Recursion Pharmaceuticals: A Sub-$375m Cost Reset Buys Runway into Early 2028, but $3.63 Already Pays for About 66% of a $1.28bn Platform Option

The assignment's description of the REC-4881 “new” Phase II readout contains a material dating error. The 43% three-month median polyp reduction is not a post-June-2026 disclosure. Recursion published the fuller Phase 1b/2 TUPELO data on December 8, 2025, off a November 25, 2025 cutoff. It reported a 43% median reduction at week 13 among 12 evaluable patients and a 53% median reduction at week 25 among 11 patients, after the planned treatment period had ended; 73% of the latter group had at least a 30% durable reduction. Preliminary 43% data in an even smaller cohort had been reported in May 2025. The August 2026 materials add regulatory-pathway framing, upper/lower gastrointestinal analyses and future-readout timing, but they are not the first disclosure of the efficacy result.

That chronology changes price attribution. RXRX closed at $3.32 on June 8, 2026 and $3.63 on September 4, a roughly 9% gain. The stock fell after the August 5 Q2 release, closing around $3.17 as the revenue miss competed with the narrower loss and lower expense guidance, then recovered about 15% into the September 4 close. No fresh 43% REC-4881 efficacy disclosure landed during that recovery. Calling the rebound “clinical de-risking” would be inaccurate. Catalyst anticipation, the lower burn rate, broad biotech appetite and a stock with unusually heavy short interest are better explanations; how much each contributed cannot be established from public evidence.

REC-4881's own evidence carries weight. FAP has no approved drug therapy in Recursion's cited market framing, and a reproducible reduction in upper and lower gastrointestinal polyp burden could matter clinically. Yet the dataset is small and uncontrolled, and polyp burden is a surrogate, not a proven reduction in colectomies, cancer or mortality. At the 4 mg dose, the August materials showed treatment-related adverse events in most patients, three Grade 3 events, no Grade 4/5 events, and four treatment discontinuations. What endpoint and trial design FDA will accept for registration matters nearly as much as another small-cohort efficacy update.

REC-1245 and REC-7735 matter more to the “AI-native” claim than their current rNPVs suggest. REC-1245 is an RBM39 molecular-glue degrader that Recursion says went from 204 synthesized compounds to a development candidate in about 18 months. Through the March 31, 2026 cutoff, 16 Phase I patients showed no dose-limiting toxicity and no Grade 4/5 treatment-related events. But tested exposures were still below the exposure associated with tumor regression in the company's mouse models, and no clinical efficacy signal had yet been established. REC-7735, a mutant-selective PI3Kα H1047R inhibitor, was selected in roughly ten months and is claimed to be more than 100-fold selective for H1047R over wild-type PI3Kα; at the base date it still had no human efficacy evidence.

Those examples support one part of Recursion's AI argument: selected discovery/design cycles can run shorter and use fewer synthesized molecules. The company compares roughly 330 compounds per program and around 1.5 years to candidate selection against industry figures around 2,500 compounds and four years. REC-617, REC-1245 and REC-7735 are program-level anecdotes consistent with the claimed efficiency. The harder proof is missing: a statistically credible improvement in clinical likelihood of approval. Industry datasets put overall Phase-I-to-approval success below 10%, with Phase II still the largest attrition point. Recursion has no approved drug and too few internally originated clinical programs to show that its faster discovery engine produces a higher clinical success rate.

Leadership and ownership need the same correction for timing. Najat Khan became CEO and president on January 1, 2026 after the board announced the succession on November 4, 2025. Chris Gibson became chair on the same date, then completed his board term at the June 17, 2026 annual meeting without seeking reelection. NVIDIA's entire 7.71 million-share holding had already been sold by December 31, 2025, with the exit disclosed in February 2026. All of that changed how the market viewed Recursion's external endorsement and founder continuity. But it was known before the June 8 prior report, except for Gibson's actual June board departure, which had itself been announced beforehand. None of it should be used to explain a fresh post-June surprise.

Qualitatively, this is a company in transition. It spent the 2010s and early 2020s proving that industrialized phenomics could attract capital and pharma partners, then expanded through chemistry, data and Exscientia acquisitions. The cost of that ambition became too high for the post-2021 biotech capital environment. The 2025–26 response has been to prune, consolidate and force the platform into fewer, more consequential clinical bets. The next stage will be judged on human data, partner option exercises and dilution, not on dataset size.

Vertical history, financial evolution and capital-market narrative

Recursion began in Salt Lake City in 2013 out of a specific scientific frustration. Co-founder Chris Gibson had been an M.D./Ph.D. student at the University of Utah working on disease biology under Dean Li; the early idea was to use automated microscopy and phenotypic screening to observe what disease and treatment did to cells, at a scale no human researcher could inspect manually. Gibson left medical training after his Ph.D. to build the company. Blake Borgeson supplied computational and engineering leadership. Recursion's early institutional identity followed from those backgrounds: experimentally generated images first, algorithms second, with software used to navigate biology instead of merely searching the scientific literature.

Behind that identity sat a bet that drug discovery suffered from low throughput and too much dependence on human hypotheses. Recursion set out to industrialize the experimental layer itself. Large automated laboratories would perturb cells, create high-dimensional phenotypic images and generate proprietary data; machine learning would then find relationships among genes, diseases and chemical compounds. That architecture separated it from pure in-silico companies. It also made for a capital-heavy version of AI drug discovery, because proprietary wet-lab data, microscopes, robotics, compute and scientists all had to be financed before any medicine reached patients.

Bayer supplied the first major external proof. In August 2020, before the IPO, Bayer signed a five-year research collaboration initially focused on roughly ten fibrosis projects. The agreement later expanded and, in 2023, was reoriented toward oncology as Bayer's strategy changed. That arc says something useful about platform partnerships: a large-pharma customer can validate the technical relationship while changing therapeutic priorities, which makes individual programs and revenues less durable than a conventional subscription contract.

Recursion went public in April 2021 at $18 per share. Including the underwriters' option, it issued about 27.9 million shares and raised roughly $463 million net. The pitch centered on “industrializing drug discovery,” an operating system built on massive proprietary biological data, automated experimentation and machine learning. The market heard that story in the final phase of the zero-rate biotechnology boom, when long-duration scientific optionality commanded unusually high valuations.

Its public-company history splits into four stages.

The first stage, 2021 through 2022, was platform-scale construction. Roche and Genentech added a transformative collaboration, including a $150 million upfront payment that cut 2022 operating cash burn sharply. Revenue rose. But the economic lesson was already visible: one large upfront cheque could make a year's cash flow look dramatically better without creating anything resembling recurring operating profitability. Cash used in operations was $158.6 million in 2021 and only $83.5 million in 2022, largely because that Roche payment arrived during 2022.

The second stage, 2023 through much of 2024, bought breadth with capital. Recursion picked up Cyclica and Valence capabilities, deepened chemistry and machine learning, upgraded BioHive computing and entered a $50 million strategic investment and AI-model relationship with NVIDIA in July 2023. NVIDIA bought more than 7.7 million shares at the time. The announcement produced exactly the kind of external-validation re-rating that later became a recurring feature of RXRX trading: investors read the identity of the strategic partner as evidence about the platform's quality, even though the investment did not remove drug-development risk.

It was also when the equity base began to expand quickly. At year-end 2022 Recursion had about 191.0 million common shares across its classes; by year-end 2023 the number was about 234.3 million. Financing cash flow in 2023 was $140.1 million against $287.8 million of operating cash burn. Equity capital was the balancing item between scientific ambition and internal cash generation.

Exscientia changed the scale of the organization. The companies announced the all-stock transaction in August 2024 at an indicated value around $688 million and closed it on November 20, 2024. Exscientia became wholly owned, pairing Recursion's phenomics and automated biology with AI-guided chemistry and adding partnered and internal programs. Recursion issued more than 100 million shares for acquisitions in 2024; the accounting value of shares and options issued for acquisitions was about $630 million.

The strategic rationale was clear. Recursion had spent years claiming an advantage in discovering disease biology and linking phenotypes to compounds. Exscientia supplied a more developed precision-chemistry engine. In theory the combined company could move from biological insight to optimized molecule without handing work between separate organizations. The financial consequence arrived at once: bigger headcount, more clinical programs, duplicated sites and higher R&D/G&A. The 2025 annual report says the full-year inclusion of Exscientia lifted the cost base and contributed to a record $644.8 million net loss.

The third stage was integration meeting a hostile funding environment. In June 2025 Recursion announced a roughly 20% workforce reduction, as biotechnology funding conditions and policy uncertainty forced management to choose between organizational breadth and runway. The subsequent 10-K describes strategic prioritization, site consolidation and lower headcount. That was more than a routine cost program. It marked the point where the capital market stopped rewarding maximum pipeline breadth and began demanding evidence that the combined platform could concentrate resources around assets with realistic clinical and partnership value.

The fourth stage began with the November 2025 leadership succession and is still running. The board appointed Najat Khan CEO effective January 1, 2026, while Gibson moved first to chair and then left the board after the June 2026 annual meeting. Khan had joined Recursion in 2024 and served as chief R&D and commercial officer before taking the top job. Her background fits the new stage: clinical prioritization, portfolio economics and commercial discipline now matter more than proving the platform can generate ever-larger datasets.

NVIDIA's equity exit reinforced the transition. Its latest disclosed position fell from 7.71 million shares at the end of Q3 2025 to zero at December 31, 2025. The February disclosure hit RXRX sharply in early trading, though the stock recovered much of that reaction. No public filing gives NVIDIA's reason. I read the sale as the removal of a signaling asset, not as evidence that NVIDIA had inside knowledge about Recursion's clinical prospects. The two are not the same thing: the companies' technical relationship and NVIDIA's ownership were never economically identical.

Across that arc, the financial record reads:

SEC filings give the net-loss, cash-flow, capex and share-denominator series directly; revenue growth came entirely from collaborations and grants, never from product sales.

Revenue grew sevenfold from 2021 to 2025 while the underlying economics deteriorated, because expenses grew much faster. Conventional “revenue CAGR” is close to useless here. A $30 million map milestone can move one year's growth rate dramatically while saying little about next year's recurring revenue. The annual report also says two customers represented substantially all operating revenue in 2025; in 2024, one customer represented primarily all operating revenue.

Earnings quality is an odd question here, because there are no positive earnings to assess. Cumulative 2021–2025 net losses were roughly $1.86 billion; cumulative operating cash burn was approximately $1.26 billion. The roughly 0.68 ratio of operating cash loss to accounting net loss reflects large non-cash items, including stock compensation, depreciation/amortization and acquisition accounting, plus the timing of partnership cash. In 2025 alone, stock-based compensation was $111.2 million and depreciation/amortization $83.7 million. These accounting add-backs help cash flow relative to net loss but represent real dilution or consumption of previously funded assets.

The balance sheet is in much better shape than the income statement. At June 30, 2026, cash and equivalents were $545.7 million, restricted cash roughly $11.1 million, total assets $1.21 billion and total liabilities $297.7 million. Goodwill plus identifiable intangibles were about $443 million, mostly from acquisitions. There is no near-term conventional debt-refinancing problem here; the financial risk comes from cash burn and equity issuance.

Share dilution is the permanent feature investors should keep beside every pipeline slide. The year-end common-share count rose from 234.3 million in 2023 to 396.8 million in 2024, driven heavily by Exscientia and financing, and the company reported 535.34 million Class A shares outstanding as of July 31, 2026. That last figure sits about 128% above the 2023 year-end common-share count cited in the statements. The comparison crosses share classes and acquisition-related issuance, so read it as an order-of-magnitude indicator, not a clean same-class dilution statistic.

Recursion added another financing instrument in February 2026: an at-the-market program with TD Cowen permitting up to $300 million of Class A common-stock sales. No shares had been sold under it through June 30, leaving the full $300 million available. At the September 4 price of $3.63, issuing the whole amount would take roughly 83 million shares before fees, about 15% of the July 31 share count. Nothing forces the company to use the ATM soon. But its existence makes “runway into 2028” a different statement from “no dilution before 2028.” Management can rationally raise earlier if clinical data open a favorable window.

Governance changed more than a simple CEO title swap suggests. The April 2026 proxy showed ten directors before Gibson's departure, eight of whom the board classified as independent. After the annual meeting the remaining/elected slate consists of Najat Khan, Franziska Michor, Blake Borgeson, Zachary Bogue, Namandjé Bumpus, Zavain Dar, Robert Hershberg, Dean Li and Elaine Sun, subject to any later appointment not disclosed in the materials reviewed here. The proxy had specified that Hershberg would remain lead independent director after Gibson left, pending appointment of a new chair.

The dual-class structure is already gone. Class A carried one vote per share, Class B ten. At December 31, 2025 Gibson and affiliates controlled all 5,547,334 outstanding Class B shares and about 9.6% of total voting power. The Q2 2026 10-Q reports that all previously outstanding Class B shares converted into Class A during the three months ended June 30, 2026, leaving none outstanding at that date. Gibson's board service ended in the same month, so the concentrated founder voting power disappeared with the conversion rather than at a later charter deadline, and the outstanding equity now votes as a single class.

The latest company-compiled 5%-holder register in the April proxy identified Vanguard at 47.19 million Class A shares, or 9.0%; BlackRock at 37.82 million, or 7.2%; and ARK Investment Management at 37.32 million, or 7.1%. The underlying ownership reports generally reflected December 31, 2025 positions, so this is a lagged register, not a September 2026 real-time one. NVIDIA was absent after its complete Q4 2025 exit.

Capital markets keep changing the label attached to RXRX. It arrived as an “industrialized drug discovery” growth platform at $18. The 2023 NVIDIA deal turned it briefly into an AI-adjacent strategic-validation trade. Exscientia encouraged a vertically integrated TechBio narrative. Then the 2024–25 biotech funding reset, mixed legacy clinical results and cost expansion pushed investors back toward a cash-burn framework. The current price of $3.63 is almost 80% below the IPO issue price, yet the share count is vastly larger. A low nominal share price says little about valuation.

Conventional historical multiples mislead just as badly. At $3.63 the equity is worth about $1.94 billion. Subtract June cash plus restricted cash and enterprise value is around $1.39 billion, roughly 18 times 2025 revenue and nearly 49 times annualized H1 2026 revenue. That apparent multiple expansion is mostly a denominator problem caused by milestone timing. The useful valuation framework for Recursion has shifted from price-to-sales toward net cash plus pipeline rNPV plus the value of future partner economics.

Business model, moat, industry and horizontal peers

Everything in the platform business starts with proprietary experiments. Recursion perturbs cells genetically or chemically, captures high-dimensional phenotypic information, folds in chemistry and other biological/clinical data, and uses machine learning to find relationships that may be useful for target identification, drug design and patient selection. Exscientia added AI-guided medicinal chemistry; Cyclica and Valence broadened computational chemistry and machine-learning capability. Partners can pay upfront fees, research funding and progress milestones, then make option or licensing decisions that create further milestones and royalties.

The internal pipeline runs on much of the same machinery but keeps more economics inside Recursion. That creates a basic capital-allocation tension. A partnered program hands much of the downstream cost and risk to a pharma company, but it caps Recursion's economics at milestones and royalties. A wholly owned drug can eventually be worth billions if approved, but Recursion has to finance clinical development and possibly commercialization itself. The 2025–26 restructuring effectively acknowledges that the company cannot fund every platform-generated hypothesis as an internal asset.

Today's program set is narrower than the combined post-Exscientia organization first suggested. The August 2026 materials show five clinical-stage programs: REC-4881, REC-617, REC-1245, REC-3565 and REC-4539. REC-7735 had an IND cleared and was expected to begin its first-in-human ZINNIA study in the second half. REC-102 remained earlier-stage. That also corrects a loose formulation in the assignment: the CDK7 inhibitor REC-617 was still an active Phase I/II program in the latest disclosure, with updated clinical data discussed in the 2025 10-K. It had not disappeared from the pipeline.

Cost structure explains why losses grew faster than revenue. Core fixed and semi-fixed expenses cover scientists, engineers, clinical staff, laboratory and computing infrastructure, data systems, leases and central functions. Variable costs cover external datasets, CRO work, consumables and partner-specific service activity. Q2 showed that some supposedly “platform” expense is quite flexible: lower Tempus purchases accounted for $19.6 million of the quarterly platform-R&D decline. Clinical expense is less flexible once patients are dosed and studies are running. Q2 clinical R&D rose about 21% even as total R&D fell 30%.

Operating leverage works differently here from a software platform. A new pharma milestone can generate a high incremental accounting contribution in one period, while ongoing research obligations can produce cost of revenue before or after milestone recognition. Q2's $11.49 million cost of revenue against $7.30 million of operating revenue is the mismatch in one line. Royalties from an approved partnered drug could eventually give Recursion excellent economics. Today's platform has no stable gross-margin profile that can be extrapolated.

I see three defensible sources of competitive advantage.

First, Recursion has spent more than a decade building a proprietary wet-lab/data-generation stack. The asset that counts is the linkage between perturbation, phenotype, chemistry and biological context, not raw byte count. Competitors can buy public datasets and GPU compute; recreating years of standardized proprietary experimentation is harder. That Roche/Genentech, Bayer and Sanofi keep funding programs supports the proposition that the dataset and experimental engine have external scientific value.

Second, the Exscientia combination filled a real capability gap. Recursion's historic strength was phenotypic biology and target discovery. Exscientia's precision-chemistry platform gave the merged organization a better route from target to optimized molecule. REC-1245, REC-617 and REC-7735 now let investors test whether that integration produces differentiated molecules on clinically relevant timelines.

Third, established pharma relationships create repeat-play potential and some switching friction, because collaborators have already built joint maps, selected targets and spent scientist time. Genentech's move to a first previously unexplored neuroscience target in 2026 matters more than a new marketing partnership would: it is a partner carrying platform output forward into another stage of discovery.

The claimed AI efficiency remains a provisional moat. Recursion compares about 330 compounds synthesized per program and around 1.5 years to candidate selection against industry reference points of approximately 2,500 compounds and four years. REC-617 was selected after 136 new compounds in under 11 months; REC-1245 after 204 compounds in around 18 months; REC-7735 in roughly ten months. Those are falsifiable process metrics, which makes them better evidence than vague claims about “AI transforming pharma.”

Clinical productivity is still unproven. The BIO/Biomedtracker historical work cited by industry sources puts overall Phase-I-to-approval likelihood below 10%, with rare-disease programs generally doing better and oncology worse. A faster candidate-selection cycle creates economic value only if quality holds. Recursion cannot yet show an internally controlled dataset demonstrating superior Phase I, Phase II or approval success against matched industry programs. Five clinical-stage programs and zero approvals are too small a denominator.

REC-4881 shows the distinction especially clearly. Its human efficacy result gives the company a potentially valuable drug and supports the biological insight that MEK inhibition can alter FAP polyp burden. It does not establish that Recursion's de-novo chemistry engine produces better drugs, because REC-4881's development history differs from REC-1245 and REC-7735. For platform validation, the strongest future evidence would be repeated clinical success among molecules whose target discovery and chemistry were both materially generated inside the current operating system.

The industry itself is still in a validation stage. “AI drug discovery” covers software vendors, antibody discovery engines, computational chemistry companies, multi-omics platforms and clinical biotechs using machine learning internally. The profit pool sits downstream: a successful approved drug can produce billions of dollars, while discovery tools capture a smaller share unless the platform owner retains drug rights or earns meaningful royalties. This economic structure is why Recursion has progressively become more like a drug developer even while retaining its platform identity.

Financing cycles matter almost as much as scientific ones. Clinical biotech assets carry cash flows many years out and depend on repeated equity or partnership financing, which makes their capital-market values sensitive to risk appetite and discount rates. Recursion is its own case study: the IPO, subsequent placements, ATM issuance, 2024 offering and acquisition stock all funded a company whose cumulative operating cash flow stayed deeply negative.

Regulation is still conventional biopharma regulation. AI may shorten target identification or medicinal chemistry, but it does not relax the requirement to establish safety and efficacy in humans. That makes REC-4881's next regulatory discussion economically important: the market needs to learn whether FDA will accept a development path centered on endoscopic polyp-burden endpoints, whether a randomized study is required, how long treatment must continue and what safety burden is acceptable for a preventive or chronic-use setting. The 2025 10-K explicitly warns that regulators can require additional studies and lengthen development.

No public company is a clean comparator. Schrödinger is the best reference for a real software-plus-drug-discovery hybrid. AbCellera fits a partner-driven discovery platform moving toward owned drugs. Absci is a smaller generative-biologics analogue. Relay Therapeutics works from the other direction: it shows what the market will pay for computationally informed clinical oncology assets with no large platform-revenue narrative attached.

Market capitalizations use September 4 closes. Schrödinger reported $32.5 million of software revenue plus $23.0 million of drug-discovery revenue in Q2 and $418.8 million of cash, restricted cash and marketable securities at June 30. AbCellera reported $4.1 million Q2 revenue, a $55.4 million net loss and more than $565 million of cash/marketable securities, plus $110 million of available non-dilutive government funding.

Schrödinger built a real software business and attached drug-discovery optionality to it. Its customers can license physics-based computational tools without buying into Schrödinger's internal pipeline. That gives its revenue a dimension Recursion lacks: software demand can be judged independently of clinical milestones. Q2 2026 software revenue was still $32.5 million, even during a transition toward hosted licensing. Recursion has broader proprietary phenomics and wet-lab scale, but far less recurring commercial revenue.

AbCellera took a different path. Its original economics centered on antibody-discovery partnerships and downstream milestones/royalties; it has increasingly used the platform to build owned clinical assets. Its Q2 revenue was even lower than Recursion's, so a simple sales multiple would make AbCellera look absurdly expensive at a $3.49 billion market cap. What the market is valuing instead is liquidity, partner optionality, antibody infrastructure and the internal pipeline. Peer price-to-sales ratios applied to RXRX would create false precision for the same reason.

Absci, valued at about $1.51 billion on September 4, sits on the generative-protein side of AI drug discovery. Its competitive threat is conceptual: if generative biologics platforms can build compelling partner economics with less laboratory infrastructure, Recursion's broader stack may prove costly instead of advantaged in some modalities. Recursion is still focused primarily on small molecules and phenotypic biology, so the overlap is incomplete.

Relay Therapeutics, at about $4.17 billion, is the cleaner benchmark for the internal oncology pipeline. Investors value Relay mostly through individual clinical assets and precision-oncology data. Its higher market capitalization does not make RXRX cheap. It shows how fast a market can assign billions to a clinical program once human efficacy is convincing. Recursion has produced no comparable late-stage evidence yet.

Recursion's niche is broad-platform challenger plus clinical portfolio. Breadth is its advantage and its burden. One organization can generate biology, design molecules, run partner programs and increasingly support clinical enrollment. The same breadth helped create the expense base that had to be cut by roughly 40% versus 2024 pro-forma cash operating cost. The moat will strengthen if the post-restructuring organization keeps its cycle-time advantage while delivering multiple human efficacy signals. It weakens if cost cuts reduce discovery output, or if the current pipeline attrits at ordinary industry rates.

Current fundamentals, clinical evidence and price-move attribution

Q2 2026 worked as a stress test: weak platform revenue against a forceful operating reset.

Those figures come from the filing, which attributes lower Roche revenue to the completion of certain project phases in the comparison period.

External consensus services differed slightly on expected EPS, around a loss of $0.23–$0.24, against a reported $0.25 loss. Revenue of $7.67 million came in roughly a third below street expectations around $12 million. What matters is direction, not false precision around a one-cent consensus difference: partnership revenue materially missed, operating expenditure fell faster than expected, and the balance sheet stayed adequate.

H1 makes the reset clearer. Total revenue fell to $14.14 million from $33.97 million. R&D fell to $177.5 million from $258.3 million, G&A to $76.1 million from $101.3 million, and the H1 net loss narrowed to $248.5 million from $374.4 million. Across the half the company spent $46.8 million less on Tempus records in platform R&D, while clinical investment rose. So the expense reduction mixes durable headcount/site savings with less external data acquisition; investors should not assume every dollar of savings is permanent and free of any effect on platform throughput.

One more useful data point sits in the quarter-end balance sheet: $140.4 million of current and non-current unearned revenue. That liability is cash received for obligations not yet fully recognized as revenue. It is evidence that accounting revenue understates some contracted/cashed partner activity in a given quarter, though it is not the same thing as future profit or remaining contract value.

Two contrary platform signals landed in the same quarter. Accounting monetization was poor. Commercial and scientific continuity was intact: Genentech had selected another target, Sanofi had delivered a 2026 milestone and Bayer's agreement remained in force under the terms disclosed in the annual report. That is why I read the Q2 revenue collapse as evidence of lumpy economics, not evidence that the partnership franchise is falling apart.

REC-4881 is the largest single internal valuation component in my model. The November 25, 2025 cutoff showed a 43% median reduction in total polyp burden at week 13 among 12 evaluable patients. Among 11 evaluated at week 25 after the treatment period, the median reduction was 53%; nine of eleven kept reducing and eight had at least a 30% durable reduction. The August materials added median reductions in both upper and lower gastrointestinal disease.

Durability is the intriguing part, because it raises the possibility that a finite period of pathway inhibition can produce persistent benefit. The sample is far too small to assume that result survives a larger controlled study. Regression to the mean, endoscopic measurement variability and cohort selection matter more with a dozen patients than they will in a registrational population. The upcoming cohort and the FDA discussion should move valuation on two axes at once: efficacy confidence and required development cost.

Recursion's market-opportunity framing needs a haircut. It has cited an addressable opportunity above $10 billion, built from roughly 50,000 diagnosed FAP patients and rare-disease oral-drug price benchmarks around $260,000–$360,000 annually. That calculation is a theoretical multiplication of prevalence and benchmark price. It does not establish that every diagnosed FAP patient would be eligible for chronic MEK inhibition, tolerate therapy, receive the drug or support that pricing. My base case assumes $1.25 billion of global peak sales, roughly an order of magnitude below the headline TAM.

REC-1245 is still an early safety/PK experiment. As of March 31, 2026, 16 patients had been treated. Recursion reported no dose-limiting toxicities, no Grade 4/5 treatment-related events and predictable dose-dependent exposure. The limitation that counts: exposure at the reported dose levels stayed below the range associated with tumor regression in the preclinical models. The next update needs to show that clinically tolerable dosing reaches pharmacologically meaningful exposure; a response before or near that level would be a bonus.

REC-617 has more human experience but modest efficacy evidence. The 2025 annual report describes 29 heavily pretreated patients across monotherapy dose escalation, one confirmed durable partial response in an ovarian-cancer patient and five more patients with durable stable disease. The maximum tolerated dose was set at 10 mg once daily. Enough to keep a CDK7 asset alive; nowhere near enough to establish best-in-class status in a competitive oncology mechanism.

REC-7735 may become the cleaner platform proof. Recursion says the molecule was designed in around ten months, has greater than 100-fold selectivity for PI3Kα H1047R over wild type and produced preclinical antitumor activity without the hyperglycemia signal expected from broad PI3Kα inhibition. The IND has been cleared. Human therapeutic index is now the first question: can a dose with adequate mutant inhibition be reached without recreating class toxicities?

Since the previous report, the price path is short enough to separate events from narratives.

Historical pricing sources place the June 8 close at $3.32 and the September 4 close at $3.63; the Q2 session finished down after the August 5 release.

The June 17 annual meeting removed Gibson from the board, and the April proxy process had already signaled it. I assign little of the stock's subsequent move to a surprise governance change. The CEO succession had been effective since January. NVIDIA's exit had been public since February. Both still shaped the narrative discount attached to RXRX, but by June 8 they were background variables.

August 5 is cleaner. The revenue miss hit the immediate platform narrative. The expense reduction and runway guidance supported the balance-sheet narrative. A stock that sold off and then recovered over the following month fits investors weighting forward clinical catalysts and financing runway above a single milestone-revenue quarter. That is my inference from the timing, not something management can confirm.

The REC-4881 Phase II result cannot explain that recovery as new information, because its main efficacy numbers had been disclosed eight months earlier. The November 2 additional readout and the FDA interaction can explain anticipatory positioning. That is the key distinction between “de-risked by data” and “re-rated by sentiment”: the first requires new evidence about probability of technical/regulatory success; the second can happen because investors become more willing to pay for evidence already known, or for a coming catalyst.

Broad biotech conditions and rates are legitimate background variables for a loss-making company with cash flows many years away, though I do not assign a precise fraction of the June-to-September RXRX move to them. The September 4 XBI level was 163.81, and RXRX's reported beta and very high short interest let the stock amplify sector moves. MarketWatch reported short interest around 178 million shares, roughly 37% of float, as of mid-August. That can turn ordinary risk-on flows or catalyst positioning into outsized share moves without changing rNPV.

I found no disclosed RXRX-specific index addition or deletion in the materials used here that would justify treating index flows as a primary post-June catalyst. There were also no disclosed sales under the new $300 million ATM through June 30. Post-quarter issuance, if any, becomes fully visible only in later filings unless separately disclosed. Without evidence, the price path should not be narrated as a financing-driven selloff or as a new-clinical-data rally.

Bull/bear disagreement can be put more sharply than “AI works versus AI does not work.” Bulls are betting that the cost reset buys enough time for REC-4881 to establish a plausible registrational route, that one or more earlier oncology programs produce human proof, and that partner programs keep converting into cash milestones. Bears are betting that ordinary biotech attrition will overwhelm the discovery-speed advantage, leaving a company whose impressive infrastructure must still be financed by shareholders.

My reading of the evidence is that Q2 improved solvency much more than it improved business quality. The revenue miss does not negate the partnership franchise, yet the company still has no repeatable margin and cash-generation model. REC-4881 raises the value of one asset; the broader thesis has to be proved by the next generation of Recursion-designed molecules.

Valuation, margin of safety, risks and catalysts

A P/E, EBITDA or conventional DCF framework would misvalue Recursion. Earnings and free cash flow are negative, product revenue is zero, collaboration revenue is episodic, and the most important assets carry contingent future cash flows. So my primary method combines program-level risk-adjusted NPV, a separately valued platform/partnership option, and current net liquidity. Peer market capitalizations are a reasonableness check, not the valuation engine.

The cash-flow passthrough test makes that choice unavoidable. Over 2021–2025, operating cash flow totaled approximately negative $1.26 billion against cumulative net losses of about $1.86 billion, an OCF/net-loss magnitude ratio around 0.68. The gap comes from stock compensation, depreciation/amortization, partnership-cash timing and other non-cash accounting items. It is not healthy earnings conversion.

Recursion does not separately disclose maintenance versus growth capex. It states that ordinary repairs and maintenance are expensed, while PPE purchases have included laboratory equipment, computing upgrades and other infrastructure. PPE purchases fell from almost $40 million in 2021 to $6.5 million in 2025. Treat all 2025 PPE purchases as a conservative proxy for maintenance capex and 2025 owner cash earnings land near negative $378 million, against negative $372 million of operating cash flow. Either way, P/E and FCF yield are economically meaningless.

Cash here is a runway asset, not a permanent liquidation floor. The $556.8 million June balance is being spent to create the clinical data embedded in the rNPVs below. Credit the full cash balance while ignoring future R&D and the same value gets counted twice. My program models subtract expected development costs, and a separate central-cost reserve covers corporate/platform expenditure that cannot be assigned cleanly to one asset.

For probability assumptions I start near historical biotechnology base rates and move only where program-specific evidence warrants it. Industry analyses have historically put overall Phase-I-to-approval probability below 10%, with oncology below the all-industry average and rare-disease programs higher. REC-4881 gets a 35% base probability on the strength of a human Phase II efficacy signal in a rare disease; the early oncology programs sit around high-single-digit to 10% probabilities. These are my assumptions, not company guidance.

Those rNPVs use a 12% nominal discount rate, commercial contribution margins in the high-50% range before tax, staged future development costs, launch ramps and finite post-launch economics. They come out deliberately much smaller than company TAMs multiplied by probability of success. The resulting base pipeline value is about $0.82 billion. Small changes to timing, price, treated population and probability move these values materially.

REC-4881 carries most of the pipeline rNPV because it is the only program with a reasonably interpretable human efficacy signal. My $1.25 billion peak-sales assumption is a fraction of Recursion's greater-than-$10-billion addressable-market framing. A path to $2–3 billion of peak sales would need convincing durability, manageable chronic-use toxicity, broad treatment eligibility and strong pricing. Failing to secure a practical registrational endpoint could cut the asset's value even if the biological effect is real.

REC-1245 gets a large theoretical peak-sales number and a very low current rNPV, because human efficacy is unknown. The program would revalue sharply if dose escalation reaches target exposure and produces objective responses with manageable toxicity. REC-7735 sits in the same position: a mutant-selective PI3Kα medicine could address a large oncology population, but preclinical selectivity is no substitute for human therapeutic index.

I give the platform a separate option value, because folding partner economics into the internal assets would miss the Roche/Genentech, Sanofi and Bayer programs. I value that option off expected future milestone, option and royalty cash flows, heavily discounted because most contractual maxima require multiple successful events in sequence. The base platform/partnership value is $1.28 billion; conservative and optimistic cases use $0.75 billion and $2.15 billion. Those figures are valuation assumptions, not booked backlog. The factual anchor is the set of active collaborations and achieved milestones.

The scenario analysis is:

This is valuation-scenario analysis within a research framework, not investment advice.

The share denominators deliberately exceed today's 535.34 million shares. A pre-revenue company with a $300 million unused ATM and an early-2028 runway should be valued on probable future ownership, not on today's basic share count alone. The conservative case assumes roughly 9% additional dilution from the July share count, the base about 6%, the optimistic case roughly 5%. Drawing the ATM in full at today's price would be far more dilutive than any of those.

Peer valuation does not justify a higher number by itself. RXRX's $1.94 billion market cap sits above Schrödinger's $1.52 billion, despite Schrödinger generating materially more Q2 revenue, and below AbCellera's $3.49 billion and Relay's $4.17 billion. Those gaps mostly reflect differences in pipeline assets, commercial models and investor expectations, not stable multiples.

The expectation gap concentrates in four events. The market needs evidence that REC-4881's effect survives a larger dataset; clarity from FDA on a realistic pivotal path; evidence that REC-1245 can reach active exposure; and confirmation that partner milestones keep arriving even while quarterly revenue stays lumpy. A fifth event, REC-7735's first human PK/safety profile, comes later and may overlap the edge of the current cash runway.

The independent margin-of-safety check is harsher than the base case. The current price of $3.63 stands about 30% above the $2.80 conservative intrinsic value, so the conservative scenario offers zero discount. The most fragile base assumption is the $1.28 billion platform option value. Cut it to 70% and base value falls by roughly $0.67 per share, from $4.40 to about $3.73, almost exactly the current price. The market is already paying for meaningful partner conversion before any approved drug exists.

The prescribed “flat earnings for three years” test is especially unfavorable. Recursion has negative earnings and negative owner cash flow; keeping those flat would consume capital rather than compound it. There is no positive earnings yield to set against a positive government-bond yield. On that test, there is no margin of safety at this buy price.

Margin-of-safety sufficiency verdict: none.

The first permanent-loss risk is REC-4881 clinical/regulatory disappointment. I assess probability as medium and impact as high. The observable indicators are November efficacy breadth, adverse-event/discontinuation rates and the FDA development-path update. A requirement for a large, long-duration randomized trial would push launch later, raise development cost and lower rNPV even if the drug keeps reducing polyps. A materially weaker larger-cohort response would attack both the PoS and the peak-sales assumptions.

The second is financing dilution. Probability is high on a 3–5-year horizon, and so is impact. Early 2028 sits comfortably beyond the 2026 catalysts but nowhere near commercialization. The unused $300 million ATM lets management issue stock well before cash gets scarce. Watch quarterly cash, share count, ATM disclosures and partnership cash inflows. If cash falls below roughly $350–400 million without a meaningful clinical re-rating, the probability of preemptive equity issuance rises.

The third is platform monetization shortfall. Probability medium to high; impact high, because $1.28 billion of my base equity value comes from partner optionality. The things to watch are new target selections, development-candidate milestones, option exercises, royalty-bearing program progression and gross contribution from collaborative research. If the Roche/Genentech, Bayer and Sanofi relationships stay scientifically active but produce no recurring option/milestone cash, the market will eventually stop paying a platform premium.

The fourth is ordinary oncology attrition. Probability is high at the individual-program level and medium-to-high for several simultaneous failures. REC-1245, REC-617, REC-3565, REC-4539 and REC-7735 give the portfolio diversification, but early-stage oncology historically has low approval rates. Failure across two or three would not threaten near-term solvency. It would undermine the claim that Recursion's discovery system improves molecule quality.

The fifth risk is sentiment interacting with financing. It has medium probability and medium direct fundamental impact, but can become high impact if the company must issue equity during a sector drawdown. The unusually high short interest can amplify both rallies and selloffs. A 30–40% stock decline by itself destroys no scientific value; issuing tens of millions of shares after such a decline permanently transfers a larger fraction of future program economics away from existing holders.

Here is the tracking dashboard I would use:

The November 4 earnings date is an external estimate, not a company-announced date, as of the information reviewed here; the company-specific clinical catalyst on November 2 is firmer.

Financial tracking belongs on cash and share count, not on quarterly revenue “beats.” A $30 million milestone can create a spectacular revenue growth rate while changing total pipeline probability very little. For clinical tracking, the variables are sample size, effect consistency, dose/exposure and safety, not management adjectives. Partnership tracking should follow the irreversible economic steps: target acceptance, candidate selection, option exercise and cash receipt.

Cross-synthesis, research conclusion, uncertainties and source set

Looking vertically, Recursion has proven one capability beyond reasonable dispute: it can build and operate a large-scale experimental/computational drug-discovery system that sophisticated counterparties are willing to engage and fund. Bayer partnered before the IPO; Roche/Genentech later committed a major upfront payment and kept accepting maps/targets; Exscientia was integrated to fill the chemistry gap; Sanofi programs kept producing milestones after the acquisition. A company with no useful technology would be unlikely to sustain that sequence across more than a decade.

Its capital-market success came from a mixture of genuine technical capability and an era that funded long-duration scientific narratives richly. The 2021 IPO provided nearly half a billion dollars. Strategic partnerships brought more. NVIDIA's 2023 investment attached the name of the most valuable AI hardware company to Recursion at exactly the moment generative AI became a dominant market theme. Equity markets supplied the rest, repeatedly. That capital let Recursion build infrastructure years before any product revenue existed.

That tailwind faded before the science could mature. By 2025 the combined Recursion/Exscientia organization was producing a $645 million annual loss, spending hundreds of millions in operating cash and asking the market to finance a broad platform plus many clinical options. The 20% workforce reduction and the 2026 expense reset were rational responses. They also exposed the contradiction at the center of the old operating model: discovery scale has value only if the organization can afford to move the best discoveries through the far more expensive clinical funnel.

Khan's version of Recursion is beginning to look more investable operationally because it is narrower. Platform R&D is down, clinical R&D is protected, site and headcount costs have fallen, and the runway reaches a cluster of catalysts. The tradeoff is that scientific productivity now has to be maintained with fewer resources. The next two years will test whether the 2024 peak expense base was wasteful breadth or necessary investment behind the platform's claimed advantage.

Horizontally, integration is Recursion's sharpest distinction. Schrödinger has a more mature standalone software business. AbCellera has deep antibody-engineering infrastructure and a formidable liquidity position. Absci works more narrowly on generative biologics. Relay lets investors buy computational precision-oncology assets without underwriting a large discovery-platform overhead structure. Recursion offers a longer chain, from phenotypic biology through chemistry and increasingly into clinical development. That integration can generate uniquely informed programs; it can also make cost discipline harder.

The biggest market misjudgment may run in both directions. Bears can understate the partner franchise by treating a weak quarterly revenue number as though partners were leaving. Q2's decline partly reflected project-phase completion, and Sanofi and Genentech kept advancing work. Bulls can overstate validation by reading every partner milestone or discovery-speed statistic as proof that clinical success rates will be higher. The truth currently sits at different levels of the funnel: commercial willingness to use the platform is proven; superior clinical productivity is not.

The same distinction applies to REC-4881. It is a genuine de-risking event relative to the pre-human stage. A 43% median polyp reduction across 12 patients, with persistence in the follow-up set, is better evidence than target-validation slides. The evidence was already public in December 2025. So it changes intrinsic pipeline value relative to a report written before December 2025, but it should already have been available to a June 8, 2026 analysis. This re-research should not pretend August 2026 created that evidence.

What August did change was the financing and catalyst map. Below-$375-million 2026 cash-operating-expense guidance, combined cash/restricted cash of $556.8 million and a stated early-2028 runway materially lower the probability of a forced near-term financing. The clinical calendar is more concentrated now. That is a real improvement in security design: existing shareholders have a better chance of seeing 2026–27 data before a financing becomes unavoidable.

The runway still does not cover the full proof period. REC-7735, one of the most important de-novo tests of the integrated AI-chemistry stack, may only start delivering meaningful human PK/safety evidence near the edge of it. Any pivotal REC-4881 program, later REC-1245 expansion or subsequent registrational program runs far beyond early 2028. Over a 3–5-year horizon, dilution belongs in the base case.

The capital-market security has changed less than the operating organization. RXRX is still a portfolio of highly convex outcomes funded by a melting cash balance. Positive data can make rNPV jump faster than cash is consumed; negative data can leave the same operating infrastructure supported by fewer valuable programs. The $300 million ATM is in effect a call option for management on future stock-market enthusiasm. Existing shareholders pay for that flexibility through possible dilution.

Over the next 12 months, REC-4881 dominates. The questions are size-adjusted consistency of polyp reduction, safety, FDA alignment and whether the market can tell a potentially registrational path from another exploratory Phase II cohort. REC-1245 exposure/response data and REC-7735 trial initiation are secondary, though potentially important.

At three years, the central variable changes. Recursion needs evidence that the operating system can repeatedly create clinically viable molecules. One approved or convincingly pivotal asset would transform how the market treats every earlier program, because the platform would then have a human-success precedent. Repeated early oncology failures would do the opposite: discovery speed would be reclassified as an R&D efficiency metric, not a moat.

At five years, the company needs either commercial product economics or a partnership royalty stream with visible probability of commercialization. A business still leaning primarily on episodic discovery milestones in 2031 would deserve a platform-services valuation plus net cash, not a pharmaceutical platform premium. The cost base would then be judged against partner gross profit rather than theoretical pipeline breadth.

The independent re-research therefore lands in an interesting place relative to the June prior report. I independently reach a similar broad fair-value neighborhood, but for different reasons, and with a different reading of the “new” facts. The cost reset improves near-term financial survival. The main REC-4881 efficacy evidence predates the prior report and should not be counted as new de-risking since June. Q2's revenue collapse sharpens the criticism of current platform economics, while Genentech/Sanofi progression stops me from concluding that the partner franchise is deteriorating.

My base intrinsic value of $4.40 sits above the current $3.63, but roughly 21% of point-estimate upside is not adequate compensation on its own for the dispersion between $2.80 and $7.82. A clinical-stage biotechnology stock with an unused $300 million ATM, no product revenue and several early oncology assets can cross that whole range on a handful of datasets. The conservative case sits materially below today's price. Base-case upside notwithstanding, that is why the valuation earns no margin-of-safety verdict.

The earlier report's $2.80 “ideal buy” reference is materially more generous than my independently derived purchase discipline. Under the explicit rule in this assignment that a buy price must sit at least 20% below conservative intrinsic value, my $2.80 conservative value puts the upper buy threshold around $2.24. The difference comes from prospective dilution, and from separating the platform's theoretical contractual maxima from risk-adjusted realizable economics. I am not treating a low share price as cheap.

The central thesis is now stronger clinically and safer financially, while remaining unproven economically. The cost reset buys time. REC-4881 provides a legitimate clinical asset. Roche/Genentech, Sanofi and Bayer give the platform external validation. Yet investors still fund virtually the entire gap between partner revenue and the expense required to discover and develop medicines.

Core bull reasons:

• REC-4881 produced a 43% median week-13 polyp-burden reduction in the disclosed Phase II cohort with continued median improvement at week 25, creating a credible rare-disease asset where Recursion frames the market as lacking approved pharmacotherapy.
• FY2026 cash operating expense guidance below $375 million and $556.8 million of combined cash/restricted cash extend runway into early 2028, covering several near-term clinical and regulatory catalysts.
• Roche/Genentech, Sanofi and Bayer continue to provide economic/scientific third-party validation, including a fifth Sanofi milestone in February and ongoing Genentech target progression.
• REC-1245, REC-617 and REC-7735 give investors multiple tests of whether integrated AI-guided chemistry can produce differentiated human-stage drugs, instead of one flagship binary outcome.

Core bear reasons:

• Q2 operating revenue of $7.30 million was below $11.49 million cost of revenue before almost $131 million of additional R&D/G&A, so present platform economics do not finance the organization.
• The share base has expanded dramatically through acquisitions and financing, and a fully unused $300 million ATM remains available; at $3.63, full use would represent roughly 83 million new shares before fees.
• REC-4881's efficacy dataset remains small and uncontrolled, while the value of the asset depends heavily on FDA accepting a feasible registration path and chronic-use safety.
• The company's AI speed metrics demonstrate process efficiency but still lack the clinical outcomes needed to establish a higher approval rate than conventional biotech.
• REC-1245 and REC-7735 remain too early to offset a REC-4881 failure today; one had not yet reached preclinical tumor-regression exposure in patients and the other had no human data at the base date.

The first pre-mortem script is clinical plus financing. Suppose the November 2026 REC-4881 update shows a weaker effect in the expanded cohort, and FDA requires a larger randomized, longer-duration study with a harder endpoint. My REC-4881 rNPV could fall from roughly $580 million toward $200–250 million. If REC-1245 then reaches pharmacologically active exposure in 2027 without objective responses, and REC-7735 human dosing slips, the internal pipeline discount widens just as cash moves toward $300 million. Management could reasonably sell $150–200 million of ATM stock near $2.25–$2.75. Issuing 60–80 million shares at those prices, with platform option value falling below $700 million, gives a credible path to a $1.5–$2.0 stock, around 45–60% below today's price.

The second script attacks the platform. Imagine that through 2027 the Roche/Genentech maps continue scientifically but produce no meaningful license option, Bayer does not exercise major candidate options and Sanofi milestones slow. Platform revenue stays episodic while collaborative cost of revenue periodically exceeds recognized revenue. Schrödinger, AbCellera, Absci and large pharma's own internal AI groups keep advancing candidates, which makes “AI-enabled discovery” less scarce as a capital-market category. My $1.28 billion base platform option would then compress toward $400–500 million. Even if REC-4881 survives, the equity could lose around half its value, because the market would price Recursion as a conventional multi-asset biotech instead of a premium discovery platform.

The research conclusion follows from those two scripts. Recursion today is a more disciplined company than the organization that entered 2025. Management has taken out real cost, protected clinical work, extended runway and given investors a small but meaningful Phase II asset around which to value the internal portfolio. The balance sheet supports the next wave of decisions instead of forcing one now.

At $3.63, the security still asks the shareholder to pay in advance for successful conversion of some platform optionality. My base value is $4.40. The conservative value is only $2.80, and the platform assumption on its own can pull base value to around $3.73 if cut by 30%. That leaves very little valuation cushion before ordinary biotech uncertainty enters the picture. I would turn more constructive after FDA alignment on REC-4881 plus larger-cohort efficacy, or at a price that compensates for those unknowns.

【Company-profile scores】

• Fundamental quality: medium
• Growth: high
• Moat: medium
• Financial soundness: medium
• Management credibility: medium
• Valuation attractiveness: medium
• Risk level: high
• Suitable investor type: event-driven / high-risk speculation

【Investment rating】

• Rating: Watch
• One-line thesis: Cost cuts fund the next catalysts, but current valuation still prices meaningful partner and clinical success before either is securely established.
• Ideal buy price: see dedicated line below.
• Acceptable hold price: 3.75–5.05 USD
• Clearly overvalued price: 8.60–10.00 USD
• Current-price classification: outside the three bands
• Whether to wait for a better price: yes. A price at or below $2.20 provides the required discount to conservative value; alternatively, FDA alignment plus larger-cohort REC-4881 confirmation could justify paying more. The opportunity cost is missing a catalyst-driven re-rating before those conditions appear.
• Target holding horizon: 3–5 years for fundamental ownership; 6–18 months for event-driven exposure.
• Expected annualized return: conservative approximately -8.3%, base +6.6%, optimistic +29.2%, assuming three-year convergence toward the scenario values and no dividend.
• Max-loss risk: roughly 50–70% in a combined REC-4881/regulatory disappointment, weak early-oncology data and low-price financing scenario.
• Reassessment-trigger signals: REC-4881 expanded-cohort median polyp reduction below 25%; FDA requiring a materially longer or larger pivotal program than assumed; REC-1245 unable to reach active exposure before dose-limiting toxicity; cash below $400 million without offsetting milestones; Class A shares above 575 million before a major clinical de-risking event.

【Ideal Buy Price】1.90–2.20 USD

Basis: the range is at least 20% below the $2.80 conservative intrinsic-value estimate, incorporating prospective dilution and a substantial haircut to platform and pipeline optionality.

【Valuation Range】

• current: 3.63 (close as of 2026-09-04)
• bear (conservative · ideal buy zone): [1.90, 2.20]
• base (fair · acceptable hold zone): [3.75, 5.05]
• bull (optimistic · above the clearly-overvalued line): [8.60, 10.00]

The current price sits in the gap between the ideal-buy and acceptable-hold bands. That gap is deliberate. At $3.63, base-case upside exists, but the conservative-case discount required for a new high-risk position is absent.

Research uncertainties remain material. First, partner agreements disclose enormous nominal milestone ceilings but too few program-by-program probabilities to value them tightly. Second, FDA has not yet provided the public registrational clarity needed to estimate REC-4881 pivotal cost, duration and endpoint risk with confidence. Third, institutional ownership data lag; the April proxy is the best company-compiled 5% register located, not a September real-time one. Fourth, ATM sales after June 30 cannot be ruled out until subsequent disclosures. Fifth, Recursion's AI productivity comparisons use company-selected process metrics, and matched external evidence on clinical success rates remains unavailable.

The primary source set underlying this report is Recursion's 2025 Form 10-K, Q1 and Q2 2026 Forms 10-Q, August 2026 corporate/earnings materials, the 2026 proxy and annual-meeting filing, the December 2025 REC-4881 release, the IPO prospectus and collaboration disclosures. Market pricing is checked to September 4, 2026. Peer numbers use current market data and the latest cited SEC filings. Secondary sources are used mainly for market reactions, consensus and NVIDIA's disclosed portfolio event where useful; conflicting consensus estimates were not treated as primary facts.

Other tickers mentioned

NVDA.US: former strategic shareholder whose complete 2025 stake exit removed an important capital-market endorsement while leaving the scientific merits to stand independently.

SDGR.US: closest listed hybrid reference for combining commercial computational-drug-discovery software with an internal/partnered drug pipeline.

ABCL.US: partner-driven discovery-platform comparator with substantial liquidity and increasing internal clinical exposure.

ABSI.US: generative-biologics peer illustrating a narrower AI-native discovery model than Recursion's phenomics-plus-chemistry stack.

RLAY.US: computational precision-oncology comparator useful for valuing Recursion's internal clinical assets separately from its platform narrative.

TEM.US: data supplier whose lower usage materially contributed to Recursion's 2026 platform-R&D cost reduction.

ROG.SW: largest platform partner, through Roche and Genentech, whose completed project phases drove most of the Q2 revenue decline and whose first Validated Target Option is the clearest near-term test of partner conversion.

SAN.PA: large-pharma collaboration counterparty whose inherited Exscientia programs have continued producing progress milestones.

BAYN.XETRA: earliest major external validator, whose 2020 fibrosis collaboration was later reoriented toward oncology, showing that a platform partnership can survive a change of therapeutic priority.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.