AIPher raises ¥100M angel for AI that owns drug-R&D decisions

AIPher (艾斐智药; 36Kr's English report calls it AIPharma Intelligence) is an AI-native drug-discovery company founded in July 2026 from the Tsinghua AIR research group of Professor Lan Yanyan and Academician Zhang Yaqin, with ByteDance-algorithm veteran Gao Bowen as core member. Its stack has three layers: DrugCLIP, a two-tower contrastive model that embeds protein pockets and small molecules in one high-dimensional space for genome-scale virtual screening; PharmAgents, an engine that weighs conflicting, incomplete evidence and proposes the next experiments; and a learning loop that feeds wet-lab results and expert corrections back into the system.

Make AI the decision-maker in drug R&D, not a tool: build the system from scratch, verify agents in wet labs, and get paid through co-development, not services.

Angel round at the 100-million-yuan level announced 2026-09-03, led by Xianghe Capital with Xinglian Capital — the venture arm of Zhipu AI, per 36Kr's Chinese report — and an unnamed industrial fund participating. Proceeds fund DrugCLIP and PharmAgents R&D, next-generation AI4S research, and the independent and partnered pipeline.

Gao Bowen worked on recommendation algorithms at ByteDance, returned to Tsinghua AIR in 2022 for a doctorate, and co-authored the Science paper behind the DrugCLIP line of work. The company recruits for the closed loop from day one: R&D leads with medicinal chemistry and structural biology backgrounds, and a business-development head with nearly 20 years in pharma who has moved multiple drugs into Phase II/III.

AIPher casts itself as the third generation of Chinese AI pharma: first-generation companies solved single-point tasks with AI, second-generation ones pursued full-process AI with automated labs, and this generation lets AI take responsibility for R&D decisions. It rejects the relay model where AI computes, medicinal chemistry executes and biology receives, and also rejects pure AI technology services, choosing co-development deals with milestone payments, license-out and revenue sharing so it earns only as pipelines advance.

As of early September 2026 the angel round is announced, the lead pipeline sits at Pre-PCC, and the team is using the capital to iterate DrugCLIP and PharmAgents while pushing independent and collaborative pipelines. No revenue or clinical milestones are disclosed.

A blank-sheet organization escapes the legacy structure where every modality and disease business line builds its own siloed AI models that cannot learn from one another.

The Science-published DrugCLIP work and Tsinghua AIR pedigree — plus a Zhipu-linked investor in the round — signal that the foundation and network already exist.

Day-one wet-lab, medicinal chemistry and pharma-business talent, plus five-plus co-development partners, answers investors' demand for reproducible proof over papers and demos.

Milestone-payment and license-out co-development aligns the company's earnings with pipelines advancing, in a track crowded with startups selling AI services whose value is hard to verify.

Org design is the moat: a startup built from scratch can make AI the decision-maker where incumbents with tool-embedded lines cannot — provided wet-lab proof and expert review ride along from day one.

As of 2026-09-06 AIPher is two months old with a 100-million-yuan-level angel announced and no further public developments: no revenue, clinical or hiring disclosures beyond the round statement. The company's claims about DrugCLIP throughput and hit rates are self-reported and unverified by independent tests, and Pre-PCC is far short of clinical evidence; like every AI-drug startup it faces the long, expensive road from screening to a marketed medicine.