ARTV: An "immune reset" bet. Can you trust management's market size?
Deconstructing Artiva's AlloNK valuation chain: from 1.5 million patients down to the 35,000 that actually matter
Artiva Biotherapeutics has no revenue and no products approved. Its lead asset AlloNK is an off-the-shelf natural killer (NK) cell therapy made in bulk from healthy donor cord blood. Combined with rituximab (an anti-CD20 monoclonal antibody with decades of clinical experience), it is designed to provide refractory rheumatoid arthritis (RA) patients with a one-time “immune reset.”
In May 2026, the company shared early clinical data: 71% ACR50 response rate (defined as 50% or greater improvement in joint swelling, tenderness, and other disease activity markers) among 7 refractory RA patients with 6-month follow-up. It also announced FDA alignment on a single ~150-patient randomized controlled Phase 3 study as the registration pathway, and concurrently closed a $300 million equity raise.
The stock has moved from a 52-week low of $1.47 to over $14, and has since pulled back to current levels (~$9). What does the market think is going to happen? And are they right?
Starting with the biology: what AlloNK actually does
RA is caused by a rebellion of B cells in the immune system. These B cells produce misdirected antibodies that attack the synovial membrane lining your joints, causing chronic inflammation, swelling, and ultimately joint destruction.
The predominant approach to RA therapy today is “continuous suppression”: taking medications daily or weekly that dial down your immune system. Most patients find some point on the RA treatment ladder where things improve enough to live normally. But for about 13% of patients who have tried biologic drugs and failed two or more medications with different mechanisms of action, these drugs don’t work. They are labeled D2T-RA, for difficult-to-treat rheumatoid arthritis.
Instead of forcing patients onto another suppressive medication, AlloNK attempts to reboot their immune system completely. First, rituximab marks the bad B cells for destruction by binding to CD20 proteins on their surface. Then a large infusion of donor-derived NK cells finds and kills the rituximab-coated B cells in a process called ADCC (antibody-dependent cellular cytotoxicity). If the depletion is thorough enough, the immune system may regenerate without the autoimmune memory, effectively patching the bug.
There is one important caveat: prior to infusion, patients must receive low-dose chemotherapy conditioning (specifically cyclophosphamide plus fludarabine) to temporarily suppress their own immune system, creating a survival window for the donor NK cells. The need for chemotherapy conditioning is the single biggest bottleneck in ARTV’s commercial story. You have to convince patients who are not at risk of dying from rheumatoid arthritis to receive chemotherapy.
Early data: tiny sample, enormous delta
71% from 7 patients is statistically worthless. Right?
Except: even at the low end of the 95% confidence interval (~29%), AlloNK plus rituximab still beats the high end of external literature control (~27%) for rituximab monotherapy.
So we know it’s better than the standard of care. We just don’t know by how much.
Running 500,000 Monte Carlo simulations with empirically calibrated open-label-to-RCT shrinkage factors (historical RA programs suggest shrinkage of roughly 12 to 20%, lower than many assume) and literature-anchored control arm assumptions (rituximab only ACR50 of approximately 20 to 27%), we calculate a 77 to 85% likelihood that AlloNK plus rituximab achieves statistical significance in Phase 3. After down-stepping for risks not captured by the model (enrollment population differences, clinical meaningfulness thresholds, safety signals, and execution risk), we use 60% as our composite probability that the Phase 3 data will be positive and large enough to re-rate the stock meaningfully.
The core of the valuation chain: how many patients will actually use this drug
This is the fundamental investable question. It isn’t whether or not Phase 3 works (it’s probably going to work). It’s how much the drug can make if it does work.
Artiva management says there are 150,000 to 200,000 eligible patients in the United States. This figure is featured prominently in the May 8, 2026 8-K filing (Exhibit 99.3, slide 12), complete with four footnoted citations of underlying medical literature. We tracked each paper down to the original study.
Paper 1: Roodenrijs 2021 (Ann Rheum Dis 2021; 80:31-35). This paper is actually the EULAR consensus on what D2T-RA means. A group of doctors sitting down and deciding on diagnostic criteria. Nothing in the paper talks about prevalence or how common D2T-RA is. It contributes zero to answering the question “how many patients is 150,000 to 200,000.”
Paper 2: Watanabe (KURAMA cohort, Immunol Med 2022; 45:35-44). This is a Kyoto University cohort study. They report the prevalence of D2T-RA as 7.9% of all RA patients, the lowest of the 4 papers cited by Artiva. If we apply that percentage to the United States: 7.9% x 1.5 million = 118,500. That doesn’t even reach management’s low end of 150,000 possible patients. This number is not mentioned at all in any of the company presentations. It was cited, but Artiva ignored its actual number.
Paper 3: Jung 2023 (KOBIO registry, Arthritis Res Ther 2023; 25:174). In plain English: “Among 2,321 RA patients treated with b/tsDMARDs, 271 (11.7%) were diagnosed with D2T RA.”
The denominator of that sentence: 2,321 patients who are currently taking biologic meds. NOT all patients with RA. The 11.7% figure presented by management is a percentage of patients taking biologics, yet Artiva used the 1.5 million total RA population as the denominator.
Applying the numbers correctly: 11.7% x 575,000 Americans currently treated with biologic medications = 67,000 patients. Still nowhere close to the range management presented on its slides.
Paper 4: Paudel 2024 (BRASS cohort, Rheumatology 2025; 64(3):1102-1110). This came from Brigham and Women’s Hospital in Boston. It has the highest percentages of any paper cited: 14.4% prevalence of D2T among all RA patients, and 22.3% prevalence of D2T among biologic-treated patients. But Paudel himself writes explicitly in the discussion section that BRASS is a single academic medical center with a 65% b/tsDMARD exposure rate, far above the national average of ~38%, and that the results “may not be applicable to other patient populations.” Using the highest value from an academic referral center whose own author warns against generalization, as an anchor for national market sizing, is selective citation.
So management took the 14.4% from BRASS (the highest academic single-center figure) and the 11.7% from KOBIO (whose denominator is actually the biologic-treated population, not all RA), combined them into a “10 to 15%” range, and multiplied by 1.5 million total RA patients to arrive at 150,000 to 200,000. A definition paper with no data was cited for legitimacy. A study yielding 7.9%, the lowest estimate, was cited but its number was buried.
Do the math, and it breaks.
150,000 to 200,000 eligible patients divided by 575,000 Americans currently taking biologics = 26 to 35% of biologic patients are D2T.
In plain English: if management’s number is true, then approximately one out of three patients taking a biologic medication qualifies for treatment with AlloNK.
The KOBIO registry says only 11.7% are D2T. The meta-analysis of international biologic-treated RA patients says only 13.2% are D2T. Even the BRASS study from Boston shows only 22.3% of their patients were D2T-RA.
Zero sources come close to the 26 to 35% implied by management.
The correct approach requires three filtering steps, each grounded in a specific published source.
A recent meta-analysis published in Annals of the Rheumatic Diseases in 2025 by Xie et al. (Ann Rheum Dis 2025; 84(12), PMID 41188120) pooled 23 individual studies from 13 countries with a total of 27,987 patients with RA. The paper reports two critical subgroup figures: D2T prevalence of 10.9% in studies using an all-RA denominator, and 13.2% in studies restricted to b/tsDMARD-exposed populations. The full text, on page three of the methods section, explicitly states that the treated subgroup includes “patients exposed to b/tsDMARDs.” In clinical epidemiology, “exposed to” is standard terminology for ever-used, not currently-on.
Step one: 1.4 million Americans are diagnosed with RA. Only about 37% (~520,000) have ever used a biologic medication or JAK inhibitor. That number comes from the Optum Clinformatics database (biologic point-prevalence ~20% in 2016 to 2021, with cumulative ever-use higher), cross-validated against Artiva’s own 8-K slide deck figure of 575,000/1,500,000 = 38%. The EULAR D2T definition requires failure of at least two distinct-mechanism b/tsDMARDs. Patients who have never used a biologic drug cannot have failed two of them. Exclude them from the denominator.
Management multiplied 11.7% (the pooled average of both subgroups) by 1.5 million total RA patients to get their figure of 150,000 to 200,000. This is a mismatch that can be algebraically proven wrong.
If the 23 studies pooled by Xie et al. really represent a 10.9% D2T rate among all RA patients, and a 13.2% D2T rate among those who’ve ever been exposed to biologics, then the implied biologic-treated proportion = 10.9% / 13.2% = 83%. But the actual proportion of Americans who’ve ever used biologics to treat their RA is only 38%. The 83% vs. 38% gap reveals that the all-RA studies in the meta-analysis must be coming from populations where an above-average proportion of patients have used biologics, typically academic centers or highly established registries. The BRASS cohort (Paudel 2024, Rheumatology 64(3):1102) confirms this: its biologic-treated proportion is 65%, and 65% x 22.3% (D2T among treated) = 14.5%, matching its measured all-RA D2T rate of 14.4%. Internally consistent, but Paudel himself cautions that these results “may not be applicable to other patient populations.”
Step two: 520,000 biologic-treated x 13.2% = approximately 69,000 D2T patients. Here, the denominator (ever-used) and the rate (from the ever-used subgroup studies) are properly matched. The KOBIO biologic registry (Jung 2023, Arthritis Res Ther 25:174) independently reports 11.7% D2T among treated patients, in the same range as the meta-analysis 13.2%, reinforcing confidence in this estimate.
Step three: within the same meta-analysis by Xie et al., 5 studies (3,516 RA patients including 319 with D2T-RA) used musculoskeletal ultrasound with power Doppler imaging to separate out two categories of D2T patients. If power Doppler signal was detected (active blood flow in joint synovium = genuine ongoing inflammation), they called it PIRRA (persistent inflammatory refractory RA), at 47.1% (95% CI 33.3 to 61.4%). When there was no signal, the patients were labeled NIRRA (non-inflammatory refractory RA), at 52.9%. NIRRA patients’ elevated DAS28 scores reflect central pain sensitization and comorbid fibromyalgia inflating subjective pain components, not active joint inflammation. An independent 2024 ultrasound study in 85 patients validated a similar split: 56% PIRRA vs. 44% NIRRA. AlloNK’s mechanism is clearing B cells to eliminate inflammation. It has no rationale in NIRRA patients.
Eligible U.S. population: 520,000 x 13.2% x 47% ≈ 32,000 to 35,000. Roughly one-fifth of management’s figure.
From 35,000 patients to peak sales
Of those 35,000 eligible patients, roughly 5% per year would actually undergo treatment (benchmarked against CAR-T’s real-world penetration of 5.4% per SEER-Medicare data in oncology; AlloNK’s outpatient, off-the-shelf profile lowers the access barrier but RA is non-fatal, and the two forces roughly offset). That translates to approximately 1,750 patients per year. Net price per treatment course: approximately $105,000 (below CAR-T’s $300,000 to $500,000, above the ~$50,000 to $60,000 annual net cost of biologics; requires 18+ months of treatment-free durability to justify to payers). Ex-U.S. revenue from Artiva’s territory (excluding Asia-Pacific, which belongs to licensor GC Cell; primarily Europe) adds roughly 25%.
Global peak net sales: approximately $230 million, one-seventh of the $1.64 billion implied by management’s assumptions.
Platform optionality: narrative or substance
If RA succeeds, AlloNK could theoretically expand to other B-cell-driven autoimmune diseases: Sjogren’s disease (SjD), systemic sclerosis (SSc), lupus (SLE), and others.
Two external facts heavily constrain the value of these options.
Novartis has a drug called ianalumab which has reported positive results in two Phase 3 trials in SjD, received FDA Breakthrough Therapy Designation, and is expected to file for approval imminently. It is a once-monthly subcutaneous injection requiring no conditioning whatsoever. If approved in 2027, the vast majority of SjD patients will choose this option. The treatment burden is orders of magnitude lower than AlloNK. AlloNK’s positioning in SjD shrinks to a second-line niche of patients who fail ianalumab, a population that will take years to accumulate after ianalumab’s launch.
Hematopoietic stem cell transplantation (HSCT) in SSc has three positive randomized controlled trials behind it, the highest level of clinical evidence, and is formally recommended by both EBMT and EULAR. Yet in 2023, only 51 autologous HSCTs for SSc were performed globally. A therapy with Grade 1 evidence and guideline endorsement has a global annual volume of 51. Our original assumption of 10% annual penetration for AlloNK in SSc, implying 2,500 U.S. patients per year, was off by a factor of 50 relative to this reality.
After adjusting for competitive dynamics and real-world penetration benchmarks, total platform option expected value comes to approximately $225 million. The most robust component is label expansion within RA itself (same disease, same mechanism, broader patient eligibility). The two largest options, SjD and SSc, were significantly compressed by external competition and historical penetration data.
Conclusion
The probability-weighted expected value of the RA-only indication, at a 60% success probability, is approximately $370 million. Adding platform optionality of $225 million brings the total to roughly $595 million.
Current fully diluted market capitalization is approximately $550 million.
The market is pricing this name close to our independently verified expected value. The scientific bet underlying the pipeline (that Phase 3 will likely read out positive) is real. But the addressable market embedded in management’s narrative is overstated by roughly five-fold. This is not fraud; it is the kind of selective citation that biotech founders routinely engage in during capital raises. Investors need to run their own funnel to correct for it.
Disclaimer: This article is a personal research note and does not constitute investment advice. The author may or may not hold positions in the securities mentioned and may buy or sell at any time without notice. Biotechnology investing carries extreme risk; clinical trial outcomes are highly uncertain; and all probability estimates and valuation calculations herein are based on public information and subjective judgment that may contain material errors. Readers should conduct their own due diligence and consult a licensed professional before making any investment decisions. No sell-side analyst views were cited in this analysis.