{"id":35847,"date":"2026-04-14T17:25:00","date_gmt":"2026-04-14T15:25:00","guid":{"rendered":"https:\/\/askme.it\/insights\/ai-drug-discovery-in-2026-between-promise-and-production\/"},"modified":"2026-05-28T17:54:38","modified_gmt":"2026-05-28T15:54:38","slug":"ai-drug-discovery-in-2026-between-promise-and-production","status":"publish","type":"insights","link":"https:\/\/askme.it\/en\/insights\/ai-drug-discovery-in-2026-between-promise-and-production\/","title":{"rendered":"AI drug discovery in 2026: between promise and production"},"content":{"rendered":"
\n

The analysis published by Boston Consulting Group with Wellcome tracked around 73 AI-derived molecules in clinical pipelines as of 2023, growth that by 2024 took the segment above 30% of the clinical pipelines of AI-native biotech firms. During the same period, however, no drug discovered or designed entirely by artificial intelligence models has obtained approval from the FDA or EMA. The 2026 scenario is exactly this: a promise that has generated impressive pipeline numbers and no regulatory approval, a race to the model that produced AlphaFold 3 and then three open source alternatives in six months, a market consolidation that left BenevolentAI on the field and merged Recursion with Exscientia.<\/p>\n

The turning point that arrived in 2025 has a precise name: rentosertib, a TNIK inhibitor for idiopathic pulmonary fibrosis developed by Insilico Medicine. It is the first molecule in which both the biological target and the chemical structure were identified and designed by proprietary generative models, and on June 3, 2025 Nature Medicine published its Phase IIa results with a significant clinical efficacy signal. Four months later Insilico listed on the Hong Kong Stock Exchange with the largest Asian biotech IPO of 2025.<\/p>\n<\/section>\n

\n

The data that separates chemistry from biology<\/h2>\n

The central reference for measuring the real maturity of the sector is the analysis by Jayatunga and colleagues published in Drug Discovery Today in June 2024 with the title “How successful are AI-discovered drugs in clinical trials?”. The numbers are sharp: for AI-derived molecules the success rate in Phase I is between 80 and 90%, against a historical industry average of around 50-65%. In Phase II, where efficacy on real patients is measured for the first time, the success rate collapses to 40%, in line with the traditional Big Pharma benchmark. The operational translation is clear: AI optimizes what we can measure computationally, namely drug-likeness, ADMET profile, binding affinity, but does not yet predict biological efficacy in a sick human being.<\/p>\n

The BCG estimate of AI’s potential impact on reducing preclinical phase time and costs remains significant, between 25 and 50%, but these percentages apply to the part of the process that precedes humans. The bottleneck has shifted upward, into target validation, and remains downward, in clinical biology. For pharmaceutical companies planning AI investments, the message is that the measurable return today is in target discovery, literature mining, omics analysis and trial design. Molecule generation is the most visible part but not the most profitable.<\/p>\n<\/section>\n

\n

AlphaFold 3 and the commodity moment<\/h2>\n

On May 8, 2024 Nature published the AlphaFold 3 paper, developed by Google DeepMind and Isomorphic Labs. It is the first model capable of predicting in a single architecture the interactions between proteins, nucleic acids, small molecules, ions and modified residues. For industrial drug discovery workflows it has immediate effect on in silico target validation and on prediction of ligand-protein binding poses. The initial choice to release the model with commercial use restrictions and without public weights drew criticism from the scientific community and triggered a response that within a few months zeroed out the competitive advantage.<\/p>\n

On October 10, 2024 Chai Discovery published Chai-1 with open weights and 77% performance on the PoseBusters benchmark, against 76% for AlphaFold 3. On November 19, 2024 MIT Jameel Clinic released Boltz-1 under MIT license, the first fully open source AlphaFold 3-class model. EvolutionaryScale published ESM3 with 98 billion parameters trained on 2.78 billion proteins, capable of generating GFP variants equivalent to 500 million years of natural evolution. For European pharmaceutical companies planning the technology stack for 2026, this means that AlphaFold 3-level structural prediction has become a commodity. Competitive value has shifted to data, integration with the experimental laboratory and regulatory validation.<\/p>\n<\/section>\n

\n

The 2024-2025 consolidation: fewer players, more compute<\/h2>\n

The “AI drug discovery winter” narrative that circulated in the trade press between 2023 and 2024 was disproved by the 2024-2025 numbers: not a winter, but a consolidation. On August 8, 2024 Recursion and Exscientia announced their merger, closed on November 20 with an equity premium value estimated at around $688 million. Entering 2025 the combined company had about 800 employees; between January and February 2025 it cut the workforce by 20% and deprioritized four pipeline programs. Insilico Medicine listed on HKEX on December 30, 2025 raising $292 million, with an opening capitalization of around $2.5 billion and a retail subscription oversubscribed approximately 1,427 times.<\/p>\n

BenevolentAI, listed on Euronext Amsterdam since 2022, closed its public chapter: after the failure of the BEN-2293 Phase IIa trial for atopic dermatitis in April 2023 and two rounds of layoffs that cut about 180 employees, on March 13, 2025 it delisted via merger with Osaka Holdings and now operates as a private company with runway extended to 2027. AbCellera held the opposite course: no layoffs in 2024-2025, two advanced programs in clinic by the end of 2025 (ABCL635 in Phase 2, ABCL575 in Phase 1) and 19 cumulative clinical molecules from the platform. Atomwise reduced headcount by 30% in December 2023 and completed the leadership transition with new CEO Steve Worland in February 2025. The message for companies evaluating AI partnerships in 2026 is that the supplier market has narrowed to a dozen operational players, and that the choice between the asset-centric model (selling drugs) and the service-centric model (selling software or services to Big Pharma) is now the main survival criterion.<\/p>\n<\/section>\n

\n

Big Pharma’s bets: over three billion in headline value<\/h2>\n

On January 8, 2024 Isomorphic Labs announced two simultaneous deals: with Eli Lilly for $45 million upfront and up to $1.7 billion in milestones, with Novartis for $37.5 million upfront and up to $1.2 billion in milestones. On March 31, 2025 Isomorphic closed its first external financing round for $600 million led by Thrive Capital, with participation from GV and Alphabet, aimed at bringing its proprietary programs into the clinic. Eli Lilly signed a collaboration with OpenAI on June 25, 2024 focused on antimicrobials against resistant pathogens. Sanofi announced on May 21, 2024 a collaboration with OpenAI and Formation Bio for AI software across the entire drug development cycle, in continuity with its internal “plai” platform scaled in June 2023 with Aily Labs.<\/p>\n

On the European front BioNTech acquired InstaDeep on January 10, 2023 for a total value of up to 562 million pounds (362 upfront in cash and shares, 200 in contingent earnout), closing on July 31, 2023 with about 240 InstaDeep employees transitioning. AstraZeneca reinforced its axis with Absci in December 2023 for up to $247 million on an AI-designed oncology antibody. Aggregate estimates of the headline value of pharma-AI partnerships over the 2020-2024 period range between $50 and $70 billion, a figure that should be read knowing that the great majority is composed of milestones conditional on the achievement of technical and regulatory targets, and that realized value remains a much smaller fraction.<\/p>\n<\/section>\n

\n

European 2026 is the year of governance<\/h2>\n

The EU AI Act entered into force on August 1, 2024 and obligations for high-risk systems apply from August 2, 2026, with sanctions up to 35 million euros or 7% of global turnover. The EFPIA position of April 22, 2024 clarified that the majority of AI uses in pharmaceutical R&D fall under the “scientific research and development” exemption provided by articles 2(6) and 2(8), and that typical discovery uses do not fall within the high-risk categories of Annex III. AI uses in regulatory submissions, manufacturing release and pharmacovigilance remain openly under the high-risk regime. EMA published on September 30, 2024 its Reflection Paper on the use of AI in the medicinal product lifecycle, adopted by CHMP and CVMP, with a risk-based approach centered on data quality, model transparency and lifecycle management.<\/p>\n

Italy has three concrete observation points. Domp\u00e9 farmaceutici leads the Exscalate4CoV consortium, funded by Horizon 2020, with a virtual screening platform capable of processing over 3 million molecules per second that identified raloxifene as a candidate for COVID-19 brought into Phase III through AIFA. Chiesi has signed drug design partnerships with Iktos on the Makya generative software and holds an equity position in Cyclica; in 2024 it invested 24.3% of revenue in R&D, ranking eleventh in Europe in the EU Industrial R&D Investment Scoreboard published by the JRC. Human Technopole, in Milan’s MIND district, hosts five centers including Computational Biology and Structural Biology and works on oncology drug discovery with functional genomics and AI. For CIOs and regulatory leads of Italian pharmaceutical companies, 2026 is less the year of model choice and more the year of building the governance framework: documentation, validation, audit trail. The technology is available. Compliance is the constraint that decides.<\/p>\n<\/section>\n","protected":false},"excerpt":{"rendered":"

BCG counted 73 AI-derived molecules in clinical pipelines, but success rates collapse from 80-90% in Phase I to around 40% in Phase II. Insilico Medicine’s rentosertib is the first entirely AI-designed molecule to show an efficacy signal in Phase IIa, published in Nature Medicine in June 2025. An honest 2026 map: what works, what doesn’t, what is changing in Europe.<\/p>\n","protected":false},"featured_media":35849,"menu_order":0,"template":"","insights_category":[968],"insights_tags":[879,878,715,881,970],"class_list":["post-35847","insights","type-insights","status-publish","has-post-thumbnail","hentry","insights_category-ai-and-pharma","insights_tags-alphafold","insights_tags-drug-discovery","insights_tags-eu-ai-act-en","insights_tags-insilico-medicine","insights_tags-pharmaceutical-research"],"acf":[],"_links":{"self":[{"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights\/35847","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights"}],"about":[{"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/types\/insights"}],"version-history":[{"count":1,"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights\/35847\/revisions"}],"predecessor-version":[{"id":35848,"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights\/35847\/revisions\/35848"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/media\/35849"}],"wp:attachment":[{"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/media?parent=35847"}],"wp:term":[{"taxonomy":"insights_category","embeddable":true,"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights_category?post=35847"},{"taxonomy":"insights_tags","embeddable":true,"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights_tags?post=35847"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}