Scientari · Industry Intelligence · Updated July 2026
A living map of AI-native biotech.
Companies whose core methodology is learned models — neural networks, foundation models, generative AI — applied to drug discovery, diagnostics, and clinical trial design. Editorially curated, not comprehensive. How we classify →
—
Companies tracked
—
Pharma deals captured
—
Clinical assets
Scientari · Industry Intelligence
The AI-Native Biotech Tracker
Text size
Recent Headlines
ScionxBio launches from Eleven Therapeutics — xRNA spun out for cardio-metabolic disease. July 1, 2026. Eleven Therapeutics rebrands as ScionxBio (per co-founder Yaniv Erlich), a preclinical RNA-therapeutics company operating independently since January 2026. ScionxBio takes the xRNA modality (chemically-modified, long-acting mRNA) and the TERÅ platform — combinatorial chemistry + deep learning to engineer RNA with sustained protein expression — and focuses it exclusively on cardio-metabolic disease. The move follows a strategic split: Eleven's two platforms proved suited to different modalities, so rather than merge them the company is spinning each into its own vehicle. ScionxBio is the first; a second spin-off (built around Eleven's other platform) is in formation and unannounced. Dr. Adi Gilboa-Geffen (Eleven's former CTO) is CEO; co-founder Dr. Yaniv Erlich becomes Chairman; co-founder Shaul Ilan becomes COO.
Why it matters: A clean example of the AI-native RNA thesis maturing into focus: rather than spreading a platform across every modality, Eleven split its two technologies into single-purpose companies — ScionxBio for long-acting xRNA, plus a second spin-off still to come. Long-acting RNA for cardio-metabolic disease targets the same commercial territory as the GLP-1 boom — infrequent dosing is the battleground — so an AI/combinatorial-chemistry platform optimizing RNA durability is aimed squarely at a proven market. Tracker note: ScionxBio enters as the active company; Eleven moves to Legacy as the retired parent brand. Two things to watch: a dedicated ScionxBio financing round (it launched on Eleven's ~$22M seed base with no new raise disclosed), and the identity of the second spin-off — likely built on Eleven's delivery/RNAi work tied to the Novo Nordisk partnership.
Anthropic launches Claude Science — and starts its own neglected-disease drug programs. June 30, 2026. A standalone AI workbench for researchers, elevated to peer status with Claude Code and Claude Cowork; now in beta for Pro/Max/Team/Enterprise on macOS and Linux. Not a new model — runs on existing Claude models (incl. Opus 4.8). 60+ curated skills/connectors across genomics, single-cell, proteomics, structural biology, cheminformatics; native NVIDIA BioNeMo Agent Toolkit access to Evo 2, Boltz-2, OpenFold3; a coordinating agent dispatches sub-agents while a separate reviewer agent checks every citation and calculation. Alongside the launch, Anthropic said it will pursue its own preclinical drug discovery targeting neglected diseases — building on its ~$400M Coefficient Bio acquisition (April 2026) and the John Jumper (AlphaFold) hire. Tracker company Manifold Bio is a named beta user, running end-to-end target nomination against its proprietary data.
Why it matters: Notable precisely because of where this tracker draws its line. Per our criteria, Big Tech AI labs are tracked through their spinouts and partnerships, not as company entries — so Anthropic stays represented via the Owkin × Anthropic infrastructure (and the resulting Sanofi K Pro deployment) and, increasingly, as substrate the way NVIDIA is, rather than as a Companies row. But a frontier lab standing up its own internal pipeline is a genuine category shift: if the Coefficient Bio work yields a spinout or a named clinical asset, it would graduate to a tracked entity under our spinout rule. Second thread: Claude Science productizes the “foundation model as pharma substrate” pattern the Owkin integration previewed in Jan 2026 — now available to every paying researcher rather than a single pharma partner. Watch whether the reviewer-agent provenance model (code + environment + full history attached to every figure) becomes a reproducibility bar other AI-bio platforms have to match.
METiS TechBio × Boulevard Bio — ~$1.62B for an AI-designed trispecific T-cell engager. June 30, 2026. Deerfield-backed Boulevard Bio licenses exclusive worldwide rights to MTS-128, a trispecific TCE designed end-to-end on METiS's NanoForge platform; $20M upfront + up to $1.6B milestones + tiered royalties. Largest disclosed preclinical trispecific TCE out-license on record. METiS (HKEX:7666) is the first listed AI-powered drug-delivery company, having IPO'd just 47 days earlier (~$269M, May 2026). Targets and preclinical data undisclosed.
Why it matters: Two firsts in one deal. First, it's the debut major out-license from a newly public AI-delivery company — a proof point that HKEX-listed TechBio can command premium preclinical biobucks from US capital. Second, the modality is notable: most AI drug design to date has centered on small molecules and standard biologics; a fully AI-engineered trispecific TCE pushes learned-model design into more complex multi-target architectures. The heavy backloading ($20M against $1.62B topline) fits the now-familiar AI-bio pattern where headline numbers buy reputation more than near-term cash — watch for MTS-128 preclinical data to substantiate or deflate the valuation.
Lilly TuneLab 2.0 launches with multi-task foundation models — Lilly replaces the original single-task models with two new architectures: ChemLab (33 small-molecule ADMET endpoints predicted simultaneously, multi-task transformer) and AbLab (18 antibody developability endpoints from sequence). Federated training is underway on 90+ contributed datasets; federated ChemLab release targeted for July 2026. Tamarind Bio joins as scaled inferencing partner; Rhino Federated Computing continues operating the federated learning environment. New: model fine-tuning and API-based inferencing for embedding into partner workflows. ~70+ partner companies onboarded since September 2025 launch, targeting 150 by year-end.
Why it matters: The most-developed example of pharma-native federated AI in production. Lilly bears the infrastructure cost, partner biotechs contribute private data without disclosure, all benefit from a stronger model — a flywheel pattern that didn't exist as a working product 18 months ago. If the July federated ChemLab release outperforms internal single-pharma ADMET models on held-out validation sets, it raises a structural question for external AI-bio platforms: the data moat may now be federated, and ADMET specifically may be moving from competitive differentiator to precompetitive infrastructure. Watch for Novartis, AZ, or Pfizer to respond with their own federated rails.
Owkin × Sanofi — 5-year K Pro license + multi-year AI agents collaboration. First major productization of the Owkin × Anthropic agentic infrastructure announced Jan 2026. Builds on existing €90M Sanofi partnership (2021).
Why it matters: Validates K Pro as a productized AI scientist that top-10 pharma will license at scale, not just collaborate around. The first major customer of the Owkin × Anthropic infrastructure stack — early proof the foundation-model-as-pharma-substrate pattern works commercially.
Chai Discovery × Pfizer — AI drug discovery license. Second top-10 pharma in 6 months (after Lilly Jan 2026). Chai is now a $1.3B AI-bio unicorn ($230M cumulative). Board includes ex-Pfizer CSO Mikael Dolsten.
Why it matters: Confirms the “year of deployment” thesis Endpoints called in January. Two top-10 pharma deals in six months plus a $1.3B unicorn valuation signal foundation-model AI biotech crossing from speculative to operational. The pattern matters more than the single deal.
DIOSynVax (Cambridge spinout) — first-in-human readout for AI-designed vaccine. Pan-Sarbeco coronavirus 'super-antigen' published in Journal of Infection. 39 volunteers, safe, modest immune response across SARS-CoV-2, SARS, and bat coronaviruses. First vaccine antigen designed entirely by computer simulation tested in humans.
Why it matters: First peer-reviewed proof that a fully AI-designed vaccine antigen clears the first-in-human safety bar. Joins Insilico's Rentosertib (small molecule, Jun 2025) and Generate's GB-0669 (mAb, Oct 2025) as parallel AI-biologic FIH firsts — three modalities, three independent validations within 18 months. The category threshold is real.
FDA adopts ICH M15: General Principles for Model-Informed Drug Development — first internationally harmonized framework explicitly governing AI/ML evidence in drug regulatory submissions. Model risk stratification, standardized assessment tables, Model Analysis Plans.
Why it matters: Regulatory infrastructure now explicitly accommodates AI/ML evidence in drug submissions. Removes a persistent uncertainty about how AI-derived inputs would be evaluated — and gives Insilico, Unlearn, and peer companies a shared framework to file against.
MSU lab publishes GPS in Cell — deep learning predicts gene expression changes from chemical structure alone. Screened 7M+ compounds across two diseases. Discovered a novel HCC lead with sub-micromolar potency and in vivo efficacy; identified an anti-fibrotic IPF compound validated in patient-derived precision-cut lung slices. Code and web portal open-sourced.
Why it matters: Two AI-discovered IPF compounds within nine months — Insilico's Rentosertib (foundation-model approach, Jun 2025) and now GPS (transcriptomic-reversal approach, Mar 2026), validated in different ways and from different methodological lineages. The convergence strengthens IPF as AI-bio's proving ground. The open-source code + web portal continues a pattern of meaningful AI-bio methodology emerging from academia with code released — a structural counterforce to closed commercial platforms.
CZ Biohub unveils ESMFold2 + open ESM Atlas (1B+ proteins) — AI "world model" of protein biology claimed to surpass AlphaFold3. Open-source counterweight to closed commercial AI-bio platforms. Built by Alex Rives's team (formerly EvolutionaryScale, now Biohub).
Why it matters: Open-source counterweight to closed commercial protein foundation models. With Alex Rives leading and $500M behind it, Biohub becomes the open infrastructure for the next decade of biology AI — a structural counterforce to Isomorphic's closed-platform model.
Isomorphic Labs Series B — $2.1B (Thrive Capital-led; sovereign AI funds participate). Largest AI-bio equity round to date. First clinical trials by end-2026 (delayed one year from original target).
Why it matters: Largest AI-bio equity round to date, raised with sovereign-AI fund participation. AI drug discovery is now treated as national-strategic infrastructure, not just biotech VC. The closed-platform thesis is well-funded for the long haul.
Gain Therapeutics — positive Phase 1b interim data for AI-discovered allosteric Parkinson's drug. GT-02287 (oral, brain-penetrant GCase modulator) showed biomarker evidence for disease-modifying activity in patients with idiopathic and GBA1-mutation Parkinson's: reduced plasma neurofilament light chain (NfL), reduced α-synuclein aggregation markers, and changes consistent with restored lysosomal/mitochondrial function. Phase 1b extension to 12 months underway. NASDAQ:GANX. Magellan™ allosteric platform.
Why it matters: Two distinct firsts. First, the binding site itself was AI-discovered — most AI drug discovery targets known active sites; allosteric sites have been historically computational-light. Second, the readout went beyond safety/tolerability to biomarker evidence of disease modification at Phase 1b — rare for an early-stage neuro program. Joins Insilico Rentosertib (Phase 2a IPF, Jun 2025) and Generate GB-0669 (Phase 1 FIH, Oct 2025) as third independent clinical-progression validation for AI-derived therapeutics, now spanning three modalities: foundation-model small molecule, ML-designed biologic, AI-discovered allosteric small molecule.
Insilico — Rentosertib Phase 2a in Nature Medicine — first published clinical proof-of-concept for an AI-discovered + AI-designed drug. TNIK inhibitor for IPF.
Why it matters: The single most important precedent for the field's central claim. First published clinical efficacy signal for a drug both discovered and designed by AI — moves Rentosertib from validation case to category proof point.
Profluent × Eli Lilly — up to $2.25B milestones + royalties for AI-designed recombinases enabling kilobase-scale DNA editing. "Holy grail" gene editing meets foundation models.
Why it matters: $2.25B for AI-designed recombinases extends the AI-discovery pattern from small molecules and biologics into kilobase-scale DNA editing. Lilly's third major AI-bio deal of 2026 — they're systematically building exposure across modalities.
Iambic × Takeda — multi-year AI discovery partnership worth up to $1.7B+, including NeuralPLexer model access for Takeda.
Why it matters: NeuralPLexer model access embedded into Takeda's discovery workflow. Model-as-product becomes a real commercial pattern beyond compound licensing — pharma is buying the AI itself, not just its outputs.
Roche acquires PathAI for $1B — landmark exit for AI digital pathology.
Why it matters: First $1B+ exit for an AI digital pathology company. Establishes the playbook for diagnostics → big-pharma acquisition in the AI-bio category. Other AI pathology players now have a comparable to point to.
Novo Nordisk acquires Cellular Intelligence — STEM-PD Parkinson's program + AI cell-therapy platform.
Why it matters: Big pharma acquiring AI-bio for a therapeutic asset (STEM-PD Parkinson's program), not for platform IP. Different exit pattern than PathAI — substance over infrastructure. Signals two distinct acquisition theses now coexist.
Owkin × Anthropic — agentic biology infrastructure integrated with Claude for Healthcare & Life Sciences.
Why it matters: First major LLM-foundation-model integration into a biology-AI platform. Sets the technical and commercial stage for the Sanofi K Pro deployment 5 months later. The partnership architecture is now visible: foundation model + domain data + pharma customer.
Where AI actually compresses drug development
Click any stage to filter companies below ↓
Traditional
~14 years
Discovery
~3.5 yrs
Preclinical
~3 yrs
Clinical Phase 1 – 3
~6.5 yrs
FDA
~1 yr
AI-enhanced
~10 years
Discovery
~1.5 yrs
Preclinical
~1.5 yrs
Clinical Phase 1 – 3
~6 yrs
FDA
~1 yr
Validated case: Insilico Medicine's Rentosertib (ISM001-055) reached preclinical candidate in 18 months at a budget of ~$2.7M — about one-third the time and one-tenth the cost of conventional discovery, per Zhavoronkov (Bloomberg, Nov 2023). Phase 2a clinical proof-of-concept for IPF published in Nature Medicine, June 2025. The first AI-discovered and AI-designed drug with published clinical PoC.
Clinical compression is modest but real. McKinsey 2025 finds AI-driven site selection accelerates enrollment by 10–15%, NIH's TrialGPT cuts patient screening time by 40% at same accuracy, and AI-prevented protocol amendments save ~260 days per trial on average. Together these translate to roughly 8–12% end-to-end Phase 1–3 compression — meaningful but bounded by biology, which dictates treatment and follow-up duration. FDA review timelines have not yet moved.
Clinical compression is modest but real. McKinsey 2025 finds AI-driven site selection accelerates enrollment by 10–15%, NIH's TrialGPT cuts patient screening time by 40% at same accuracy, and AI-prevented protocol amendments save ~260 days per trial on average. Together these translate to roughly 8–12% end-to-end Phase 1–3 compression — meaningful but bounded by biology, which dictates treatment and follow-up duration. FDA review timelines have not yet moved.
Engagement key
Internal pipeline
Pharma partnerships
Self-serve / SaaS
Open-source models
Region
Type
Focus
| Company | Type | Modality / Focus | Region | Stage | Funding |
|---|
Collective view of clinical-stage assets developed by AI-native biotechs. Programs are assigned to their most advanced phase. Preclinical includes lead-optimization and IND-enabling work; partnered programs run by big pharma are tracked separately under Pharma Deals.
AI-discovered or AI-optimized programs being advanced inside big pharma — either developed internally on AI platforms or in-licensed/acquired from AI-first biotechs. Includes platform partnerships where AI is the explicit basis of the collaboration.
Pharma
Type
Companies actively open to pharma partnerships, platform deals, or self-serve access — the AI-bio BD shortlist. Excludes pure-internal pipeline plays and any company in a hiring freeze or restructuring. Use the legend to filter by what kind of engagement you need.
Mode
Type
Companies applying learned models to clinical trial operations — patient recruitment, protocol design, digital twins / synthetic control arms, site selection, and trial outcome prediction. This is the layer between drug discovery (Companies tab) and commercialization (deliberately out of scope). Many already-tracked companies (Tempus, Owkin, Dandelion) span both Companies and Clinical AI; here we include pure-play clinical-AI companies.
Focus area
AI-native biotechs that have been acquired, merged, or folded into larger organisations. Their technology and teams continue inside pharma and platform companies — the exit price, where disclosed, gives a sense of how the market has valued AI-bio capability over time.
Editorial Note
What counts as AI-native?
The line between AI-bio and traditional computational drug discovery is genuinely fuzzy, and getting fuzzier as classical platforms add machine learning layers. This tracker takes a clear editorial position to stay useful: we include companies whose core methodology is learned models — neural networks, foundation models, generative AI, large language models trained on biology — applied to drug discovery, diagnostics, gene editing, or clinical data infrastructure.
- Included. Foundation models for protein, RNA, or small-molecule design (Profluent, Generate, Isomorphic, Boltz, Iambic).
- Included. ML on multimodal clinical / hospital data, federated learning, digital pathology (Owkin, PathAI, Caris).
- Included. Generative chemistry combined with physics-based validation, as long as ML is on the critical path (Charm, Relay, Schrödinger, OpenEye post-2022).
- Included. Autonomous lab / AI-scientist agents (Medra, FutureHouse, LILA, Atomistic Insights).
- Excluded. Pure classical FEP / MD / docking platforms with no learned models in the loop.
- Excluded. Traditional CROs that have added AI capabilities as one service among many (e.g., Domainex).
- Excluded. Pure cheminformatics, structural biology, or HPC-as-a-service for chemistry (e.g., classical FEP cloud platforms).
- Excluded. Pharma marketing / commercialization tech (e.g., DTC audience targeting, KOL management SaaS) — downstream of clinical, not part of drug discovery.
- Adjacent (separate tab). Clinical trial AI (recruitment, protocol design, digital twins) — tracked separately in the Clinical AI tab to keep the main Companies list focused on discovery / biology.
- Ecosystem subtypes. The Ecosystem category covers three distinct archetypes that differ in where their moat sits: Patient Data Platforms (Tempus, Owkin) where proprietary multimodal patient data is the differentiator; Generative Bio Platforms (Cradle, Boltz, Converge) where model architecture and training compute are the differentiator; and Research Tooling (FutureHouse, LatchBio, Medra, Potato) where workflow integration with scientists is the differentiator. Selecting the Ecosystem type filter reveals subtype chips.
The classical / AI line moves over time. OpenEye and Schrödinger qualify today because their newer offerings (ROCS X, LiveDesign ML, AutoDesigner, Generative Glide, federated learning integrations) are substantively learned-model approaches, not just physics with a model bolted on. Big Tech AI labs (Meta FAIR, Google DeepMind, NVIDIA, OpenAI, Anthropic) are tracked through their spinouts and partnerships, not as parent-company entries: Isomorphic Labs covers DeepMind, the ESM lineage (originally Meta FAIR, then EvolutionaryScale) now lives at Chan Zuckerberg Biohub as of Apr 2026, NVIDIA shows up in deal records and as a recurring investor. Institutional research orgs (CZI Biohub) are tracked when their AI/bio output is consequential enough to warrant inclusion. Suggestions and rebuttals welcome.