Replicating the "Semiconductor Moment": A "bottleneck deal" in AI-driven pharmaceuticals is taking shape. Who will be the next "shovel seller"?
AI is reshaping the underlying logic of drug discovery, and an investment opportunity similar to the "bottleneck deal" in the semiconductor industry is quietly taking shape in the pharmaceutical supply chain.
On October 4, Freda Duan, a partner at top-tier US dollar fund Altimeter Capital, recently published an article pointing out that a "bottleneck trade" is taking shape in the AI-driven drug development (AIDD) sector, and its structure is highly similar to the chain-like market trend that previously extended from GPUs to HBMs, networks, and power in the semiconductor field.
The collective entry of cutting-edge AI labs is the most significant signal. Anthropic has assembled a dedicated life science research team and established a wet lab; OpenAI has built the GPT-Rosalind model family specifically designed for biology and drug discovery; Anew Labs, an independently operated subsidiary of ByteDance, already has an internal drug development library; and Isomorphic Labs has explicitly stated that it will advance AI-designed drugs into human clinical trials.
Meanwhile, signals from the supply chain are equally clear. DNA synthesizer Twist Bioscience (TWST) expects its AI drug discovery orders to achieve triple-digit percentage growth for both fiscal years 2026 and 2027. GenScript's AI drug discovery business doubled year-on-year in the first half of fiscal year 2026, and channel research shows that its actual validation capacity has reached approximately 8,000 designs per day, and is expected to increase to 16,000 designs per day by the end of 2026.
This logic is highly similar to the bottleneck transmission path in the semiconductor industry. The scarcity of GPUs has driven a successive surge in demand for HBM memory, network devices, and power and cooling systems.
In the field of AI-driven pharmaceuticals, bottlenecks are also being passed down the value chain: AI model generation and design → DNA synthesis → protein production → experimental verification → preclinical testing, with each link potentially becoming the next beneficiary .
(Supply chain data shows that upstream discovery and early preclinical research and development have already begun to feel the impact of increased experimental volume.)
Cutting-edge AI labs are being rapidly established, ushering in a paradigm shift in pharmaceutical discovery.
In her article, Freda Duan describes the current landscape of the AI-driven pharmaceutical industry as a true "regime change."
Anthropic has elevated life science research to a strategic level, and its first major project completed in its wet laboratory was the use of the Claude agent to identify a previously uncharacterized enzyme system associated with CRISPR-like DNA repetitive sequences.
OpenAI's GPT-Rosalind model family covers medicinal chemistry, protein engineering, genomics, and experimental workflows.
ByteDance’s former AI drug discovery division is now operating independently, renamed Anew Labs, which integrates basic models, wet laboratory validation and its own drug pipeline. ByteDance still holds a majority stake.
Isomorphic Labs has explicitly stated that its goal is to advance AI-designed drugs into human clinical trials.
It is worth noting that the commercialization paths of each institution differ significantly.
- Anthropic is currently positioned closer to the research and platform level, and has made it clear that it does not intend to become a traditional commercial clinical-stage biotechnology company.
- ByteDance's Anew Labs is getting closer to becoming a biotechnology platform with its own drug pipeline;
- Isomorphic laboratories fall somewhere in between, with a clear commitment to independently advancing drugs into clinical trials.
This means that the commercialization paths for AI-driven drug discovery are diverging: licensing to large pharmaceutical companies, collaborations with milestones and patent fees, or directly incubating independent companies. For the downstream supply chain, regardless of the path chosen, the increased demand for experimental work will become a real phenomenon.
Why is AI actually increasing demand for wet labs?
AI has improved efficiency, so why has it increased the amount of experiments conducted in wet labs? The answer lies in the fact that AI has fundamentally changed the economics of hypothesis generation.
First, while the cost of assuming drug generation approaches zero, experimental validation becomes a new bottleneck. In traditional drug discovery, the cost of expert screening of candidate molecules is extremely high, and only the highest-confidence candidates can enter the expensive validation process.
AI makes designing candidate molecules almost costless, greatly increasing "opportunities for application" and shifting the bottleneck from the design stage to the biological validation stage.
Second, failed experiments are also valuable, with negative samples serving as training data. Anthropic provides a typical example: Claude designed 1,320 protein binders, and Adaptyv converted these numerical sequences into DNA, expressed the proteins, and tested their binding affinity; ultimately, only 354 successfully bound.
However, the other 966 failed cases were not wasted—what wasn't expressed, what didn't bind, what had poor affinity—these are all valuable negative labels used to train the next round of model iterations. Traditional drug discovery only asks "Is candidate X effective?", while AI drug discovery also asks "What can this experiment teach the model?"
Third, wet labs are transitioning from manufacturing drugs to manufacturing training data. TWST explicitly states that its business is shifting towards generating structured experimental results data from AI-designed sequences that can be directly input into models. GenScript's sequence-to-data workflow can return results in 4 to 7 days.
In some workflows, clients may no longer be concerned with obtaining the physical protein itself, but rather with feeding the structured experimental results back into the model. TWST and GenScript are therefore increasingly approaching the role of a "biological data foundry".
Supply chain data confirms the surge in demand, with production capacity entering a phase of rapid expansion.
The article points out that, in addition to theoretical logic, traceable quantitative signals have emerged at the supply chain level.
TWST expects its AI drug discovery orders to grow at a triple-digit percentage rate in FY26, with similar growth expected in FY27. GenScript's AI drug discovery business is projected to double year-over-year in the first half of fiscal year 2026, with its platform boasting an industrial-grade validation capacity of over 4,000 designs per day and a complete sequence-to-data workflow.
Channel research indicates that actual progress is faster. Current actual production capacity is approximately 8,000 designs per day, and is expected to reach approximately 16,000 designs by the end of 2026.
From the perspective of the supply chain path, the beneficiaries are, in order: AI model → sequence design → DNA synthesis (TWST, etc.) → protein production (GenScript, etc.) → experimental validation → sequencing (ILMN, TXG, etc.) → automated equipment → preclinical testing (CRO).
In terms of market size, if AI is simply a more efficient R&D tool, the related spending pool represents $300-400 billion annually in global pharmaceutical R&D expenditure. The opportunity will be even greater if AI truly expands the number of viable drug projects, as it will also drive incremental demand for DNA synthesis, protein production, experimental validation, and preclinical work.
Furthermore, a rough estimate can be made from the perspective of the AI company's own expenditures. If Anthropic achieves $80 billion in annualized revenue by 2026, investing just 1% of that in AI drug discovery would already represent an annual investment of approximately $800 million.
How is this different from 2020/21?
Freda Duan cautioned that investors familiar with the topic would instinctively draw parallels to the AI drug discovery boom of 2020-2021 and the subsequent crash.
Exscientia's DSP-1181 molecule progressed from discovery to Phase I clinical trials in less than a year, but was subsequently terminated because it failed to meet the Phase I evaluation criteria for the target mechanism. There were two clear reasons for the failure at that time: the macro environment tightened dramatically, coupled with the harsh reality that "faster discovery does not equal better clinical biology".
However, Freda Duan believes that the context of 2026 is fundamentally different from that of 2020/21, as that cycle occurred before the emergence of the GPT.
The core logic for 2020/21 is to use computation to reduce low-value wet laboratory screening. AI is a search efficiency tool with the goal of reducing the amount of physical experiments and identifying drug candidates faster with less synthesis and fewer tests .
The core logic in 2026 is completely different : use AI to significantly expand the hypothesis space, and then use wet labs to generate real-world validation data. The internal cycle is: AI → Experiment → Data → Better AI → More Experiments. The sheer number of usable experiments created by AI makes physical biology itself a bottleneck.
Based on existing data, a 2024 study showed that the success rate of AI-discovered drugs in Phase I trials was about 80% to 90%, higher than the industry's historical average; the success rate in Phase II trials was about 40%, roughly the same as the historical average, but the sample size was still limited.
How will bottlenecks propagate along the blockchain, and what are the risks associated with transactions?
Freda Duan emphasized that current DNA synthesis and protein production are the most immediate bottlenecks.
As more drug candidates are pushed into the preclinical stage, animal testing may become the next limiting factor . Primate prices are already close to previous highs, CRO capacity remains tight, and the further push of AI-driven drug discovery into the preclinical stage will only exacerbate this pressure.
Analysts believe this deal has a long potential lifespan, and it's unlikely to have a fundamental logical conclusion until the Phase I/II clinical trial results are available after 2028. Key milestones:
From 2026 to the end of 2027 , Isomorphic's first batch of AI-designed drugs will enter human clinical trials, and more AI-native projects will advance to the IND-enabling stage;Between 2028 and 2030 , clinical trial results will begin to reveal whether AI-designed drugs are truly superior to traditional drugs.
In the near term, if experimental budgets stop growing, AI-generated designs cannot be translated into effective wet laboratory results, or production capacity catches up with demand too quickly, upstream "picks and shovels" transactions will be under pressure.
In the long run, if AI drugs perform well in the discovery stage and Phase I, but fail in Phase II/III with a normal failure rate, the entire investment logic will face a fundamental challenge, which is why Phase II clinical results are so crucial.
UCLA professor Quanquan Gu succinctly summarized this logic:
Biology may become the next frontier for AI’s recursive self-improvement—designing in computers, testing in labs, turning experimental feedback into semi-synthetic training data, improving models, and repeating the cycle.
Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.