Analysis: AI drug development shifts from model race to closed-loop ecosystem; Anthropic, ByteDance build wet labs

Recent moves by Anthropic, ByteDance, and other technology giants signal a strategic pivot in AI-driven drug discovery—from competing on model capabilities to building integrated experimental ecosystems.

From General Tools to Laboratories: Building Systemic Infrastructure

Anthropic has established a wet laboratory in the San Francisco Bay Area. ByteDance's AI drug discovery subsidiary, Anew Labs, has completed its first round of external financing as an independent entity, raising $290 million. Eli Lilly is also among the players expanding in this space.

According to interviews conducted by Shanghai Securities News with executives from multiple listed AI drug discovery companies and industry insiders, the competitive focus is shifting toward "who can build industry-wide systemic infrastructure" and successfully establish a complete closed loop of "data—model—experiment—data."

Previously, tech giants focused on general-purpose large language models. Now, many are extending their capabilities from the digital world into physical laboratories to close the experimental loop.

Ma Jian, co-founder and CEO of XtalPi, noted that the barrier to entry in AI drug discovery extends beyond models. Pure computation and prediction in AI-driven drug design (AIDD) are no longer sufficient.

From the tech giants' perspective, extending capabilities into drug R&D laboratories is a natural progression. Mao Hua, senior partner and managing director at Frost & Sullivan China, explained that as computing power becomes more abundant and large models iterate rapidly, functional differences between models are narrowing. For tech companies to build competitive advantages in pharmaceuticals, they must secure access to automated experimental platforms, proprietary data assets, and pipeline intellectual property.

Lai Caida, co-founder, chairman, and CEO of METiS Pharmaceuticals, pointed to Anthropic, OpenAI, and Nvidia as key players driving this trend.

Demand-side dynamics are also evolving. According to securities analysts, large pharmaceutical companies are increasingly integrating AI into drug discovery, preclinical research, and R&D workflows. AI investment is transitioning from experimental "trial use" to a standard line item in R&D budgets. The division of labor is becoming clearer: tech companies provide models and computing power, pharmaceutical companies contribute scenarios and pipelines, and experimental capabilities are a shared area for reinforcement.

However, the gap between capability upgrades and commercial returns remains. Zhang Chenyang, senior consulting director at Frost & Sullivan China's Healthcare Division, noted that while large models are evolving from辅助 prediction to autonomously generating hypotheses, designing molecules, and calling upon wet-lab experiments for validation, it remains to be seen whether clinical success rates, R&D timelines, and return on investment will materialize.

From Model Competition to Closed-Loop Competition: The Data Flywheel as the True Moat

Compared to the previous wave of AI drug discovery enthusiasm, the driving forces have changed—and so have the evaluation criteria.

Lai Caida emphasized that AI can generate hypotheses faster, but whether those hypotheses hold up ultimately depends on real-world experiments, animal studies, clinical trials, and long-term medical validation. Therefore, the priority going forward is not simply increasing computing power, but building infrastructure that connects AI with real biological experiments.

The shift in evaluation criteria has also redirected competitive positioning toward the true bottleneck. Zhang Chenyang stated that while high-quality data appears to be a surface-level bottleneck, the real challenge lies in establishing an experimental system that continuously generates high-quality data and drives autonomous model iteration—particularly standardized wet-lab experiments, negative result data, and a "prediction—experiment—feedback" closed loop.

Ma Jian added that what the AI drug discovery industry truly lacks is a "data flywheel combined with an engineered dry-wet lab closed loop."

Ren Feng, co-CEO and chief scientific officer of Insilico Medicine, believes that independent AI drug discovery companies have increasingly clear advantages: automated dry-wet lab closed loops with high-quality data, validated AI innovation capabilities, continuously expanding areas of innovation, and deep large-model training platforms.

Multiple industry insiders believe that the key competitive differentiator is shifting from individual AI tools to the ability to form a complete closed loop of "data—model—experiment—drug and clinical validation." Whoever successfully runs this loop holds the "ticket" to the next phase.

From Going It Alone to Ecosystem Co-opetition

How does the influx of tech giants into AI drug discovery affect the industry? The consensus among insiders is clear: this is not a zero-sum game, but an ecosystem reshaping.

Ren Feng predicts that all biotech and even pharma companies will eventually be AI-enabled, and the entry of tech giants only accelerates this trend. AI drug discovery companies and tech giants are not direct competitors. The optimal model is for each to focus on their strongest areas: large tech companies provide general-purpose foundation models, computing power, and basic algorithmic capabilities, while AI drug discovery companies leverage their expertise in biology, wet-lab experience, and clinical validation. Together, they can achieve a 1+1>2 effect.

Notably, the division of labor is shifting from individual efforts to ecosystem co-opetition. Tech giants contribute foundation model capabilities and computing power; global pharmaceutical companies bring clinical and commercialization resources; and independent AI drug discovery companies offer data flywheels and dry-wet lab closed loops. These three parties collaborate through co-building, co-creation, and licensing models to form a new industrial ecosystem.

Looking ahead, Ma Jian believes the trend toward industry consolidation will continue. On one hand, AI drug discovery has entered the clinical validation phase, where moats are shifting from individual models to systemic capabilities. Companies lacking data accumulation and engineering closed loops will face greater challenges. On the other hand, advances in model and experimental capabilities are lowering barriers in vertical domains, with opportunities moving from "general platforms" to "deep verticals."

Industry evaluation standards are also changing. Insiders note that the criteria are shifting from valuation and narrative to data and milestones. In this sense, tech giants entering with capital, computing power, and talent are acting as "catfish" that energize the entire industry, accelerating AI drug discovery from concept validation to value realization.

(Source: Shanghai Securities News)