AI in Pharmaceutical Drug Product Design: The 2026 Solution Landscape

AI in pharmaceutical drug product design is, in 2026, a decision-support technology, not autonomous formulation. The most mature tools combine historical formulation knowledge, material data, mechanistic models, and design-of-experiments (DoE) and Quality-by-Design (QbD) principles to recommend formulation strategies, screen excipients, predict solubility and bioavailability, and de-risk scale-up, with the strongest results in oral solid dosage forms. They reduce trial-and-error, but experienced formulators still own the decisions.

The provider landscape splits into six groups: formulation-design copilots (BASF ZoomLab, TaBlitz, FormulationAI, Formulaite.ai), enterprise and CDMO-embedded platforms (Thermo Fisher OSDPredict, BIOVIA, Merck KGaA/EMD), mechanistic and PBPK modeling (Simulations Plus GastroPlus, Certara Simcyp, Schrödinger), materials informatics (Citrine, Uncountable), process and manufacturing AI (Aizon, DataHow, AspenTech, Siemens gPROMS), and the data infrastructure that feeds all of them (Benchling, Dotmatics, TetraScience, IDBS). The FDA’s January 2025 draft guidance frames regulated use around a risk-based credibility assessment tied to a defined context of use (COU). The best approach is a hybrid stack: structured data, mechanistic modeling, AI prediction, and human judgement.

Table of Contents

• Where AI Is Already Useful in Formulation Development
• How Regulators View AI in Drug Product Development
• The AI Formulation Solution Provider Landscape
• Formulation-Strategy and Oral Solid Dose Copilots
• Enterprise and CDMO-Embedded Formulation Platforms
• Mechanistic, PBPK, and Molecular Modeling
• Materials Informatics and R&D Experiment Platforms
• Process Development, Scale-Up, and Manufacturing AI
• The Data Infrastructure Layer
• Practical Pros of AI in Formulation Development
• Practical Cons and Risks
• Frequently Asked Questions
• Key Takeaways
• Sources

Where AI Is Already Useful in Formulation Development

AI in drug product design is currently strongest in five areas, all of which support the formulator rather than replace them:

• Preformulation and excipient selection: predicting API-excipient compatibility, solubility enhancement routes, stability risks, and likely formulation platforms.

• Solubility-Enabling Platform Selection: choosing between solubility-enabling platforms such as cyclodextrin complexation, amorphous solid dispersion, lipid systems, nanocrystals, liposomes, and self-emulsifying systems.

• Oral solid dosage design: tablet geometry, manufacturability, compressibility, tooling considerations, and patient-centric design.
• Predictive modeling and digital twins: models for dissolution, permeability, bioavailability, first-in-human (FIH) dose estimation, packaging, scale-up, and process robustness.
• Knowledge management: AI-assisted retrieval of formulation history, prior failed experiments, excipient performance, and manufacturing lessons learned.

The value concentrates on poorly soluble compounds. An estimated majority of new small-molecule candidates fall into Biopharmaceutics Classification System (BCS) class II or IV, where dissolution or permeability limits exposure, so tools that rank solubility enhancement routes early save significant bench time. The main limitation is data quality: formulation data are often sparse, proprietary, inconsistent, and full of negative results that are rarely published.

How Regulators View AI in Drug Product Development

Regulators are preparing for wider AI use across the drug product lifecycle, including manufacturing. The FDA’s January 2025 draft guidance introduces a risk-based credibility framework that assesses an AI model against a defined context of use (COU): the specific question the model answers and the consequence if it is wrong. The higher the model influence and the higher the decision risk, the more credibility evidence is required.

For formulators, the practical implication is that a model used to prioritize early excipient screening carries a lighter evidentiary burden than one informing a specification, a control strategy, or a clinical dose. Regulated use therefore needs a documented COU, model versioning, input-data traceability, human expert review, and a credibility assessment proportional to risk. AI does not remove the formulator from the loop; it changes what must be documented around their judgement.

See and download are infographic on The AI Formulation Solution Provider Landscape:

The AI Formulation Solution Provider Landscape

The table below maps the providers most relevant to pharmaceutical formulators. The sections that follow describe each provider in text, grouped by the job it does best.

Provider / Platform	Focus	Strengths	Limitations
BASF ZoomLab®	AI-enabled (machine-learning–supported) digital platform for drug product development – especially oral solid formulation support and solubilization	Developed specifically for formulators; predicts next formulation steps; covers BASF excipients plus partner portfolios (e.g., Meggle, IFF, Lipoid, Budenheim) within the Virtual Pharma Assistant ecosystem.	Positioned as AI-enabled rather than a standalone AI tool; depth depends on access level / premium features.
TaBlitz / TaBlitz Labs	AI-supported tablet design, tablet geometry, manufacturability, compression and formulation analysis	Strong focus on oral solid dosage forms; combines tablet design, formulation insight, compression characterization, USP <1062>-aligned methods, and right-first-time tablet development.	Primarily tablet-focused; less relevant for injectables, biologics, liquids, semisolids, or advanced drug delivery systems.
FormulationAI	Academic/free web-based AI platform for in silico drug formulation design	Broad formulation strategy coverage; published in Briefings in Bioinformatics; useful for early screening and education.	Academic/research orientation; may not provide enterprise workflow, GMP documentation, audit trail, or direct regulatory-ready validation.
Formulaite.ai	AI formulation for phytopharmaceutical, nutraceutical, and herbal products	Uses quantum chemistry, GNN-based ADMET modeling, and multi-compartment PBPK simulation to predict bioavailability, excipient interactions, activity, and shelf-life stability for herbal and nutraceutical formulas.	Not a pharmaceutical drug product platform; built for polyherbal and phyto-pharmaceutical formulas, lipophilic molecules, and excipient modeling across nutraceutical and wellness formats.
Thermo Fisher / Patheon OSDPredict™	AI/ML toolbox for oral solid dose formulation and development services	Combines predictive models for solubility, permeability, bioavailability, FIH dosing, packaging, and scale-up planning.	Appears embedded in Thermo Fisher/Patheon CDMO services rather than a broad independent software product; access may depend on collaboration model.
BIOVIA Formulation Design / Dassault Systèmes	Formulation data management, recipe management, calculations, optimization workflows	Enterprise-grade formulation informatics; useful for structured formulation knowledge, collaboration, and digital continuity.	Broader CPG/formulation platform, not pharma-specific AI drug product design only; implementation can be complex.
Merck KGaA / EMD AI formulation tools	AI-assisted formulation, especially poorly soluble drugs	Strong science and excipient/materials background; positions AI as a route to faster formulation solutions for poorly soluble compounds.	Public information suggests internal/service-linked tools rather than a fully open standalone platform.
Simulations Plus GastroPlus® / ADMET Predictor®	PBPK, absorption, dissolution, bioavailability, IVIVC, virtual bioequivalence	Very strong for mechanistic and regulatory-relevant modeling; complements AI with physiologically based models.	Not a pure formulation AI platform; requires modeling expertise and high-quality input data.
Certara Simcyp® / Phoenix	PBPK, clinical pharmacology, model-informed drug development	Strong regulatory familiarity; useful for formulation impact on exposure and dose decisions.	More clinical pharmacology/PBPK than practical excipient or tablet design.
Schrödinger LiveDesign / Materials Science tools	Molecular modeling, drug–excipient interactions, solid-state risk, molecular properties	Strong computational chemistry base; useful for mechanistic understanding of interactions and stability.	More molecule/materials modeling than end-to-end formulation workflow.
Citrine Informatics	Materials informatics and AI for product/material development	Good for structured experimental learning, materials selection, and optimization workflows.	Not pharma formulation-specific by default; requires customization and data integration.
Uncountable	R&D data platform and ML for formulation-heavy industries	Strong for capturing formulation experiments, building ML models, and reducing duplicated work.	Broad formulation/chemicals focus; pharma validation and GxP use need careful implementation.
Aizon	AI for pharmaceutical manufacturing and process optimization	Stronger in GMP manufacturing, process monitoring, deviation reduction, and digital plant intelligence.	Less focused on early formulation design.
DataHow	Hybrid modeling, process development, bioprocess and pharma manufacturing analytics	Useful for process modeling, scale-up, and combining mechanistic and data-driven models.	More process/manufacturing-oriented than excipient selection or tablet design.
AspenTech / gPROMS (Siemens) / PSE-type tools	Process modeling, digital twins, scale-up, manufacturing simulation	Valuable for process understanding and scale-up robustness.	Usually not “AI formulation design” out of the box; needs expert modeling.
Benchling / Dotmatics / TetraScience / IDBS	Scientific data infrastructure and AI-ready R&D data capture	Important enabling layer: structured data, experiment history, metadata, ELN/LIMS integration.	They enable AI but are not formulation prediction engines by themselves.

Formulation-Strategy and Oral Solid Dose Copilots

Formulation-design copilots recommend the next experiment rather than run it, and four platforms anchor this group.

BASF ZoomLab is an AI-enabled, machine-learning-supported platform built specifically for formulators, focused on oral solid formulation support and solubilization; it predicts next formulation steps and covers BASF excipients plus partner portfolios such as Meggle, IFF, Lipoid, and Budenheim inside the Virtual Pharma Assistant ecosystem.

TaBlitz concentrates on tablets, combining tablet geometry, manufacturability, compression characterization aligned with USP General Chapter <1062>, and right-first-time development, which makes it powerful for oral solids but less relevant for injectables, biologics, liquids, or semisolids.

FormulationAI is a free, peer-reviewed academic web platform (published in Briefings in Bioinformatics) covering drug-cyclodextrin systems, solid dispersions, phospholipid complexes, nanocrystals, self-emulsifying systems, liposomes, and solubility prediction, well suited to early screening and education but without enterprise workflow, GMP documentation, or audit trails.

Formulaite.ai applies quantum chemistry, graph-neural-network (GNN) ADMET modeling, and multi-compartment PBPK simulation to phytopharmaceutical, nutraceutical, and herbal formulas rather than mainstream drug products.

Enterprise and CDMO-Embedded Formulation Platforms

Enterprise and CDMO-embedded platforms target organizations that need formulation AI inside a validated, collaborative environment.

Thermo Fisher’s Patheon OSDPredict is an AI/ML toolbox for oral solid dose development that combines predictive models for solubility, permeability, bioavailability, FIH dosing, packaging, and scale-up planning; it appears embedded in Patheon’s CDMO services rather than sold as standalone software, so access typically follows a development collaboration.

BIOVIA Formulation Design from Dassault Systèmes provides enterprise-grade formulation informatics: recipe and formulation data management, calculations, and optimization workflows that preserve institutional knowledge and digital continuity, though it is a broader consumer-and-industrial formulation platform rather than a pharma-only tool, and implementation can be complex.

Merck KGaA (operating as EMD in North America) positions AI as a route to faster formulation of poorly soluble compounds, backed by deep excipient and materials science; public information suggests these are internal or service-linked tools rather than a fully open standalone platform.

Mechanistic, PBPK, and Molecular Modeling

Mechanistic modeling platforms complement data-driven AI with physics and physiology, which matters when regulators ask why a prediction holds.

Simulations Plus GastroPlus and ADMET Predictor model absorption, dissolution, bioavailability, in vitro-in vivo correlation (IVIVC), and virtual bioequivalence, letting formulators test how a particle size or dissolution change propagates to exposure before running a clinical study; the trade-off is that they require modeling expertise and high-quality input data.

Certara’s Simcyp and Phoenix focus on physiologically based pharmacokinetic (PBPK) modeling and clinical pharmacology, carry strong regulatory familiarity, and are most useful for connecting a formulation change to systemic exposure and dose decisions rather than for excipient or tablet design directly.

Schrödinger’s LiveDesign and Materials Science tools bring a computational chemistry base to drug-excipient interactions, solid-state and crystal-form risk, and molecular property prediction, which supports mechanistic understanding of stability and incompatibilities but is more molecule-and-materials modeling than an end-to-end formulation workflow.

Used together, they turn formulation hypotheses into testable, exposure-level predictions before clinical work begins.

Materials Informatics and R&D Experiment Platforms

Materials informatics platforms apply machine learning to structured experimental data so that each new study improves the model.

Citrine Informatics is built for materials and product development, supporting materials selection, sequential learning, and optimization workflows; it is not pharma-formulation-specific out of the box and needs customization and data integration to fit drug product work.

Uncountable is an R&D data platform with machine learning aimed at formulation-heavy industries, strong at capturing formulation experiments, building predictive models, and reducing duplicated work, though its breadth across chemicals and consumer products means pharma validation and GxP use require careful implementation.

Both address the recurring problem that formulation knowledge is scattered across notebooks and spreadsheets: by structuring experiments, including the negative results that are rarely published, they make a formulation program’s own history usable as training data.

Process Development, Scale-Up, and Manufacturing AI

Process and manufacturing AI applies once a formulation exists and the question shifts to making it robustly at scale.

Aizon focuses on GMP manufacturing, process monitoring, deviation reduction, and digital plant intelligence, which makes it stronger in production than in early formulation design.

DataHow specializes in hybrid modeling that blends mechanistic and data-driven approaches for process development, scale-up, and bioprocess analytics, useful when neither pure first-principles nor pure machine learning alone is sufficient.

AspenTech, Siemens gPROMS FormulatedProducts, and PSE-type tools provide process modeling, digital twins, and scale-up simulation that improve process understanding and robustness, but they are generally not turnkey AI formulation design and require expert modeling to set up.

For formulators, this group de-risks the transition from a bench recipe to a manufacturable, transferable process, where QbD design space thinking and AI prediction converge.

The Data Infrastructure Layer

The data infrastructure layer is the enabling foundation beneath every other tool in this landscape, because AI is only as good as the data it learns from.

Benchling, Dotmatics, TetraScience, and IDBS provide electronic lab notebook (ELN), laboratory information management system (LIMS), and scientific data management capabilities that capture experiments, instrument output, and metadata in a structured, AI-ready form. They are not formulation prediction engines themselves; their role is to make formulation history findable, accessible, interoperable, and reusable (the FAIR data principles) so that a prediction model has clean, traceable inputs.

Skipping this layer is the most common reason AI formulation initiatives stall: without structured, well-governed data, even the best model produces unreliable output. Input-data traceability from these systems also directly supports the credibility documentation that regulated use requires.

Practical Pros of AI in Formulation Development

AI can reduce the number of experimental cycles, prioritize excipients, predict likely failure modes, improve DoE design, and preserve institutional formulation knowledge that would otherwise leave with departing scientists. It is especially valuable for poorly soluble APIs, low-material early development where every milligram counts, tablet manufacturability, and early risk assessment. Used as a screening and prioritization layer, it lets formulators spend bench time on the few candidates most likely to succeed rather than exhaustively testing a large design space.

Practical Cons and Risks

The main risks are overconfidence in plausible but wrong predictions, black-box outputs that cannot be explained to a reviewer, poor transferability between APIs, missing negative data, weak explainability, and insufficient model validation. A model trained on one compound class can fail silently on a structurally different molecule.

For regulated use, companies need a clear context of use, model versioning, input-data traceability, human expert review, and a documented credibility assessment proportional to the decision’s risk. Treating AI output as a hypothesis to be confirmed at the bench, not a conclusion, keeps these risks manageable.

Frequently Asked Questions

Can AI fully automate pharmaceutical formulation in 2026? No. The mature applications are decision-support systems, not autonomous formulation. AI prioritizes excipients, predicts properties, and proposes strategies, but experienced formulators select, test, and confirm. The most reliable setups pair AI prediction with mechanistic modeling and bench validation rather than acting on predictions directly.

Which AI tools help most with poorly soluble (BCS class II or IV) drugs? Formulation-strategy platforms such as BASF ZoomLab and FormulationAI rank solubility enhancement routes (solid dispersions, cyclodextrins, lipid and self-emulsifying systems, nanocrystals), and Merck KGaA/EMD positions its AI specifically for poorly soluble compounds. Simulations Plus GastroPlus then predicts how the chosen route affects absorption and bioavailability.

What is a context of use in the FDA’s AI guidance? The context of use (COU) is the specific question an AI model answers and the consequence if it is wrong. The FDA’s January 2025 draft guidance ties required credibility evidence to the COU: higher model influence and higher decision risk demand more validation. A screening model needs less evidence than one informing a specification.

Why is data quality the main limitation for formulation AI? Formulation data are sparse, proprietary, inconsistent, and dominated by unpublished negative results. Models trained on thin or biased data generalize poorly across APIs. This is why data infrastructure platforms (Benchling, Dotmatics, TetraScience, IDBS) that structure experiment history are often the prerequisite for any successful AI initiative.

What is the difference between formulation-design AI and PBPK modeling? Formulation-design AI (ZoomLab, TaBlitz, OSDPredict) recommends excipients, strategies, and tablet designs from data and material properties. PBPK modeling (GastroPlus, Simcyp) uses physiology and physics to predict how a formulation behaves in the body. They are complementary: design AI proposes the formulation, PBPK predicts its in vivo consequence.

How should a formulation group combine these tools? Build a hybrid stack: structured data infrastructure as the foundation, mechanistic and PBPK modeling for physiology, AI prediction for screening and prioritization, and experienced human judgement for decisions. No single platform covers preformulation through scale-up, so most groups integrate a formulation copilot, a modeling tool, and a data layer.

Key Takeaways

• AI in drug product design is decision support in 2026, strongest in oral solid dosage forms, and does not replace formulation scientists.
• The provider landscape spans formulation copilots, enterprise and CDMO platforms, mechanistic and PBPK modeling, materials informatics, manufacturing AI, and data infrastructure.
• Poorly soluble (BCS class II or IV) compounds are where AI-driven strategy selection delivers the most bench-time savings.
• The FDA’s January 2025 draft guidance ties required model credibility to a defined context of use and decision risk.
• Data quality, not algorithms, is the binding constraint; structured, traceable experiment data is the prerequisite for reliable predictions.
• The best results come from a hybrid stack: structured data, mechanistic modeling, AI prediction, and experienced human judgement.

Sources

1. U.S. Food and Drug Administration. Artificial Intelligence in Drug Development. https://www.fda.gov/about-fda/center-drug-evaluation-and-research-cder/artificial-intelligence-drug-development (accessed 2026-06-09).
2. United States Pharmacopeia. General Chapter <1062> Tablet Compression Characterization. USP-NF (accessed 2026-06-09).
3. Yang, J., et al. FormulationAI: a novel web-based platform for drug formulation design driven by artificial intelligence. Briefings in Bioinformatics, 25(1), bbad419, 2024.
4. FormulationAI. In silico drug formulation design platform. https://formulationai.computpharm.org/ (accessed 2026-06-09).
5. BASF Pharma Solutions. ZoomLab / Virtual Pharma Assistant. https://pharmaceutical.basf.com/global/en/pharma-solutions/about-basf-pharma/digitalization/zoomlab (accessed 2026-06-09).
6. TaBlitz Labs. AI-supported tablet design and formulation analysis. https://www.tablitzlabs.com/ (accessed 2026-06-09).
7. Formulaite.ai. AI formulation for phytopharmaceutical, nutraceutical, and herbal products. https://formulaite.ai/ (accessed 2026-06-09).
8. Thermo Fisher Scientific / Patheon. OSDPredict in silico modeling for oral solid dose. https://www.patheon.com/us/en/our-capabilities/small-molecule/oral-solid-dose/in-silico-modeling.html (accessed 2026-06-09).
9. Dassault Systèmes BIOVIA. AI-powered formulation development. https://www.3ds.com/products/biovia/ai-powered-formulation-development (accessed 2026-06-09).
10. Merck KGaA / EMD. Harnessing AI to speed up drug formulation. https://www.emdgroup.com/en/research/science-space/envisioning-tomorrow/harnessing-ai-to-speed-up-drug-formulation.html (accessed 2026-06-09).
11. Simulations Plus. GastroPlus and ADMET Predictor. https://www.simulations-plus.com/ (accessed 2026-06-09).
12. From In Vivo Predictive Dissolution to Virtual Bioequivalence: A GastroPlus-Driven Framework. PMC. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12030460/ (accessed 2026-06-09).
13. Certara. Simcyp PBPK Simulator. https://www.certara.com/software/simcyp-pbpk/ (accessed 2026-06-09).
14. Schrödinger. Pharmaceutical formulations and delivery. https://www.schrodinger.com/materials-science/solutions/pharmaceutical-formulations-delivery/ (accessed 2026-06-09).
15. Citrine Informatics. Materials informatics platform. https://citrine.io/platform/ (accessed 2026-06-09).
16. Uncountable. R&D data and machine learning platform. https://www.uncountable.com/company (accessed 2026-06-09).
17. Aizon. AI for pharmaceutical manufacturing. https://www.aizon.ai/ (accessed 2026-06-09).
18. DataHow. Hybrid modeling for process development. https://datahow.ch/ (accessed 2026-06-09).
19. Siemens. gPROMS FormulatedProducts. https://www.siemens.com/global/en/products/automation/industry-software/gproms-digital-process-design-and-operations/gproms-modelling-environments/gproms-formulatedproducts.html (accessed 2026-06-09).
20. Benchling, Dotmatics, TetraScience, and IDBS. Scientific data infrastructure (ELN/LIMS). https://www.benchling.com/, https://www.dotmatics.com/, https://www.tetrascience.com/, https://www.idbs.com/ (accessed 2026-06-09).

This article is for informational purposes for pharmaceutical industry professionals and does not constitute regulatory, formulation, or procurement advice. It is not an endorsement of any vendor or platform. AI tool capabilities, product names, and access models change over time; always confirm current functionality, validation status, and GxP suitability against each vendor’s official documentation and applicable regulatory guidance (FDA, EMA, ICH) for your dosage form, route of administration, and market. Pharma Excipients International AG is not affiliated with the platforms discussed and is not a vendor of formulation AI software.