Insilico Medicine says artificial intelligence and laboratory infrastructure in China have reduced some early drug-discovery programmes from years to roughly one. Its fastest project reached preclinical candidate nomination in nine months, with a typical timeline near 13 months.
That is a meaningful acceleration, but candidate nomination is only the start of development. Clinical trials, manufacturing and regulatory review still determine whether a medicine reaches patients.
AI narrows the compounds sent to the laboratory
Insilico uses generative systems to identify biological targets, design molecules and rank compounds for testing. Researchers then synthesise candidates and verify their biological activity and drug-like properties experimentally.
The company says its programmes usually nominate a preclinical candidate within 12 to 18 months after testing 60 to 200 molecules. Conventional discovery may take about four and a half years to reach a comparable point, according to chief executive Alex Zhavoronkov.
Insilico has not published a direct comparison between otherwise equivalent AI-supported and conventional programmes. Different diseases, targets and laboratory practices make simple timeline comparisons difficult. The relevant question is whether models reduce synthesis and failure without discarding promising chemistry.
Since 2021, the company reports generating 31 preclinical candidates, with 13 receiving investigational new drug clearances. Its workflow divides work geographically: AI teams operate in Montreal and Abu Dhabi, while much experimental validation and scale-up occurs in Shanghai.
Automation in the Shanghai facility handles parts of sampling and compound screening. That connection between model design and fast physical testing is crucial. Predictions gain value only when laboratories can return evidence quickly enough to guide the next design cycle.
China adds speed through infrastructure and regulation
Zhavoronkov attributes part of the improvement to operating costs, research capacity and China’s regulatory environment. He estimates that pharmaceutical laboratories in the country can remove roughly two years from traditional candidate development.
China has expanded from generic ingredient manufacturing into novel medicines, supported by contract research organisations, biotechnology companies and trial centres. A Pfizer executive said clinical development there can be three times faster and cost about half as much as European work.
An eligible Class I innovative-drug trial application can use a 30-working-day review pathway introduced in 2025, while complex submissions may move to 60 working days. Faster review can shorten waiting, but scientific standards and patient protection still require complete evidence.
Insilico earns more than 90% of its revenue from Western pharmaceutical customers, according to Zhavoronkov, and has agreements with companies including Eli Lilly and Takeda. It limits most software sales inside China over geopolitical concerns even while planning to expand Shanghai research.
This structure demonstrates the tension in global biotechnology: talent, software, laboratories, customers and regulation span several jurisdictions. Faster cycles depend on all of them, while export controls and political risk can disrupt the same network.
Rentosertib will test whether speed survives clinical reality
Rentosertib, an oral treatment for idiopathic pulmonary fibrosis, is Insilico’s first programme to reach Phase III. AI helped identify its target and generate and optimise the molecular structure.
A registered trial plans to enrol 320 participants across 47 Chinese centres and compare the drug with placebo for 52 weeks. Recruitment was expected to begin in August 2026, with primary completion estimated for October 2029. It was not yet recruiting when the record was updated on July 7.
The timeline shows where computational acceleration stops. A late-stage study still needs years to observe safety and lung-function outcomes. No Insilico medicine has commercial approval, and industry data has not established that AI-designed drugs succeed more often in later trials.
An earlier analysis reported 80% to 90% Phase I success among AI-native pipelines but about 40% in Phase II, broadly matching its historical comparator. The Phase II sample was too small for a firm conclusion.
Automation is also changing work. Zhavoronkov estimates that about 40% of Insilico’s software-side roles could be automated or displaced, while laboratory scientists and engineers are retrained to manage evaluation and robotics. This is not an announced reduction.
Insilico demonstrates that AI can compress design and candidate selection when paired with fast laboratories. Rentosertib will provide the harder evidence: whether earlier speed produces a safe, effective medicine after the slow and necessary scrutiny of human trials.
