AI drug discovery has been telling an efficiency story for many years. Now it is time to deliver the report card.

Among the first batch of molecules entering the commercialization validation stage, zasocitinib is a sample that is hard to ignore — it has the potential to become the first truly commercially validated product of AI-assisted drug discovery.

In December 2022, Takeda acquired this drug, then codenamed NDI-034858, from Nimbus Therapeutics for $4 billion upfront and up to $2 billion in sales milestone payments — a deal size rarely seen in the autoimmune field. More critically, this product subsequently defeated BMS's deucravacitinib in a "head-to-head" Phase III study, becoming a case of an AI-discovered molecule going head-to-head against an opponent on a mature track and winning.

Nimbus's asset factory model continues to output tradable drug candidates. At the same time, collaboration in the AI drug discovery industry is deepening: Eli Lilly's collaboration with Insilico Medicine刷新了 domestic AI drug discovery records, while Sanofi and Earendil Labs upgraded from drug-specific licensing to platform-level binding.

MNCs' competition for AI capabilities has shifted from tentative collaboration to deep bets. However, only as approval outcomes, clinical data, and milestone realizations land one by one will the answer truly become clear as to whether AI drug discovery can transform from an efficiency tool into a source of new drugs.

Chapter 1: About to Deliver the Report Card

In September just past, the FDA accepted Takeda Pharmaceutical's new drug marketing application for zasocitinib and granted it priority review for the treatment of moderate-to-severe plaque psoriasis in adults, with a PDUFA date in the first quarter of 2027. The European Medicines Agency has also accepted the marketing authorization application for the drug, with the review procedure initiated in parallel.

Supporting this filing is a comprehensive data package covering nearly 3,000 patients. Among them, two key Phase III studies, LATITUDE PsO 3001 and 3002, enrolled approximately 1,800 patients in total and both met all primary endpoints and key secondary endpoints.

Data show that approximately 70% of patients achieved clear or almost clear skin lesions (sPGA 0/1) at Week 16; most patients achieved rapid and durable lesion clearance, with clearance observable as early as Week 4, continuing to improve through Week 24, and further improving by Week 52. High levels of clearance were also achieved in difficult-to-treat areas such as the scalp, nails, palms, and soles. In terms of safety, zasocitinib was well tolerated, with no new signals identified.

What truly drew widespread industry attention to zasocitinib was the results of a "head-to-head" Phase III study announced in June. In the LATITUDE Atlas study, zasocitinib demonstrated statistical superiority over BMS's deucravacitinib on all primary endpoints and key secondary endpoints. At Week 16, more than 35% of patients treated with zasocitinib achieved complete lesion clearance, with a response rate more than 2.5 times that of deucravacitinib, and the two curves separated as early as Week 8.

The R&D starting point can be traced back to around 2016. Nimbus launched R&D targeting TYK2, aiming to develop a TYK2 inhibitor without the off-target side effects of other JAK inhibitors. TYK2 belongs to the JAK protein family, and JAK inhibitors have been widely validated in the autoimmune disease field, but off-target side effects have long constrained the therapeutic window of this drug class. Nimbus researchers used machine learning and free energy perturbation methods to computationally evaluate more than 13,000 compounds, identifying the final candidate molecule in 2020 after staged screening.

This experience was later labeled "AI drug discovery," but the actual process is more complex than the label. Nimbus R&D President Peter Tummino is quite cautious in his assessment of this drug. He stated that for a drug discovered only in 2020, the claim of "the first AI-approved drug" is not accurate. AI's role in zasocitinib's development was to accelerate computational evaluation and optimize molecular properties, rather than directly generating drugs.

Insight database shows that currently only BMS's deucravacitinib is on the market globally as a TYK2 inhibitor, approved in the U.S. for plaque psoriasis and psoriatic arthritis, and additionally approved in Japan for pustular and erythrodermic psoriasis. In 2025, the drug's global sales were $291 million, up 18.29% year-over-year.

This filing by zasocitinib could potentially make it the world's second TYK2 inhibitor. In addition, five other TYK2 inhibitors have entered Phase III clinical trials, from Hansoh Pharmaceutical (HS-10374), InnoCare Pharma (ICP-488), Wenda Pharmaceutical (WD-890), Alumis (Envudeucitinib), and InventisBio (D-2570).

Chapter 2: The "Asset Factory"

Nimbus's story goes far beyond one drug.

In 2009, Nimbus was co-founded by Atlas Venture partner Bruce Booth and Ramy Farid, CEO of computational chemistry company Schrödinger.

The company started with a virtual lab model, operating through outsourcing, and established subsidiaries under a limited liability company structure, with each subsidiary holding an independent drug program. In 2011, Bill Gates and Schrödinger co-founder Richard Friesner participated in Nimbus's seed round extension investment.

This model determined Nimbus's DNA. The company combined high-resolution protein structure data with computational chemistry; early small-molecule compound screening was provided by Schrödinger's software and technology, while Nimbus provided cash and incentives, with both parties sharing the final outcome returns. To date, Nimbus has raised a cumulative $637 million, with a pipeline covering three major areas: oncology, immunology, and metabolic diseases.

Nimbus's pipeline mainly contains two products.

The first is NDI-219216, in Phase I/II clinical development, targeting WRN helicase for microsatellite instability-high tumors. WRN is a DNA helicase and a validated synthetic lethal target in MSI tumors, with application potential in multiple tumor types including colorectal cancer, gastric cancer, and endometrial cancer.

In December 2025, Nimbus announced that the drug completed Phase I/II dose escalation. Data showed that NDI-219216 achieved more than 24 hours of WRN target engagement at multiple doses, with no dose-limiting toxicities observed, no maximum tolerated dose reached, dose escalation completed approximately nine months ahead of schedule, and antitumor activity demonstrated in multiple MSI-H tumor models.

The other program is a SIK2 inhibitor. The SIK family contains three subtypes: SIK1, SIK2, and SIK3, which serve as key signaling nodes in myeloid cells. Nimbus used structure-based drug design to develop a highly selective SIK2 inhibitor with over 300-fold selectivity for SIK1 and SIK3 subtypes.

Preclinical data show that selective SIK2 inhibition drives dual pharmacological characteristics in myeloid cells and mice, both inhibiting pro-inflammatory cytokines TNFα, IL-23, and IL-6, and upregulating pro-resolution and tissue repair factor IL-10. This dual mechanism contrasts with SIK3 inhibition and JAK1 inhibition, which instead elevate inflammatory drivers such as IL-23 and IL-12. In rheumatoid arthritis mouse models, the SIK2 inhibitor's efficacy was comparable to etanercept and upadacitinib.

Nimbus Therapeutics Pipeline

The path Nimbus represents is essentially converting efficiency advantages in the drug discovery stage into tradable assets. The company targets validated but difficult-to-drug targets, combines AI-enhanced computational chemistry with structure-based design, rapidly produces candidate molecules, and then licenses or sells them to large pharmaceutical companies for monetization.

From the transaction record, Nimbus's asset factory model has already completed multiple cycles.

In April 2016, Gilead announced the acquisition of its liver disease subsidiary Nimbus Apollo for $400 million upfront and $800 million in potential milestones, with the transaction closing in May, making Nimbus Apollo a wholly-owned Gilead subsidiary. The 2022 Takeda acquisition was the most landmark deal.

Subsequently, Nimbus established a collaboration with Eli Lilly to develop cardiovascular and metabolic drugs. In January 2026, the two parties reached a new multi-year research collaboration and global exclusive licensing agreement to develop novel oral obesity and metabolic disease treatments.

Chapter 3: A New Proposition

In the past two years, MNCs' layout in the AI drug discovery field has clearly accelerated. Deals by Eli Lilly, Sanofi, Novartis, and others have pushed AI from an efficiency tool in the drug discovery stage to an infrastructure layer spanning the entire R&D chain and embedded in the pharmaceutical system.

According to iiMedia Research statistics, the global AI drug discovery market grew from $790 million in 2021 to $1.82 billion in 2024, reaching $2.41 billion in 2025, and is expected to reach $2.99 billion in 2026. The Chinese market has also grown rapidly, from approximately RMB 160 million in 2021 to over RMB 600 million in 2025.

Alongside market expansion, the ways large pharmaceutical companies acquire AI capabilities are also changing, from early tool procurement and project collaboration to direct acquisitions and platform-level strategic partnerships. On the 2025 China market licensing collaboration transaction volume list, XtalPi and Insilico Medicine ranked first and second, respectively — AI drug discovery companies' technical capabilities are entering MNCs' core pipeline decision-making视野.

The rise in transaction amounts is more intuitive. In March 2026, Eli Lilly reached an out-licensing and global R&D collaboration with Insilico Medicine, totaling up to approximately $2.75 billion, including $115 million upfront — refreshing the record for domestic AI drug discovery collaboration amounts.

Sanofi and Earendil Labs reached a licensing collaboration in April 2025 for two bispecific antibodies, upgraded to a platform-level strategic partnership in January 2026, with total value pushed to $2.56 billion, covering multiple autoimmune and inflammatory disease programs. In December 2025, Novartis reached a collaboration with Relation Therapeutics potentially worth $1.7 billion, combining its expertise in immunodermatology with Relation's drug discovery AI platform.

Computing power and data infrastructure construction are also advancing in parallel. Pfizer and XtalPi expanded their strategic collaboration in June 2025 to jointly develop a next-generation molecular simulation platform. The R&D paradigm advocated by XtalPi — AI participating in experimental design, robots executing experiments, and experimental data feeding back to AI — is being adopted by more institutions. BMS deployed NVIDIA's DGX SuperPOD based on the Vera Rubin architecture in July 2026, becoming the first life sciences company to deploy this system.

For Chinese AI drug discovery companies, this round of MNC layout brings direct commercial opportunities. Insilico Medicine's AI platform has covered 13 of the world's top 20 pharmaceutical companies. Huashen Zhiyao, incubated by Tsinghua University's Institute for AI Industry Research, has its overseas entity Earendil Labs collaborating with Sanofi on bispecific antibodies for autoimmune and inflammatory disease programs, with the two collaborations' total potential transaction value exceeding $4.4 billion.

Currently, AI drug discovery has passed the stage dominated by single-point tools and proof of concept, and is entering a new period of technology embedding and industrial integration. Industry penetration continues to rise, and the boundary between the two business lines of AI and pharmaceuticals is becoming blurred.

The core proposition of this evolution has also changed accordingly. The question the industry needs to answer has shifted to whether AI can stably produce differentiated drugs that withstand clinical validation, and whether large pharmaceutical companies can truly embed AI capabilities into the complete decision-making chain from target discovery to clinical development.

The approval decision for zasocitinib, whether Nimbus's SIK2 inhibitor can smoothly enter clinical trials, and whether BD transactions by Chinese AI drug discovery companies such as Insilico Medicine and XtalPi can continuously deliver on clinical milestones — these will be the next key nodes for testing this proposition.

From the AI drug discovery approval sprint to the global R&D race, the industry needs more high-frequency, in-depth exchange and connection. On November 27–28, 2026, Tongxieyi will hold the 17th Taizhou Medical Expo (Medical Expo) at the China Medical City Exhibition Center. Among them, the "AI Virtual Cells and Multimodal Data Collection" special matchmaking exchange event will bring together top experts from industry and capital, aiming at domestic and international pharmaceutical and industrial development frontiers with fresh perspectives, and conducting in-depth discussions and resource matchmaking.

This year's Medical Expo will invite nearly 300 exhibitors to exhibit, exchange, and collaborate, promoting full-value-chain integrated development through an all-new immersive exhibition format. We look forward to gathering in Taizhou to explore the next journey together.

References:

The Wind of Autoimmunity Blows Toward the Next Blue Ocean; Tongxieyi

Nimbus used AI to help develop Takeda's $4B psoriasis bet. Here's where it's taking the tech next; PharmaVoice

Takeda's $6 Billion In-Licensed New Drug Filed for Marketing; PharmCube Info