Quick ReadPlain-language overview · read this first
Recursion Pharmaceuticals is a U.S. company that wants to use artificial intelligence to develop new drugs. The report's stance is “Watch”: put it on the list and monitor it, but do not act for now.
It does two things. On one side, it rents out its AI capabilities to large drugmakers, helping them look for new drug directions and collecting collaboration fees. On the other, it develops drugs itself, betting that one of them can succeed. The difficulty is that, so far, it has not had a single drug approved for market, and all of its revenue comes from scattered collaboration payments from large drugmakers. This money is uneven, unpredictable, and shrinking, which is completely different from the steady inflow of selling products. The report sums up its essence in one sentence: it is burning cash to buy the possibility that “maybe this works someday”; it has not been proved, and it has not been disproved.
How aggressively is it burning cash? It lost about $560 million this year. It still has about $660 million in cash, but it burns through almost half of that each year, meaning its financial base shrinks by half annually. By 2027, it will probably need to issue more shares to keep going, which would further dilute existing shareholders. What makes investors even more uneasy is that Nvidia, the earlier biggest marquee shareholder, has already sold out and left, while the founder has also stepped down.
So is the current price cheap? The share price is $3.315, down about 90% from its former peak. It is genuinely cheap, but the report says that is because the risks are real, not because the asset is clearly good. There is still about $1.25 of cash per share on the balance sheet, but this floor is also melting year by year. The report's ideal buy price is below $2.80. The current price is still above that, so the stance remains Watch, not chase.
The above only explains this report in plain language and is not investment advice. The stock market involves risk; invest cautiously.
LeadRecursion Pharmaceuticals (RXRX.US) is the flagship AI drug-discovery name: the largest and most broadly integrated AI-native drug company globally, with Exscientia added in 2024 to fill the precision-chemistry gap. The core thesis is a hybrid of an AI drug-discovery platform and an in-house biotech pipeline, backed by deep partnerships with NVIDIA, Roche/Genentech, Bayer, Sanofi, and others, yet still dependent on unpredictable collaboration milestones with no product revenue. Report rating Watch: real infrastructure assets and a heavily reset valuation deserve close tracking, but NVIDIA exited its stake by the end of 2025 and founder CEO Chris Gibson has stepped down while the platform remains clinically unproven.
Prices in the article are as of publication; see the valuation band above for the live price.
Research Perspective Statement
Company: Recursion Pharmaceuticals (RXRX.US, NASDAQ), based in Salt Lake City, Utah. It is a hybrid of an AI drug-discovery platform and an in-house clinical pipeline, founded in 2013, listed in an IPO in 2021-04, and completed its merger with UK AI drug company Exscientia in 2024-11. By scope across data, compute, pipeline count, and partners, it is the largest and most broadly integrated flagship among AI-native drug companies.
Entity clarification to avoid confusion: ① RXRX follows a hybrid model: it uses its platform externally to enable large pharmaceutical partners through platform collaborations, while also developing its own biotech pipeline. It is neither a pure software-tool vendor nor yet a drug company with products. ② It differs from other AI pharma/biotech names: Schrödinger (SDGR, physics-based computational software with a profitable software business), Relay (RLAY, protein dynamic-conformation targeting), AbCellera (ABCL, antibody platform), Absci (ABSI, generative antibodies), Certara (CERT, biosimulation software, profitable), Simulations Plus (SLP, modeling software), and XtalPi (2228.HK, quantum physics + AI + robotics). Do not conflate them. ③ Key personnel and shareholder changes have occurred and must be treated as current: founder Chris Gibson stepped down as CEO on 2026-01-01. Former Johnson & Johnson/Janssen chief data science officer Najat Khan became CEO. Gibson moved to chairman and will leave the board in 2026-06. The company stated this was not due to disagreements and that he would remain a strategic adviser. Its top strategic shareholder, NVIDIA, had fully exited all RXRX holdings by the end of 2025. ④ Fiscal year = calendar year, ending 12/31.
Business model in one sentence: This is a binary business of burning cash to buy platform optionality. Today, 100% of revenue comes from lumpy, unpredictable, and shrinking collaboration milestones from large pharma, with no product revenue at all. Its entire value is anchored to the assumption that the AI platform will eventually produce approvable blockbuster in-house drugs, a hypothesis neither falsified nor proven.
Currency: Share price, market cap, valuation, and financials are all in USD. The company reports in USD and trades on NASDAQ. Hong Kong/China comparables such as XtalPi (2228.HK) are denominated in HKD and marked separately.
Price anchor: This relative valuation uses the 2026-06-05 Friday close of 3.315 USD as the baseline, after a one-day -12.76% move from a prior close of roughly 3.80. Market cap was about $1.76 billion, shares outstanding about 531 million, and the 52-week range was 2.77 to 7.18 USD. The current price is about -54% from the 52-week high, about +20% from the 52-week low, and about -92% from the 2021 peak near $42.81. No PE due to persistent losses, TTM net loss about $559.78M, TTM EPS about -$1.17, FY2025 full-year EPS -$1.44, and no dividend. Because the EODHD daily API quota had been exhausted, the price was precisely cross-checked through multiple stockanalysis sources and independently red-teamed. Internal check: 3.315 x 531 million shares is approximately $1.76 billion.
Data basis and key cautions, read first: Financials use the company's primary filings as the base, with FY2025 10-K and earnings 8-K, plus Q1 2026 10-Q and earnings 8-K as the core primary sources. They were checked item by item and independently red-teamed against primary SEC filings and authoritative media. ① Two EPS bases: TTM EPS of -$1.17, corresponding to TTM net loss of $559.78M and about 478 million TTM weighted shares, versus FY2025 full-year EPS of -$1.44, corresponding to net loss of $644.76M and about 448 million weighted shares. This report distinguishes them explicitly and does not collapse them into one number. ② REC-4881 is an in-licensed old Takeda drug, not an AI de novo design. See below. This is key evidence that the platform has not proven itself. ③ Cause of the 6-05 selloff: the -12.76% move that day was driven overwhelmingly by a broad high-beta growth/AI-biotech sector selloff plus semiconductor macro weakness, with AI-biotech peers falling in tandem. It was also accompanied by small insider sales that day: CEO Khan sold 23,588 shares and Gibson sold 40,000 shares at about $3.62. This report does not describe the move as purely sector-driven with no company-specific news. ④ Net cash is a melting floor: the cash balance is large, but the company burns about half of it each year, so the bear-case floor moves down with cash burn and dilution. ⑤ Valuation method: for a binary pre-revenue biotech, any point estimate of fair value and narrow scenario band are fundamentally option/scenario valuation, not DCF intrinsic value. This report states that explicitly.
1. Conclusion First
One sentence: Recursion Pharmaceuticals is the flagship AI drug-discovery name: the largest and most broadly integrated AI-native drug company, with real infrastructure assets including automated wet labs, tens of PB of proprietary data, and supercomputing capacity, more than $500 million of blue-chip non-dilutive validation from large pharma, clean governance, and a balance sheet with material net cash. But it is a binary option with a narrow-to-mid moat (2.5/5) and an unproven clinical platform: 0 AI-discovered approved drugs, the leading asset is actually an in-licensed old Takeda drug, the only pure AI flagship has been discontinued, revenue is 100% dependent on shrinking lumpy milestones, cash burn is structurally heavy and the "net-cash floor" is melting, 2027 dilution is likely, and even the top endorser NVIDIA and founder CEO have exited. At $3.315, the stock is already down about 92% from its peak, below the prior bear-market trough, and far below sell-side consensus, so risk has been substantially flushed out. Rating: Watch, at the cautious lower edge and one step away from Avoid. Ideal buy price <= $2.80. Watch is not a weak Buy.
Four layers of logic:
Business quality: a narrow-to-mid moat (2.5/5), with a necessary separation between real assets and an unfalsified narrative. The input side is real: proprietary data, automated wet labs, and paid validation from blue-chip partners all have real replication barriers. The output side, meaning data flywheel to higher clinical success rates to approved drugs to excess returns, is still mainly a narrative. Three facts weaken it: the REC-994 flagship was discontinued; the leading REC-4881 asset is an in-licensed old drug rather than de novo; and industry-level data show Phase II success rates for AI drugs are roughly in line with traditional drugs. This is the core reason for Watch instead of Hold: the platform has not proven itself.
Why not Avoid: it is completely different from Canaan. Canaan is a case of bad business + value destruction + super-voting control + persistent dilution of outside shareholders + overextended pricing. RXRX has real assets, blue-chip cash validation, clean governance with no dual-class structure, a professional CEO, Gates Foundation ownership, material net cash, and a price that has already been heavily reset. Its quality is far above pure value-destruction names, so Avoid is not appropriate.
Why not Buy or Hold: the margin of safety does not come from cheapness; it requires platform validation. Zero approved drugs, an unproven platform, binary clinical risk, structural cash burn, likely dilution, and declining revenue mean cheapness may be a value option, or it may be a value trap with melting ice underneath. Direction is unresolved. This is a binary option, not a quality asset with a margin of safety.
Inflection point: 2026 is a decisive year of falsification or validation. REC-1245, an RBM39 degrader, has Phase 1 data due in 1H26, and REC-4881 has FDA registration-pathway discussions in 1H26. The platform will soon be falsified or validated. With NVIDIA exiting, the founder stepping back, and sentiment volatility in AI pharma, as shown by the 6-05 high-beta selloff, RXRX is in a critical post-deglamorization, early-recovery, show-me window.
Rating: Watch, cautious lower edge. This differs from Buy/Hold, which requires either margin of safety or a validated good business, and from Avoid, which applies to bad businesses or value-destruction cases. This case is a binary option with real assets but an unproven platform, a valuation that has already been heavily reset, and key catalysts about to arrive. The quality deserves close tracking on the watchlist, so it is not Avoid. But binary uncertainty, structural cash burn, and likely dilution rule out Buy/Hold. Watch is not a weak Buy. Preset downgrade triggers: if REC-1245 Phase 1 disappoints, REC-4881 fails to secure an FDA registration path, or the company uses its ATM heavily at low prices, the rating should be cut directly to Avoid. Ideal buy price <= $2.80, close to per-share net cash of $1.25 plus a first-line safety margin. This is option/scenario valuation rather than DCF intrinsic value, and that "floor" itself is melting.
2. Company Profile
2.1 What It Actually Is: A Hybrid of Picks-and-Shovels Provider and Prospector in AI Drug Discovery
It is incomplete to think of RXRX as either an AI software company or a biotech company. It is a hybrid of both:
The picks-and-shovels side, platform enablement: It uses AI plus automated wet labs to discover targets and molecules for large pharma partners, earning collaboration milestone payments.
The prospector side, in-house biotech: It advances 5 clinical programs itself, betting that one can become a drug.
If the seller of picks and shovels for AI compute is NVIDIA, and in the prior Fabrinet report the equivalent in AI optical interconnect was Fabrinet, RXRX is trying to be the picks-and-shovels provider for AI pharma itself, while also going into the field to prospect. That dual identity is its most important structural tension: it wants to sell the picks and also dig for gold, but so far it has found no gold, with 0 approved drugs, and its picks have not yet been proven better than anyone else's, because the platform has not proven itself.
2.2 Revenue Model: 100% Lumpy Milestones, No Product Revenue
RXRX's current revenue structure, verified against primary SEC filings:
100% comes from collaboration/grant revenue, with zero product revenue. The structure consists of upfront payments, milestone payments for target validation/data-map delivery/candidate nomination, future tiered royalties, and R&D funding.
It has earned more than $500 million cumulatively in non-dilutive upfronts and milestones as of FY2025: about $213 million cumulatively from Roche/Genentech, including a $150 million upfront; about $134 million from Sanofi; Bayer with potentially up to about $1.5 billion in milestones plus royalties, noting this is potential and not yet received; plus Merck KGaA and others.
Revenue is extremely lumpy and shrinking. Under ASC 606, revenue is recognized when performance obligations are completed, so quarterly figures can swing sharply. Q4'25 revenue was $35.5M, mainly from Roche data-map milestones, then Q1'26 plunged to $6.5M, down -56% YoY, because several prior-stage project phases had been completed and less Roche revenue was recognized in the current period. This is not SaaS-like stable platform revenue and cannot be valued like software.
Gross margin is near zero or even negative. FY2025 revenue was $74.7M versus cost of revenue of $70.9M, leaving gross profit of only about $3.7M. In Q1'26, cost of revenue was $12.5M against revenue of $6.5M, so gross margin was negative. Collaboration revenue is essentially cost reimbursement, not a profit engine.
2.3 Core Platform and Data Flywheel
Recursion OS: A full-stack AI operating system spanning biology, chemistry, automation, and data science. The company positions itself as "TechBio."
Automated wet labs: They generate proprietary phenomics data, including cell microscopy images, plus transcriptomics and other perturbation data at industrial scale. Capacity is up to about 2.2 million samples per week, more than 100 billion HUVEC cells per year, and millions of phenomics images per week. Automation reduces biological experiment time and cost by more than 75%.
Proprietary data at the tens-of-PB scale. The company's platform page cites about 36PB, while earlier disclosures cited about 23PB. This report uses the "tens of PB" basis. The data can be distilled into trillions of searchable biological-chemical relationships.
BioHive-2 supercomputer: 63 NVIDIA DGX H100 systems equal 504 H100 GPUs and about 2 exaflops. Completed in 2024-05, it is described as the fastest pharmaceutical industry-owned supercomputer and is fully owned and operated by RXRX. Note that NVIDIA has exited its equity stake, but compute relationship and equity ownership are separate matters.
AI models: Phenom-1, an internal phenomics foundation model; the open-source OpenPhenom, trained only on public images; and MolE for molecular-property prediction, derived from Exscientia's chemistry capabilities. The decision to open-source a model actively shows that the moat lies in data, not model architecture.
Data flywheel logic: massive standardized data to train foundation models to predict targets/molecules, then automated lab validation, then data flows back. Important caveat: the flywheel is not purely self-generated. FY2025 R&D included about $49.9M of data purchased from Tempus, so the flywheel still needs paid external data inputs.
2.4 The Missing Link Added by the Exscientia Merger
On 2024-11-20, Recursion completed an all-stock merger with UK-based Exscientia. The exchange ratio was 0.7729, about 102.3 million shares were issued, RXRX shareholders owned about 74% and Exscientia shareholders about 26%, and the transaction value was about $688M. Exscientia brought precision chemistry, generative molecule design, and automated small-molecule synthesis, extending the flywheel from target finding to molecule building and positioning the company as an end-to-end AI pharma leader. But after the merger, the company cut pipeline programs in 2025, as discussed in Chapter 3. The integration benefits have not yet translated into clinical wins, and Exscientia itself previously had an AI-designed molecule, DSP-1181, fail after Phase 1.
2.5 Company Profile and Essential Characterization
Founded in 2013 as a University of Utah spinout. Headquarters in Salt Lake City. IPO on 2021-04-16 at $18.00.
About 600 employees as of end-2025, down from about 800 at end-2024 due to an approximately 20% workforce reduction in 2025-06.
Management, current: CEO = Najat Khan from 2026-01, formerly of J&J. Founder Chris Gibson became chairman and is leaving the board in 2026-06.
Governance, key positive: no dual-class or super-voting structure, a professional manager has taken over, and the Gates Foundation holds Class A shares with a global access commitment.
Essential characterization: Recursion is the flagship AI pharma platform plus an in-house biotech hybrid. It has real infrastructure assets, blue-chip validation, clean governance, and net cash, but the platform has not proven itself clinically, revenue depends on shrinking lumpy milestones, and it burns cash structurally. It is a binary option with a narrow-to-mid moat of 2.5/5.
3. Vertical Analysis: History and Share-Price Path
3.1 From University of Utah Spinout to AI Pharma Flagship
Founded in 2013: a University of Utah spinout co-founded by Chris Gibson, Dean Li, and Blake Borgeson.
2021-04-16 IPO (NASDAQ: RXRX): issue price $18.00, about 27.88 million shares, gross proceeds of about $501.8 million. The stock rose about +82% on the first day, touching roughly $32 intraday. Note: before the IPO, in 2021-04, the company completed a 1.5-for-1 forward stock split; there have been no splits after the IPO.
2021-12 Roche/Genentech collaboration, with $150 million upfront, neuroscience + oncology, up to 40 programs.
2022-01 Sanofi collaboration; 2023-05 acquisition of Cyclica/Valence to strengthen chemistry/generative AI; 2023-07-12 NVIDIA $50M PIPE investment, a sentiment turning point that drove the stock up to about $11; 2023-11 expanded oncology collaboration with Bayer.
2024-08 announced and 2024-11-20 completed the Exscientia merger, the core strategic event, all stock, +102.3 million shares.
3.2 Major Turn: From Endorsement to Exit
2025-05 cuts to 3 pipelines, a major setback: termination of REC-994 for cerebral cavernous malformation, REC-2282 for NF2-related schwannomatosis, and REC-3964 for C. difficile. The stock fell -13.4% on the announcement day. REC-994 was its pure AI flagship. The SYCAMORE trial met the primary safety/tolerability endpoint, but long-term extension data did not maintain efficacy durability, leading to discontinuation. This was a key blow to the platform-has-not-proven-itself thesis.
2025-06 workforce reduction of about 20%, or roughly 160 employees, with about $11M of severance, tied to cash runway and pipeline rationalization.
2025 Q3+Q4 ATM issuance raised net proceeds of about $387.5M and was fully used. In 2025-11, Q3 revenue fell -80% YoY as milestone recognition dropped sharply.
2025-12-31 NVIDIA fully exited its stake, a major sentiment turn. The roughly 7.71 million shares from its 2023 $50M investment were all sold. This was disclosed in a 2026-02 13F, and after disclosure the stock fell about 12% on 2026-02-18. Around the same time, ARK bought against the trend, adding about 1.25 million shares.
2026-01-01 founder CEO succession: Najat Khan, formerly of J&J, became CEO. Gibson became chairman and will leave the board in 2026-06.
2026-05-06 Q1'26 earnings: EPS of -$0.22 beat consensus of about -$0.26, but revenue of $6.5M was far below expectations of about $16M, a roughly -60% miss, and the stock fell about 5.3% that day.
2026-06-05 -12.76% to $3.315: the overwhelming driver was a broad high-beta growth/AI-biotech sector selloff, with semiconductor macro weakness and AI-biotech peers falling together. This was accompanied by small same-day insider sales by the CEO and founder.
3.3 Share-Price History and Dilution Path, Post-IPO with No Reverse Split
| Milestone | Price / share count | Meaning |
|---|---|---|
| IPO 2021-04-16 | $18.00, first day +82% | Listing |
| All-time peak 2021-07-13 | about $42.81 intraday / about $41.33 close | Bubble top, roughly $41 to $43 basis |
| 2022 to 2023 winter | about $5 | Biotech bear market |
| 2023-07 NVIDIA investment | surged to about $11 | Sentiment high |
| Early-2024 second high | about $12 to $13 | AI speculation |
| 2026-06-05 | $3.315 | Current price |
Dilution path, the core biotech risk: weighted shares rose from about 125 million in FY2021 to about 531 million now, or about +4.2x. The 2024 Exscientia merger added 102.3 million shares, and 2025 ATM issuance added about 134 million shares, roughly +57% in a single year and the heaviest dilution. The main non-dilutive cash sources are collaboration upfronts and milestones.
Current position: about -81.6% from the $18 IPO price, about -92% from the historical peak near $42.81, and about -54% from the 52-week high of $7.18. The 52-week low of $2.77 is likely the post-IPO all-time low, below the prior bear-market trough. Strong narrative point: after NVIDIA's endorsement and the Exscientia merger, the stock made a new historical low.
4. Financial Review
4.1 Income Statement, Primary SEC Basis, USD Millions
| Item | FY2025 | FY2024 | Q1'26 | Q1'25 |
|---|---|---|---|---|
| Total revenue | 74.7 | 58.8 | 6.5 (-56.1% YoY) | 14.7 |
| Cost of revenue | 71.0 | 45.2 | 12.5 | 21.8 |
| R&D | 475.3 | 314.4 | 87.9 | 129.6 |
| G&A | 176.6 | 178.2 | 34.6 | 54.7 |
| Operating loss | (648.1) | (479.0) | (128.5) | (191.4) |
| Net loss | (644.8) | (463.7) | (117.5) | (202.5) |
| EPS, basic = diluted | -$1.44 | -$1.69 | -$0.22 | -$0.50 |
TTM net loss = 644.8 - 202.5 + 117.5 = $559.8M, matching the $559.78M price anchor. TTM EPS was -$1.17, on about 478 million TTM weighted shares. The two EPS bases cover different periods. This report uses FY2025 full-year -$1.44 and TTM -$1.17 as load-bearing figures and labels them explicitly. Do not confuse them.
Revenue quality is extremely poor: 100% collaboration milestones, zero product revenue, gross margin near zero or negative, and very lumpy quarters dominated by the recognition schedule of one partner, Roche. Revenue cannot serve as a valuation anchor or reliable cash source.
FY2025 net loss worsened by +39%, mainly due to full-year consolidation of Exscientia, acquired IPR&D, and Tempus data purchases. But Q4'25/Q1'26 already turned to sharp narrowing through expense discipline. That is the source of the Q1'26 EPS beat. The simultaneous revenue miss means it was not a clean beat.
4.2 Cash Burn and the Melting Floor
Balance sheet, 2026-03-31: cash and equivalents about $654.5M plus restricted cash about $5.5M equals cash-like assets of about $665M, versus $753.9M at end-2025. Almost zero interest-bearing debt, with only notes/finance leases of about $18.7M excluding operating leases; company-reported total liabilities about $72M. No convertible debt, therefore material net cash. Shareholders' equity was about $1.025B, and accumulated deficit was about $2.19B.
Cash per share about $1.25, about 38% of the share price, acting as a downside "floor."
Cash burn: FY2025 GAAP operating cash outflow was about $371.8M. The most realistic all-in quarterly net cash decrease was about $88.7M per quarter in Q1'26. During that quarter, no amount was drawn from the new $300M ATM, so the quarterly cash decrease was pure burn and the cash-quality signal was clean. Annualized, this is about $355M. 2026 cash-burn guidance is less than $390M.
Cash runway: management guides to early 2028, with no additional financing needed. $665M divided by about $355M per year equals roughly 7 to 8 quarters, which is internally consistent. Q1'26 maintained/reinforced early 2028, rather than extending it again.
"Net cash is a melting floor," the key point: the cash balance is large, but the company burns about $340M to $390M per year, meaning the floor shrinks by roughly half each year. A new $300M ATM facility established in 2026-02 hangs over the stock. It was not used in Q1'26 but can be used at any time. The company is highly likely to need another dilutive financing during 2027. This is the most realistic downside variable, and the bear-case "floor" will move down with burn and dilution.
4.3 Valuation Bridge
Net cash about $665M and debt about $18.7M mean material net cash. EV is about $1.76B - $0.665B, or roughly $1.1B on a consistent basis. On a stockanalysis net-cash basis, EV is about $1.18B; the difference comes from liability assumptions, and both are noted.
In other words, the market is paying about $1.1 billion above net cash for the platform plus pipeline, a premium on an unvalidated platform that has not produced an approved drug and whose collaboration revenue is still falling.
P/B is about 1.7x. EV/annual cash burn is about 3.1x. The implied platform premium in the market cap is roughly equivalent to 3 years of cash burn.
5. Moat: Narrow to Mid, 2.5/5, Real Assets Plus an Unfalsified Narrative
RXRX's moat must be separated into two layers. This is the technical core of the conclusion that the platform has not proven itself.
5.1 Input-Side Moat = Real but Not Yet Proven Through Output
| Pillar | Score | Evidence |
|---|---|---|
| Data assets | 4/5 | Tens-of-PB proprietary phenomics data, up to about 2.2 million samples per week, more than 100 billion HUVEC cells per year, and automated wet labs. Replication would require heavy capital and many years, making it one of the largest in its class. But "larger scale" does not equal "the right data," and predictive power for clinical endpoints remains unproven, so the score is capped at 4. |
| Partner network | 3/5 | Roche/Sanofi/Bayer have paid more than $500 million of real non-dilutive upfronts and milestones. Blue-chip willingness to pay for platform target discovery is the strongest external validation that the data has commercial value. But milestones are lumpy and non-recurring, partners are also building internally, Bayer's fibrosis collaboration was previously narrowed, and no project has produced an approved drug. |
| Compute | 2/5 | BioHive-2, with 504 H100 GPUs and 2 exaflops, is a leading specification, but compute can be rented or bought; NVIDIA is not exclusive; the advantage commoditizes over time; and NVIDIA has exited the equity stake. |
| Algorithms/models | 2/5 | Phenom-1 and MolE are solid but not the core moat. The active open-sourcing of OpenPhenom itself supports the point that the moat is data, not model architecture. Algorithms are diffusing rapidly across the industry. |
| Switching costs/customer stickiness | 2/5 | After data maps are delivered, pharma partners can shift to internal development. There is no Schrödinger/Certara-style recurring software seat or installed-base stickiness. On the in-house pipeline side, there is no "customer" at all. |
5.2 Output-Side Moat = Mostly Narrative for Now
The investment thesis layer, "data flywheel to higher clinical success rates to approved drugs to excess returns," is directly weakened by three facts:
So far, 0 AI-discovered drugs have been approved, across the whole industry. The first is expected around 2026 to 2028.
The only pure AI flagship, REC-994, has been discontinued. SYCAMORE met the primary safety endpoint, but efficacy durability was not maintained.
The most advanced clinical asset, REC-4881, is actually an in-licensed old Takeda drug, formerly TAK-733, which Takeda had evaluated in solid tumors and RXRX later in-licensed and repositioned for FAP. AI's role was to identify the mechanism, that MEK1/2 inhibition could rescue APC loss of function, not to design the molecule from scratch. Its data do have bright spots: Phase 1b/2, median polyp burden down -43% at 13 weeks (N=12) and -53% at 25 weeks (N=11), with 73% achieving >=30% durable response. But that is exactly the point: the asset the company calls its "first AI clinical validation" does not validate that AI can design a better drug from scratch.
Industry-level BCG data show AI drugs have higher Phase 1 success rates of 80% to 90%, but Phase II, the real efficacy test, has a success rate of about 40%, roughly in line with history. The improvement is in safety, not efficacy.
5.3 Overall Moat Score and Why It Is 2.5/5
Overall score: 2.5/5, narrow to mid, with optionality toward "mid." The input/infrastructure layer is made of real assets, real barriers, and real validation. The data asset alone can be scored at 4. But the output/investment-thesis layer has not been validated by any approved drug or uplift in clinical success rates. With no realized excess returns so far, meaning deep losses, 0 approved drugs, and a discontinued flagship, the economic moat, defined as the ability to protect long-term excess returns, is currently judged narrow. An upgrade to "mid" requires an in-house de novo pipeline such as REC-617 or REC-1245 to generate real Phase II efficacy. Continued failures or partner contraction would move the score down.
Comparison with prior reports: Canaan (Avoid, 2/5 moat, bad business with bottom-tier share and no pricing power) < RXRX (Watch at the cautious lower edge, 2.5/5, real assets but unproven platform) < Fabrinet (Watch, 3/5, strong execution in narrow-moat contract manufacturing) < Cheniere (Hold, 4/5, wide-moat infrastructure). RXRX and Canaan are both unprofitable, cash-burning story stocks, but RXRX has real assets, blue-chip validation, net cash, clean governance, and a price that has already reset. Its quality tier is clearly higher, justifying a one-notch rating gap.
6. Industry Demand: AI Pharma in the Post-Deglamorization Show-Me Phase
6.1 Sector Drivers and Market Size, with Highly Divergent Definitions
Core drivers, already established: single-drug development costs are high, about $2.6 billion per approved drug on the Tufts basis, while RAND 2025 gives a median of about $708 million. Methodologies differ and both are labeled. Roughly 90% of drugs entering clinical development fail, and Phase II success is only about 28%. Foundation models such as AlphaFold and GPU compute breakthroughs are real. Any tool that can improve this failure curve has a huge TAM.
Market size, with extremely divergent definitions: institutions estimate the 2025 market from about $2.35B to about $19.89B, a nearly 8x spread because the boundary of "AI pharma" differs: pure software versus services included versus pipelines included. CAGR is generally forecast at 20%+. This report uses the directional signal, not a single precise anchor.
6.2 Cycle Position: Early Recovery After Deglamorization, a Show-Me Phase
The 2021 bubble peak, with AI/techbio VC around $12.5B, was followed by the 2023 winter at about $4.8B and several flagship AI drug clinical failures at Exscientia and BenevolentAI. Funding then bottomed and recovered in 2024 to about $6.7B. By 2026, recovery is clearer, with the biotech IPO window reopening: Eikon raised $381M and Generate raised $400M in 2026-02 IPOs. The AI bull market has also lifted valuations and the financing environment.
Conclusion: the industry is in an early post-deglamorization recovery / show-me phase. It has moved off the 2023 winter bottom but is far from the indiscriminate 2021 frenzy. The narrative has shifted from "can AI generate hypotheses" to "can those hypotheses repeatedly survive clinical trials." That is especially demanding for unprofitable, no-approved-drug pure platform names, and RXRX is the archetype.
6.3 The Value-Capture Problem, the Most Important Point
The areas where AI can clearly create value today do not overlap with the most expensive part of the value chain:
Where AI works = early discovery/preclinical, the cheapest segment. It can compress discovery timelines by 30% to 40% and shorten candidate generation from 3 to 4 years to 13 to 18 months.
Where AI has limited proven impact = clinical trials, the most expensive, slowest, and decisive segment, which accounts for most R&D cost and failure. Biology, patient enrollment, and regulation constrain it. AI has not yet been proven to improve the roughly 90% clinical failure rate. The measurable value of AI in clinical stages today is mostly in trial operations, such as recruitment, site selection, and data automation, not in making the molecule itself more likely to become a drug.
Implication for RXRX: its entire valuation anchor is the hypothesis of "better starting molecules to higher clinical success rates," which is neither falsified nor proven. If AI's value ultimately sits mainly in preclinical speed-up, the cheapest segment, rather than clinical de-risking, the most expensive segment, then the platform will capture far less value than the narrative suggests. This is the fundamental worry behind NVIDIA's exit and the sell-side Hold stance.
The one strong counterexample: Insilico Medicine's rentosertib, the world's first drug whose target and molecule were both designed by generative AI, showed positive Phase 2a data in idiopathic pulmonary fibrosis and was published in Nature Medicine in 2025-06. It is the strongest single-point evidence so far that an AI drug can survive clinical proof of concept, but it remains a Phase 2a/single-asset case and is insufficient to prove a sector-wide success-rate uplift.
6.4 Competitive Landscape and RXRX's Position
Platform companies, closest to RXRX: RXRX, with phenomics + automated labs + end-to-end integration; Schrödinger (SDGR), with physics-based computational software and profitable software cash flow; XtalPi (2228.HK), with quantum physics + AI + robotics.
Antibody companies: AbCellera and Absci. Targeting companies: Relay. Plus large pharma internal AI, Google/Isomorphic, and Insilico, which leads in clinical validation.
RXRX's relative position: in data scale and vertical integration, it sits in the first tier of the industry narrative. In realized commercialization and clinical output, it lags SDGR, which has software cash flow, and Insilico, which has Phase 2a validation. Its differentiation is proprietary data plus an automated flywheel. Its weaknesses are 0 approved drugs, lumpy revenue that repeatedly misses expectations, and a valuation that still feels expensive.
Industry judgment, balanced: AI pharma is a real productivity tool in discovery/preclinical work, with speed gains already established. What is overestimated is the end-to-end disruption narrative. AI has not yet shown systematic improvement in the clinical success rates that determine outcomes. RXRX is a high-beta, large-scale bet on that unproven assumption.
7. Horizontal Analysis: The Scale Leader Among Unvalidated Platforms
7.1 AI Pharma Peer Comparison, Price Anchor 2026-06-05 Close
| Company | Ticker | Market cap | Cash/liquidity | Annual cash burn | Profitable? | Pipeline/revenue | 6/5 day move |
|---|---|---|---|---|---|---|---|
| Recursion Pharmaceuticals | RXRX | $1.76B | $665M | ~$340-390M, highest | No, TTM net loss $559.78M | Clinical, up to Phase II, 0 approved | -12.76% |
| Schrödinger | SDGR | $1.07B | $406M | ~$200M+ | No at company level, software segment profitable | Software, profitable, + in-house pipeline | -9.3% |
| Relay | RLAY | $2.95B | $642M | ~$290M | No | Clinical, 1 Phase 3 asset | -7.0% |
| AbCellera | ABCL | $1.72B | $655M | ~$170M | No | Just entering clinical / royalty-share model | -11.8% |
| Absci | ABSI | $1.0B | $126M | ~$118M | No | Early clinical | -12.8% |
| XtalPi | 2228.HK | ~$4.1B | ~$1.0B | Negative operating cash flow | Marginally profitable | CRO/CDMO + AI + robotics | Hong Kong stock |
| Certara | CERT | $0.84B | — | Near breakeven | Adjusted profitable | Commercial software, no in-house drugs | -4.3% |
| Simulations Plus | SLP | $0.32B | — | — | Recently profitable | Commercial software, no in-house drugs | -4.6% |
7.2 Three Core Judgments
The whole cohort sold off together on 6/5, hard evidence of a sector drawdown: RXRX/ABSI/ABCL fell -12% to -13%, SDGR -9%, and RLAY -7%. AI-biotech peers fell in tandem that day, confirming a risk-off sector selloff rather than RXRX-specific fundamental bad news. There were small same-day insider sales, but they were not the main driver. This also creates a useful echo with the prior Fabrinet report, where Fabrinet fell -13.09% on the same day. Both ends of the AI theme, software-platform RXRX and hardware contract-manufacturer Fabrinet, were hit on the same day. Fabrinet was a profitable business incorrectly sold off; RXRX is an unprofitable option that moves with the sector.
RXRX = the scale leader among unvalidated platforms: by market cap, it is mid-pack at $1.76B, not the largest, since RLAY and XtalPi are larger. By integration breadth across data, compute, pipeline, and partnerships, it is the scale leader among AI-native platforms. By cash burn, it is the highest. On profitability, it sits in the pure cash-burn tier, unlike SDGR's profitable software segment and the commercial profitability of Certara/SLP.
Valuation: cash support, platform premium still unproven. EV is about $1.1B to $1.2B. It is cheaper than RLAY/XtalPi on absolute EV, but expensive or fragile relative to SDGR/Certara, which have recurring software revenue. Investors are paying about an $1.1 billion premium for an unvalidated platform that has not produced an approved drug and whose collaboration revenue is still declining. The $665M of net cash, about 38% of market cap, limits downside, but the platform premium needs pipeline or collaboration delivery to hold.
7.3 The Paradigm Gap Versus Traditional Large Pharma
End-drug companies such as Novartis (NVS, about $286.5 billion), AstraZeneca (AZN, about $290.9 billion), and BioNTech (BNTX, about $26.0 billion) are valued on earnings multiples, DCF, and dividends. RXRX is valued on option value and story, and has never been profitable. BNTX is precisely the arc RXRX wants to follow but has not yet reached, monetizing a platform and then reinvesting. The AI bull market has lifted the premium for platform stories. The single-day -12.76% move on 6/5 exposes how violently these story stocks can de-rate.
7.4 Sell-Side Consensus, Source Attribution and Timeliness
According to stockanalysis, using S&P Global data anchored on 6-05: 8 analysts = 1 Strong Buy / 2 Buy / 5 Hold / 0 Sell = consensus Hold. The target-price range is low $3.00 / median $6.00 / mean $6.64 / high $10.00. The current price of $3.315 is already near the low end, implying roughly +100% upside to the mean target, yet consensus remains Hold. This is typical high-risk platform biotech, where target prices are binary scenario-weighted. Aggregators disagree on analyst count and targets; TipRanks has a more bullish reading and an average around $8. This report uses stockanalysis as the single source. Key signal: 0 Sell ratings coexist with short interest around 33% to 37% of float. Wall Street is highly skeptical but reluctant to formally go short, and the long-short divide is extreme.
8. Current Fundamentals
Share price: 3.315 USD on 2026-06-05, -12.76%. The 52-week range was 2.77 to 7.18, about -54% from the high, +20% from the low, and about -92% from the 2021 peak.
Market cap: about $1.76 billion; shares outstanding about 531 million.
Valuation: no PE due to losses, TTM net loss $559.78M, TTM EPS -$1.17, FY2025 full-year -$1.44, no dividend, EV about $1.1B to $1.2B, and cash per share about $1.25, or about 38% of the stock price.
Balance sheet: cash-like assets about $665M, almost zero interest-bearing debt, material net cash, accumulated deficit about $2.19B, shareholders' equity about $1.025B, and a new $300M ATM facility established in 2026-02 hanging over the stock.
FY2025: revenue $74.7M, net loss $644.8M, EPS -$1.44, gross margin near zero.
Q1'26, released 2026-05-06: revenue $6.5M, -56% YoY; net loss $117.5M; EPS -$0.22, beating consensus of -$0.26 but revenue missed badly; stock down -5.3% that day.
Cash runway: to early 2028, with annual cash burn of about $340M to $390M.
Governance/shareholders: no super-voting rights; NVIDIA exited by end-2025; founder CEO stepped down in 2026-01 and leaves the board in 2026-06; Gates Foundation holds shares.
9. Valuation: Option/Scenario Valuation, Not DCF, Cheap Because the Risk Is Real
9.1 Methodology Disclaimer, Read First
RXRX is a pre-revenue, binary, structurally cash-burning biotech with no PE and no stable cash flow to support DCF. A point estimate of fair value and a narrow scenario band are fundamentally option/scenario valuation, not DCF intrinsic value. Its value is approximately net-cash cushion + platform-credibility option + clinical-pipeline option, or rNPV, and the latter two remain unproven. The bear-case "floor," net cash, is itself melting through annual burn of about half the cash balance plus dilution. It is not a hard floor.
9.2 Scenario Analysis
| Scenario | Valuation range (USD) | Triggers |
|---|---|---|
| Bear | 1.5 to 2.8 | Key pipelines REC-1245/REC-4881 disappoint or further falsify the platform + large low-price ATM dilution in 2027 + AI theme fades, driving the stock toward the melting net-cash level. Note: the $1.5 lower end is slightly above current net cash per share of $1.25 because it assumes some remaining cash/takeout value; in reality it can approach or even fall below net cash. |
| Base | 3.0 to 5.0 | Muddle-through: runway safe, pipeline progresses gradually, theme neutral, no major readout. The current price of $3.315 sits in the lower part of this range, spanning the sell-side low of $3 to the mid-range of $5 to $6. |
| Bull | 7 to 10 | Key clinical readouts turn positive / platform receives substantive validation / a new large pharma partnership arrives / AI theme revives. This aligns with the 52-week high of $7.18 and the sell-side high of $10. |
The current price of $3.315 sits near the lower end of the base range. The market is pricing the company cautiously and on the skeptical side, consistent with consensus Hold, NVIDIA's exit, and high short interest.
9.3 Valuation Conclusion
Cheap, but because the risk is real, not because the asset is clearly good. The current price implies roughly +100% upside to the sell-side mean target of $6.64, while net cash is about 38% of market cap and the stock is close to the historical bottom. There is a melting net-cash cushion on the downside and a binary option on the upside. Whether this cheapness is a value option or a value trap with melting ice underneath depends on whether the platform can be validated. Ideal buy price <= $2.80, close to per-share net cash of $1.25 plus a first-line safety margin, meaning "buy only below net cash plus a thin cushion." That is the appropriate YMYL conservatism for a binary, structurally cash-burning, likely-to-dilute name. The current $3.315 price is above this level, so the stance is Watch, not chase.
10. Risks, Including Pre-mortem
10.1 Core Risk List, by Severity
Platform validation + binary clinical risk, highest and most central: 0 AI-discovered approved drugs. The first risk item in the FY2025 10-K is that the company's unique drug-discovery approach may not lead to successful medicines. The flagship REC-994 has been discontinued. The leading REC-4881 asset is an in-licensed old drug, reducing its evidentiary value for the platform. Failure of any key pipeline would collapse the core thesis.
Dilution/financing, high: structural heavy cash burn, with FY2025 net loss of $644.8M and accumulated deficit of $2.19B. Cash reaches early 2028, but the floor melts by roughly half each year, and the new $300M ATM established in 2026-02 hangs over the stock. Another dilution during 2027 is highly likely, making it the most realistic downside variable. Note: the 10-K contains no going-concern warning.
AI pharma theme / valuation beta, high: market cap of $1.76B against almost zero product revenue means pure platform/story valuation. Any cooling in the theme compresses valuation, as the 6-05 -12.76% move showed. Short interest around 33% to 37% of float is double-edged, amplifying selloffs while creating squeeze potential.
Integration + personnel, medium-high: after the Exscientia merger, 2025 saw a 20% workforce cut and synergies have not converted into clinical wins. Founder CEO exit plus NVIDIA, the top endorser, selling out means narrative pillars have left.
Partner dependence, medium: revenue is 100% dependent on lumpy milestones and fell -56% YoY in Q1. There is no current collaboration termination, but if a major partner reduces investment it would hit both narrative and cash. Bayer's fibrosis subprogram was previously narrowed.
Competition, medium: large pharma internal AI plus Isomorphic, XtalPi, Insilico, and many other platforms compete. The 10-K also lists a dedicated risk that AI regulation could limit AI usage.
10.2 Practical Attribution of the 6-05 Selloff
The 6-05 -12.76% move was overwhelmingly driven by a broad high-beta growth/AI-biotech sector selloff plus semiconductor macro weakness, with AI-biotech peers falling in tandem that day. It was accompanied by small same-day insider sales by the CEO and founder. It was neither a clean case of "pure sector move, zero company-specific news," nor a fundamental company-specific negative. It exposes the fact that when AI pharma sentiment cools, pure platform story stocks de-rate the most violently.
10.3 Pre-mortem: If This Is a Failed Investment Three Years From Now
The most likely script is that the platform narrative fails to translate into approved drugs and structural dilution continues. REC-4881, a small-indication in-licensed asset, may disappoint on FDA pathway or commercialization. One or two key early de novo readouts from REC-1245/REC-617 may fail, turning clinical binary risk into reality. The AI pharma theme may fade while the company keeps burning about $350M to $390M per year, forcing a large low-price equity raise in 2027 and further diluting 531 million shares. Large partners such as Roche/Sanofi may slow milestones, drying up the only revenue source. The final outcome could be a shrinking market cap around "high compute + beautiful platform but no blockbuster product," or even a takeout/privatization where the acquirer captures the remaining net cash. That is why the rating is Watch at the cautious lower edge with explicit downgrade triggers, not a weak Buy. Every link in this failure chain already has real-world warning signs.
11. Catalyst Tracking
11.1 Upside Catalysts, Also Validation Nodes
REC-1245, an RBM39 degrader for solid tumors/lymphoma, Phase 1 monotherapy data in 1H26, the nearest and most tangible de novo asset readout.
REC-4881 for FAP, FDA registration-pathway discussions in 1H26. Securing an accelerated path would trigger re-rating.
Potential Sanofi/Roche milestones over the next 12 to 18 months; REC-7735/REC-102 go/no-go decisions in 2H26.
Revival of the AI pharma theme plus a new major large-pharma partnership.
11.2 Downside Catalysts, Also Downgrade Triggers
Disappointing REC-1245 Phase 1 data / REC-4881 failing to secure an FDA registration path, leading to a downgrade to Avoid.
Heavy use of the $300M ATM at low prices, leading to a downgrade to Avoid.
Cooling AI pharma sentiment / high-beta selloffs, already seen; large partners slowing milestones.
11.3 Tracking Metrics
Pipeline readouts and regulatory milestones for REC-1245/REC-617/REC-4881 and others; collaboration milestone realization and new signings; quarterly revenue, which is lumpy, and cash-runway markers; ATM usage and share-count dilution; cash-burn rate; AI pharma theme sentiment and peer valuations; changes in short interest. 2026 is a decisive year of falsification or validation, and these nodes will directly determine whether the rating is upgraded or downgraded.
12. Zen Horizon Intersection
Vertical view, history and share-price path: From a University of Utah spinout in 2013 to the 2021 IPO at $18, first-day +82%, and a peak near $42.81, Recursion became the sector flagship through the AI pharma narrative, NVIDIA's investment, and the Exscientia merger. But in 2025 it cut pipelines, reduced staff, and diluted heavily. By end-2025 NVIDIA had fully exited, and in early 2026 the founder CEO stepped down. The stock made a new all-time low, drew down about 92%, and fell below the prior bear-market trough. This is the full arc from endorsement to endorsement exit.
Horizontal view, peer comparison: Within the AI pharma cohort, RXRX is the largest integrated, heaviest cash-burning, but still commercially and clinically unproven scale leader among unvalidated platforms. By breadth of data, compute, and pipelines, it is a first-tier narrative. By realized profit and clinical output, it lags SDGR, with software cash flow, and Insilico, with Phase 2a validation.
Intersection conclusion: The vertical picture of endorsement exit + founder departure + new low, and the horizontal picture of a large integrator whose platform remains unproven and cheap only because risks are real, reinforce the same characterization: this is a binary option with real infrastructure assets, blue-chip cash validation, clean governance, and net cash, but no clinical proof yet for the platform. Its place on the rating ladder is clear: Canaan (CAN, Avoid, bad business 2/5, value destruction, super-voting control, price overextended) < Recursion Pharmaceuticals (RXRX, Watch at the cautious lower edge, real assets but unproven platform 2.5/5, binary option, valuation heavily reset) < Fabrinet (FN, Watch, strong execution in narrow-moat contract manufacturing 3/5, but expensive) < Cheniere (LNG, Hold, wide-moat infrastructure 4/5). The mirror image with Canaan is especially clear: both are unprofitable, cash-burning story stocks, but RXRX has clean governance with no super-voting rights, real assets and blue-chip validation, and a valuation already reset. Its quality tier is clearly higher, lifting it from Avoid to Watch at the cautious lower edge.
Rating: Watch, cautious lower edge, one step away from Avoid. It differs from Buy/Hold, which require a margin of safety or a validated good business, and from Avoid, which applies to bad businesses/value destruction. This case is a binary option with real assets but an unproven platform, a valuation that has already been heavily reset, and key catalysts about to arrive. The quality deserves close watchlist tracking, so it is not Avoid. But unresolved binary risk, structural cash burn, likely dilution, and a narrow moat rule out Buy/Hold. It is an AI pharma platform lottery ticket that has already been heavily reset and whose result is still unknown. It is worth close tracking, but it is not a blind-buy stock today. Watch is not a weak Buy. Preset downgrade triggers: REC-1245 Phase 1 disappoints, REC-4881 fails to secure an FDA registration path, or the company uses the ATM heavily at low prices, in which case the rating moves down to Avoid. Ideal buy price <= $2.80, close to per-share net cash plus a first-line safety margin; this is option/scenario valuation rather than DCF, and the floor is melting.
Research Uncertainty
Two EPS bases, explicitly separated: TTM EPS of -$1.17, with net loss of $559.78M and about 478 million TTM weighted shares, versus FY2025 full-year EPS of -$1.44, with net loss of $644.76M and about 448 million weighted shares. Different periods; do not collapse into one figure.
REC-4881 is an in-licensed old Takeda drug, key point verified: the original Takeda compound, code TAK-733, was repositioned by RXRX for FAP after in-licensing. AI's role was identifying mechanism, not designing the molecule from scratch. Data are Phase 1b/2, small sample, N=11 to 12, and median-based. "First AI clinical validation" does not equal "AI designed a better drug from scratch."
6-05 selloff attribution, softened: sector/macro selloff was the overwhelming driver, with AI-biotech peers falling together, alongside small same-day insider sales. It was neither a clean "pure sector" move nor company-specific fundamental bad news. Precise peer declines such as SDGR -9% and the Philadelphia Semiconductor Index -10.3% are multi-source corroborating evidence; not every individual figure was pinned down one by one.
NVIDIA has exited and founder CEO has changed, verified: NVIDIA had sold out about 7.71 million shares by 2025-12-31, disclosed in a 2026-02 13F. CEO changed to Najat Khan in 2026-01, and Gibson leaves the board in 2026-06, not due to disagreements.
"Net cash is a melting floor": cash about $665M and runway to early 2028, but annual burn of about $340M to $390M means the floor shrinks by about half each year. The $300M ATM overhang makes 2027 dilution highly likely. The $1.5 bear lower bound assumes some remaining cash/takeout value, not a hard floor.
Valuation method: a pre-revenue binary biotech should use option/scenario valuation, not DCF intrinsic value. Scenario bands and fair_buy are risk-weighted judgments, not precise predictions.
Approximate items still labeled as approximate: accumulated deficit about $2.19B, reconcilable but not pinned to one single filing in this report; proprietary data at the tens-of-PB scale, with company basis about 36PB and earlier basis about 23PB; Bayer's about $1.5 billion refers to potential milestones plus royalties; interest-bearing debt of about $18.7M excludes operating leases, while company-reported total liabilities are about $72M; EV of $1.1B to $1.2B varies with net-cash definition.
Sell-side consensus: from stockanalysis, 8 analysts, consensus Hold, target mean $6.64. TipRanks has a more bullish reading. This report uses a single source.
This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.
Full report
Sign in to read the full report
Sign up free to unlock the full text, the Baillie growth scorecard, and full-text search.
Log in / Sign up free