AbbVie AI Drug Discovery Push Adds Iambic, Valkai and AISB Bind

Three AI moves in nine days

In late September 2026, AbbVie made three separate artificial intelligence announcements in nine days. Each one covers a different stage of drug development, and together they show the company working on discovery chemistry, clinical operations and shared industry infrastructure all at once.

The first came on September 21. AbbVie and San Diego-based Iambic announced a multi-year collaboration to discover small-molecule drugs in immunology, neuroscience and oncology using Iambic's AI platform.1 The next day, the AI Structural Biology (AISB) Network launched AISB Bind, a federated project in which AbbVie, AstraZeneca, Bristol Myers Squibb and Johnson & Johnson will jointly train a model to predict how strongly small molecules bind to their protein targets.17 On September 29, AbbVie added a partnership with Valkai, a life sciences AI platform, to apply purpose-built AI tools in selected parts of its clinical research.21

AbbVie places all of this under AI@AbbVie, which it describes as its company-wide approach to using AI across R&D, operations and other core functions.1 Read together, the three deals show a large drugmaker that is not betting on one model or one vendor. It is buying, renting and co-building different tools at each step of the pipeline.

The Iambic deal: optimizing many properties at once

The main announcement is the Iambic collaboration. It relies on two of Iambic's technologies, Enchant and NeuralPLexer. Iambic says they are built to optimize many properties of a drug candidate at the same time, instead of the usual approach of assessing each property one by one.1 The deal will specifically use Enchant v3, a multimodal transformer that Iambic says was trained on more than 6,000 molecular properties and can test many design hypotheses in parallel.1

NeuralPLexer is the part most relevant to protein structure prediction. Trade coverage describes it as a model that predicts the 3D structures of proteins bound to drug-like molecules, which supports structure-informed design.2 Another description says it combines physics with machine learning to forecast those protein-ligand structures.35 According to one report, no existing AbbVie programs are moving to Iambic. AbbVie's discovery scientists will use the tools to create new candidates.2

The financial terms are where reports disagree. AbbVie's announcement says only that Iambic gets an upfront payment, success-based milestones and tiered royalties on net sales.1 Several outlets said the amounts were not disclosed.28 Insider Monkey went further, saying AbbVie disclosed neither the upfront payment nor the milestone values.10 Yet a separate Insider Monkey piece published through Yahoo Finance reported a $40 million upfront payment, up to $670 million in milestones across two collaboration targets, and royalties from the mid-single digits to the low teens.38 The likeliest explanation is that those numbers came from Iambic's IPO paperwork rather than the joint press release. Still, readers should treat them as reported, not confirmed by AbbVie. If they are accurate, AbbVie is making a modest bet on two targets, not a broad platform deal.

A pharma partner just before an IPO

The timing points to a funding story as much as a science story. Iambic filed for a Nasdaq listing under the ticker IAM on the same day the AbbVie deal was announced. At that point it had raised about $461.8 million from investors including Nvidia, Coatue and Q Healthcare Holding, which is linked to the Qatar Investment Authority.33 It priced its IPO on October 14 at $16 a share, the midpoint of its range, selling about 9.38 million shares to raise $150 million.32 The stock rose 16% on its first day of trading, reaching $18.50.34

The AbbVie deal makes Iambic's list of big-pharma partners considerably longer. Takeda signed a deal in February 2026 that could be worth more than $1.7 billion in success-based payments, and Bayer signed one in June 2026 focused on hard-to-drug targets.2 Earlier agreements with Lundbeck and Jazz Pharmaceuticals date to 2024.2 For a company heading to the public markets, AbbVie's name helps justify its valuation. Valuation figures vary with the date and price used: $758 million at the IPO price32 versus $877 million at the first-day trading price.31

The fundamentals call for some caution. Iambic reported a net loss of $50.1 million on $12.8 million of revenue in the first half of 2026.31 It plans to spend $75 million of the proceeds on its lead HER2 drug, IAM1363, and $20 million on its AI platform.31 BioPharma Dive data cited in coverage notes that several AI-based drug companies have struggled after going public.31 Capital is still flowing into the sector. In the same week, AstraZeneca and insitro announced a $1.5 billion AI drug discovery partnership, and OrbiMed raised $3 billion in new funds.36 My reading is that Iambic's listing tests whether partnership revenue and early clinical data can support AI-biotech valuations better than the platform-only pitches of earlier years.

The industry initiative: competitors sharing data without sharing it

The less visible part of AbbVie's activity may matter more. AbbVie and Johnson & Johnson founded the AISB effort in March 2025. Using Apheris's federated infrastructure, they set out to fine-tune OpenFold3, the open-source structure prediction system from Mohammed AlQuraishi's lab at Columbia University, on their proprietary structural data.14 The reason is that public data has a gap. Public databases contain plenty of protein structures but relatively few examples of proteins bound to drug-like molecules, which are the cases that matter most for drug discovery.11

Bristol Myers Squibb, Takeda and Astex joined in October 2025.20 By September 2026 the group reported results. The five companies jointly fine-tuned a model called AISB-1-Fed on 20,167 proprietary protein-ligand structures, and each company trained it locally so that no structures were pooled.15 On a held-out set of 1,056 private structures, the share of high-quality protein-ligand interface predictions rose from 35.6% for the public OpenFold3 Preview 2 model to 52.1%, and correct ligand poses rose from 28.9% to 46.8%.15 AbbVie alone reportedly contributed more than 9,000 structures. Coverage also described a large, unplanned improvement in predicting protein-protein interfaces.12 One summary said the federated model clearly beat Boltz-2, a leading public model.16

The coverage does not agree on what the results mean. Tech Times presented the result as a decisive win over every public co-folding model.12 BreezyScroll was more careful. It noted that the findings had not been peer-reviewed when Nature reported them, that outsiders cannot reproduce the work because the structures and model weights are private, and that the gains apply to structure prediction, not binding affinity.15 The cautious reading is the more accurate one. The roughly 16-point jump is a real result, but it is a measurement the consortium made on its own held-out data.

That limitation is what AISB Bind is meant to address. The new project goes after binding prediction directly, aiming for accuracy high enough to guide virtual screening and lead optimization.17 This explains the timing. AbbVie is helping build a shared, more accurate foundation model while also paying Iambic for specialized tools on top of it. The strategy keeps the underlying models a shared industry resource while the company competes on its own targets and molecules.

Clinical development and the limits of the claims

The Valkai deal applies the same strategy later in the pipeline. AbbVie says the platform will help clinical teams produce insights faster, spend less time on manual work and potentially shorten development timelines.21 One analysis noted that AbbVie presents these as potential benefits, not measured results.25 Coverage reported that Valkai's team comes from OpenAI, Google DeepMind and other applied-AI companies.30

The financial context explains why AbbVie can make these bets now. Second-quarter 2026 revenue was $16.99 billion, and Skyrizi and Rinvoq together brought in $8.03 billion while Humira sales kept falling.10 AbbVie has a strong core business and patent expirations ahead, so improving R&D productivity is a long-term goal, not an immediate source of earnings.

A note on scope: the deals cover AI-driven small-molecule design and structure prediction. None of the announcements involves CRISPR gene editing, and none uses generative protein design to create new biologics. AbbVie's earlier work with BigHat Biosciences on AI-designed antibodies is the closest it comes to that.13

The takeaway

What AbbVie announced in September is not a breakthrough. It is a set of tools: a specialized design platform, a shared training system for foundation models, and an operations layer for clinical work. The federated results suggest that the industry's private data, combined carefully, can improve open models in ways public data cannot. Whether any of this leads to better medicines is still unproven. As several analysts noted, AI can help choose which molecules to pursue, but the biological and clinical risks that cause most programs to fail remain.108