AI Moves From Pilot to Production as Q1 Deal Wave Continues

• ADCs remain the single most-traded modality: within the window, Lilly (CrossBridge) and the still-active Gilead/Tubulis transaction both center on ADC chemistry.
• Radiopharma becomes a board-level question: the Regeneron × Telix tie-up is notable because a top-5 antibody house is now signaling that radioligand manufacturing is a capability it will partner for rather than build.
• China-to-West licensing continues to set the pace: GSK–Frontier ($40M up / $963M total) and AbbVie–Haisco ($30M up / $745M total) fit the Q1 2026 data showing average deal size up ~76% vs. 2025, led by Chinese-originated assets.
• Upfront compression is structural, not transient: across the four headline licensing deals of the week, upfronts hovered in the $30–40M range against total biobuck packages of $700M–$1B+ — consistent with the ~6% upfront ratio seen industry-wide in Q1 2026.
• What to watch next week: obesity/GLP-1 licensing (still the single highest-value therapeutic category for 2026), further Big Tech / AI-biotech crossovers following Anthropic's ~$400M all-stock acquisition of Coefficient Bio earlier this month, and potential mid-cap consolidation into the back half of the 2025–2030 patent cliff.

The strongest deployment is concentrated in early discovery, where the data is clean, workflows are computationally tractable, and the validation loop is short.

• Target identification — generative models running in-silico perturbation across cell-state foundation models are now standard at large pharma. Novartis recently published work using AI-driven gene-knockdown simulation to identify novel targets in autosomal dominant polycystic kidney disease, a workflow that would otherwise have been prohibitively slow.
• Hit and lead discovery — protein-structure prediction is used by 71% of biotech R&D leaders in Benchling's 2026 Biotech AI Report, with literature review at 76% and target identification at 58%. The OpenFold3 federated network, which pools proprietary protein-ligand structure data across multiple pharma, is fine-tuning structure prediction toward precision approaching X-ray crystallography.
• Lead optimization — the Recursion–Exscientia merger combined high-throughput phenomic imaging with automated molecular design, compressing typical optimization cycles from months to weeks. Phase I success rates for AI-discovered candidates are running 80–90% versus a historical 40–65% baseline, based on a still-small sample of completed trials. Early data, but consistently better.

The pattern is consistent: AI is most mature where inputs are well-structured (sequence, structure, omics data) and where computational predictions can be validated quickly against wet-lab results.

Three areas matter most for BD strategy over the next 18 months:

Clinical and translational layers

Discovery has been the headline story; clinical and translational work is where the cost is. Two complementary patterns are emerging. The first is foundation-model licensing: GSK's $50M licensing of Noetik's OCTO-VC virtual cell models for tumor biology in NSCLC and colorectal cancer (January 2026) signals a move from project-based AI collaborations to AI infrastructure as enterprise assets. The second is AI-enabled trial design: Unlearn.ai's PROCOVA digital-twin methodology is the first AI method to receive a formal EMA qualification opinion, and has demonstrated 10–20% variance reduction in treatment-effect estimates — translating to roughly 20–30% smaller control arms in Phase 2 and 3 trials without sacrificing statistical power. This second category is where the most measurable, near-term ROI is.

Phase II and III readouts

The AI-discovery class has crossed into the clinic at scale: roughly 175 AI-originated drugs are in development, with a growing number entering pivotal trials in 2026. Phase II data through 2025–2027 is the decisive test for whether AI-designed molecules are merely faster-to-IND or genuinely better. A successful readout from any of the leading platforms (Insilico, Recursion, Insitro, Schrödinger) will reset valuation expectations across the entire AI-pharma deal book.

Manufacturing and the regulatory layer

AI in biomanufacturing remains the most underdeveloped area — process control, batch optimization, and predictive QA are still mostly classical methods. The FDA's January 2025 draft AI guidance is still pending finalization, and in January 2026 the FDA and EMA jointly issued 10 high-level guiding principles signaling where the final framework is heading. Separately, the EU AI Act's high-risk provisions take effect August 2, 2026. Both will increase compliance work for sponsors using AI in regulatory submissions, but they will also clarify how AI-derived evidence can be defended — which should accelerate adoption rather than slow it.

The first generation of AI deals (2016–2020) was evaluation-driven and structured around single assets. The current generation, exemplified by Insilico–Sanofi ($1.2B for six targets) and Isomorphic–Lilly–Novartis (close to $3B in combined biobucks for AI-powered small-molecule discovery across multiple targets), is platform-driven and structured around multi-program, multi-year capability access. This is the structural change to plan around.

For BD teams the immediate implications are threefold: (i) "AI-enabled capability" clauses are rapidly becoming standard in licensing term sheets and should be expected in any 2026 deal; (ii) platform IP — datasets, workflow integrations, training methods — is more legally defensible than composition-of-matter patents on AI-generated molecules, given the unresolved inventorship questions across major patent jurisdictions; and (iii) the hottest hiring profile is now domain expertise plus ML fluency, with hiring managers reporting 70%+ difficulty filling these combined roles.

The takeaway is straightforward: AI is no longer a thesis to test; it is a layer being built into how pipelines, deals, and capabilities are structured. Companies that have moved into operational integration are pulling ahead. Those still in evaluation mode will find themselves negotiating from a weaker position by 2027.

For readers with assets or intros that match the mandates above, the following cross-reference summarizes what Biolink can route directly to the relevant buyer or fund.

A China-based biotech is seeking global partners for a first-in-class (FIC) immunotherapy platform targeting autoimmune diseases. The platform is built on a proprietary antigen-specific tolerance technology designed to modulate immune response without systemic immunosuppression — a mechanism that, if validated, would directly address the central limitation of currently marketed biologics in this space.

Why this asset matches the board

The lead asset is Phase II, and the platform produces multiple FIC programs in autoimmune disease — squarely within the autoimmune mandate from Western buyers. For the geographic-arbitrage fund model the Phase II GD program could also serve as a candidate for a Western Phase II/III re-read with fund capital, particularly given the clean Phase I safety profile. For Newco discussions, the breadth of indications (GD, TED, MS, T1D) plus commercial-stage validation of the parent company make this a platform-level play rather than a single-asset play.

• Featured license-out (autoimmune platform) — BD@biorichinc.com
• Asset submissions for Western buyers or fund mandates — route via Biolink Weekly BD inbox (include modality, stage, last clinical readout, territory availability).
• Role cross-reference — see the Job Board for BD&L talent searches that may pair with these mandates (e.g., Heads of BD&L, VP BD, licensing counsel).

Biolink Weekly — Section 4, BD&L Opportunity Board. Prepared April 27, 2026. Buyer and fund mandates are summarized from direct briefings; details may evolve prior to execution. Biolink does not represent any named party unless explicitly stated.