity (FAP), REC-1245 Phase 1 dose-escalation data
- New multi-target pharma partnerships leveraging the expanded post-Exscientia platform and generative chemistry
- Rare-disease/oncology niches (FAP, RBM39-degrader, CDK7) where speed-to-clinic and precision could command premium value
- Selling/licensing platform access (data, models, compute) as a recurring, higher-quality revenue stream
- Broader TechBio tailwind — if any AI-discovered drug reaches approval, the whole category (and RXRX as a leader) re-rates
Threats
- Clinical failure risk — a negative REC-4881 or REC-1245 readout would remove the near-term catalyst and hit the equity hard
- Partners can walk: milestone economics depend on Roche/Sanofi/Bayer continuing to advance programs
- Capital-markets risk - if runway shortens, equity raises at a depressed price are dilutive
- Crowded AI-drug-discovery field (Schrodinger, Isomorphic Labs/Alphabet, Insitro, Xaira) racing for the same 'first AI drug' proof point
- Skeptic narrative that AI drug discovery is overhyped and hasn't yet delivered a differentiated approved medicine
Moats, dependencies & bottlenecks
Moats
Moderate-Strong One of the largest internally-generated, consistent biological image/perturbation datasets; hard and expensive to replicate, but data alone hasn't yet proven to yield better clinical outcomes.
Rare owned supercomputer for a biotech (~#76 on the 2025 Top500); a real capability edge, though compute is increasingly rentable and NVIDIA has exited the equity.
Multi-year Roche/Genentech, Sanofi, Bayer deals with embedded workflows; partners can still decline to advance programs.
Vertical integration from phenomics to generative chemistry (post-Exscientia) is differentiated but not yet output-proven.
Dependencies
Revenue / validation Milestone revenue and non-dilutive funding depend on partners advancing programs; concentration in a handful of deals.
Pre-revenue burn (~$390M/yr) means eventual equity raises; a low share price makes them dilutive.
BioHive-2 built with NVIDIA hardware; NVIDIA sold its equity stake (Feb 2026) but remains the compute supplier of record.
The equity's re-rating hinges on 2H26/2027 readouts (REC-4881, REC-1245, REC-617).
REC-4881 registrational path depends on ongoing FDA engagement.
Advantages
- Best-capitalized public pure-play in AI drug discovery (~$665M cash, runway to 2028)
- Vertically integrated platform (phenomics data + wet-lab automation + owned supercomputer + generative chemistry)
- Proven ability to sign and monetize top-tier pharma partnerships (>$500M received)
- Deep, proprietary, internally-consistent biological dataset
- Sharpened clinical-execution leadership under CEO Najat Khan (in seat since Jan 1, 2026)
Weaknesses
- No approved drug and no product revenue after ~a decade - platform ROI unproven
- Very large, structural cash burn with dilution risk
- Lumpy, non-recurring milestone revenue
- Loss of NVIDIA as an equity validator; stock near 52-week lows
- Binary clinical risk concentrated in a few near-term readouts
Bottlenecks
- Translating platform-discovered targets into clinically differentiated, approvable drugs - the unproven step
- Cash burn vs. time-to-value: multi-year gap between spend and any product revenue
- Milestone-lumpiness makes revenue non-recurring and hard to grow linearly
- Talent and integration bandwidth to fully absorb Exscientia and rationalize the combined pipeline
- Dependence on a small number of partner relationships for external validation and cash
Top signals & trends
Top signals
bullish if positive · The clearest near-term path to a wholly-owned approval; a defined registrational path would be a major de-risking event.
bullish if positive · Potential first-in-class degrader; clean safety + early efficacy would validate platform-derived novel targets.
Continued cost discipline extends runway and reduces dilution risk.
Fresh upfronts would validate the post-Exscientia platform and add non-dilutive cash.
Watch cumulative partner payments, not any single quarter's number.
NVIDIA's full divestment (Feb 2026) removed a marquee holder and pressured sentiment.
Trends
The category is racing for a first differentiated AI-designed approved drug; RXRX is a scaled public bellwether - re-rates with category proof points.
Favors well-capitalized names; RXRX's ~$665M cash and burn cuts are a survival advantage vs weaker peers.
Structural tailwind for partnership revenue if platforms deliver validated targets/molecules.
Owned BioHive-2 is an edge, but cloud GPU rentability lowers the barrier for competitors over time.
REC-1245 and REC-617 ride the degrader/CDK7 targeted-oncology wave.
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
GPU/compute supplier that co-built BioHive-2; note: NVIDIA fully divested its ~7.71M-share RXRX equity stake in Feb 2026.
Compute and storage for large-scale ML training/inference beyond owned hardware.
Instruments, consumables and reagents feeding the automated wet-lab phenomics factory.
Neuroscience + oncology discovery collaboration; ~$213M received in upfront/milestones to date; 6 phenomaps accepted.
Multi-target discovery collaboration; ~$134M received; fifth milestone achieved.
Discovery collaboration advancing programs toward lead-series milestones.
Legacy Exscientia partners inherited via the acquisition (BMS among them).
Physics-based computational drug-discovery platform + internal pipeline; the closest public platform-plus-pipeline analog.
Alphabet/DeepMind spinout (private) built on AlphaFold; arguably the most credible deep-pocketed AI-discovery rival.
Private ML-driven drug-discovery company (Daphne Koller); machine-learning + functional-genomics approach overlapping RXRX.
Motion-based (computational) precision-oncology drug design; competes for the same 'compute-enabled better molecules' thesis.
AI/data-driven antibody discovery platform with a partnership-royalty model similar to RXRX's.
Generative-AI protein/antibody design platform; smaller but same category.
AI + clinical/molecular data platform; adjacent (diagnostics/data) rather than direct discovery, but competes for the 'AI in biopharma' narrative and data supply.
Well-funded private AI-first drug-discovery startup (backed by ARCH/Foresite); a fresh, deep-pocketed entrant.