$ABVX: funded into 2029, which changes what it's worth

$ABVX: 13F filings show specialist biotech funds are maxed out, not exiting

Development · August 21, 2026 · written at $118.19

Q2 13F data shows ABVX as a top-3 position for most specialist biotech funds holding it, several ranking it #1. That's a sign of high conviction, but it also means many of these holders may already be at their position limit, one reason for a lack of fresh buying pressure lately.

What this note is about

Second-quarter 13F filings (institutional ownership disclosures) show a pattern worth noting: among specialist biotech-focused funds that hold ABVX, it isn't a minor position for most of them, it's a top-3 holding, and for several it's their single largest position. Filings like this only show quarterly, and don't capture what happened after the July 6 equity raise, so anything since Q3 isn't reflected yet.

What the filings show

Funds like TCG Crossover, ADAR1 Capital, Artia Global Partners, Great Point Partners, Octagon Capital, Diadema Partners, and Ikarian Capital all rank ABVX as their #1 position. A second group, including Darwin Global Management, Caligan Partners, Helicon Capital, Logos Global Management, and Eversept Partners, ranks it #2. A third group has it at #3.

Why this matters both ways

The bull read: this many specialist funds treating ABVX as a top conviction position is a real signal that people who focus specifically on biotech, and study these theses closely, believe there's meaningfully more upside from here.

The complicating factor: if a fund already has ABVX as its #1 or #2 position, it's likely near or at its own internal concentration limit, meaning there's less room for that same fund to keep buying more, even if it wants to. That may help explain why the stock hasn't found the kind of sustained buying support the fundamental story alone might suggest. Since the equity raise closed in Q3 and won't show up in filings until the next 13F cycle, existing specialist holders may have increased their stakes further using the offering, without any new specialist buyers entering yet.

What this means

This isn't a fundamental data point like a trial readout, it's a positioning signal. High specialist conviction is a genuine vote of confidence in the underlying thesis, but a fund base that's already maxed out limits how much incremental buying pressure is left in the near term from that specific investor group. Worth watching whether new specialist names show up in the next 13F cycle, or whether this remains a concentrated, fully-allocated shareholder base.

$ABVX: the cancer signal isn't real, and the math already proves it

Update note · June 2, 2026 · written at $76.16 · confirmed

Obefazimod posted the best maintenance efficacy this disease has seen and the stock fell 44% on a cancer signal instead. The "7 cases" is really 2. On cases per 100 patient-years, obefazimod sits at background rate, and there is no mechanism by which it could cause cancer.

What happened

Abivax reported 44-week ABTECT maintenance data yesterday. Clinical remission deltas of 40% and 39% beat Rinvoq, the efficacy benchmark in ulcerative colitis, and the endoscopic remission delta came in at more than double Rinvoq's. A once-daily oral pill, with no lab monitoring requirement, outperforming a broadly immunosuppressive JAK inhibitor on the endpoint GI doctors care most about.

The stock briefly traded to $175 on that. Then it collapsed to the low $70s as the market decided there was a cancer signal in the safety table.

There isn't one. This note lays out why, using the data that is already public.

The "7 cases" is really 2

The number circulating is seven malignancies in the 50mg arm. Break it down:

One is not cancer. Abivax listed a case of colonic dysplasia and, in their own footnote, called it "colon cancer." Colonic dysplasia is a precancerous polyp. It is extremely common in UC patients, who have chronic gut inflammation and get frequent colonoscopies, which means it is both more likely to occur and more likely to be found. Other UC trials do not even report their dysplasia rates, because it is considered unrelated to treatment. Including it was a mistake. Calling it cancer was a serious one, and it inflated the apparent malignancy count by 50%.

Four are non-melanoma skin cancer. NMSC is a fundamentally different category. It is more than 15 times more common than breast or prostate cancer yet does not appear on lists of common cancers, because it is so rarely dangerous. It is typically cured by a sub-minute in-office procedure. The FDA separates it out so deliberately that it uses an endpoint called non-non-melanoma skin cancer. And the decisive point: no drug in history has received a black box warning for causing NMSC alone. Several drugs that unambiguously do cause NMSC have not been black-boxed for it.

That leaves two. One breast, one prostate. The two most common cancers in the world, in patients aged 65 and 53, both within routine screening age for exactly those cancers. The prostate patient had been on six prior UC therapies, of which 7 of the 11 approved advanced IBD drugs carry either a black box or a warnings-and-precautions note for malignancy.

The metric everyone used is the wrong one

Cancer risk is not assessed by counting events. It is assessed as cases per 100 patient-years, because it matters how long patients were observed, not just how many there were.

If you take only the Phase 3 50mg arm and include the dysplasia case, you get roughly 1.88 per 100 patient-years, which looks alarming. That is the number the market appears to have run.

Include the long-term Phase 2 database, which is actually the larger safety dataset in patient-years, and drop the case that isn't cancer, and the 50mg rate is approximately 0.59 per 100 patient-years.

Context for that number:

• UC background: ~0.5 → n/a
• Omvoh: ~0.6 → no cancer warning at all
• Rinvoq: ~0.6 → black box warning
• Obefazimod: ~0.59 → TBD

Omvoh and Rinvoq have effectively the same rate and opposite labels. That looks like a contradiction until you know why Rinvoq was black-boxed: it inherited a class-wide JAK inhibitor warning issued in September 2021, triggered by a different drug (tofacitinib) in a different disease (rheumatoid arthritis). Rinvoq's own UC malignancy data would never have produced that label on its own.

Obefazimod is the only drug in its class and is only being studied in IBD. There is no other dataset that could contaminate it the way Xeljanz contaminated Rinvoq.

The placebo comparison is broken

People are comparing 4 NMSC cases in the 50mg arm to 1 in placebo. That comparison does not work, because the two arms were not observed for the same amount of time.

Only 34% of placebo patients completed the study, against 82% on 50mg, because placebo patients relapsed and dropped out. Once a patient leaves a trial they stop being surveilled for adverse events. The drug arm therefore accumulated roughly twice the patient-years of monitoring despite an identical headcount.

You should expect a higher raw event count in the drug arm even if the drug carries zero excess risk. That is not a signal, it is arithmetic.

There is no mechanism

Drugs cause cancer in two accepted ways: genotoxicity, or immunosuppression. Obefazimod shows neither.

Genotoxicity was ruled out years ago in preclinical work, with negative Ames testing for both the parent compound and its active metabolite, a negative in vivo micronucleus assay, and clean three-year rat carcinogenicity studies.

Immunosuppression is ruled out by the Phase 3 itself. Overall infection rates were low, and the severe infection rate was lower on the high dose than on placebo. Compare that to JAK inhibitors, which carry a separate black box for serious infections precisely because they are immunosuppressive, which is almost certainly the source of their malignancy signal too.

The FDA weighs mechanism heavily in labelling decisions, because malignancies are rare events and small trials are noisy. Obefazimod fails every test for plausible carcinogenicity.

What a real signal looks like

Two comparisons make the contrast clear.

OX40 and Kaposi sarcoma. Five cases across two OX40 programmes, roughly 0.13 per 100 patient-years. That sounds low until you know the expected background in HIV-negative adults is around 0.00003. That is a rate more than 4,000 times above expectation, in a cancer type that was genetically predicted from OX40 loss-of-function biology, and clustered in a single rare tumour type. That is a signal.

Ocrevus and breast cancer. Six breast cancers on drug versus zero on comparator, clustered in one tumour type, roughly 3x the expected rate. Ocrevus still received no black box, only a warnings-and-precautions note, and a subsequent real-world study across ~100,000 patient-years found no elevated incidence at all. A textbook example of a phase 3 signal that looked real and wasn't.

Obefazimod has no clustering. A meningioma, a prostate cancer, a breast cancer: three different tumour types, two of which are the most common cancers there are. Clustering is what carcinogenesis looks like. Scatter is what background looks like.

What actually matters for the label

Only a black box would meaningfully damage the launch. A malignancy mention in warnings-and-precautions would not, because prescribers largely discount that section. Skyrizi, widely considered one of the safest advanced therapies available, lists hepatotoxicity and tuberculosis under W&P, neither of which occurs at above-background rates in reality.

On current data and existing precedent, a black box is not a realistic outcome.

🐻 What would change my mind

• Part 2 data showing a materially higher rate. On the assumption of roughly 550 patients in part 2 at background rate, you should expect around 3 new non-NMSC cases. Three is normal. Substantially more would be a genuine problem.
• Clustering. If new cases concentrate in one tumour type, the scatter argument collapses.
• A shorter latency pattern than expected, especially if the two existing cases turn out to have had no recent negative screening before enrollment, which would make pre-existing disease more likely and further weaken the signal, but the reverse pattern in new cases would count against me.
• An FDA advisory committee being convened on the safety database, which would signal the agency sees something the precedent base does not support.

Conclusion & My Position

My angle: the market saw seven cancers where there were two, ran the arithmetic on one arm of one trial instead of the full safety database, and compared against a placebo group that was observed for half as long. Every one of those errors is correctable with public information.

I expect the part 2 update this month to show new malignancy cases, because statistically it should, and I expect the stock to rise anyway once management finally presents them in context. If they clear the air properly, I see no structural reason the stock cannot return to the $150 to $180 range it was heading for on the efficacy data alone.

This is a deliberately falsifiable call. If the part 2 rate lands meaningfully above background, or the cases cluster, I was wrong and I'll say so.

Position at publication: long ABVX, added on the drawdown.