Biology: Finding the Target.
Open-source data and scientific models widen the search for a credible biological hypothesis.
SR ONE · PERSPECTIVES
The New Economicsof Drug Creation
Our playbook for what matters in the era of AI and China.
I
Where We Are
Biotech is entering a new phase.
The sector spent the better part of three years working through a capital hangover — too many platforms, too many narratives, and not enough drugs. US policy uncertainty acted as a headwind. Then the IPO window cracked open in late 2025 and widened further in 2026. M&A is returning (2026 has seen $90B of deal value already in the first half of the year, ahead of 2025 and 3x 2024's figure)1,2, and large pharmaceutical companies are deploying capital again as a wave of patent expirations approaches.
Two structural forces are reshaping how medicines get made: the maturation of AI in drug discovery, and the rise of Chinese biotech as a cheaper and faster development engine and competing source of innovation. Developed correctly both promise better drugs, faster timelines, lower cost. But based on what we observe within our portfolio and in the ecosystem more generally, both trends carry real execution risk and their impact is uneven.
II
AI Is Real
But value still accrues to drug hunters.
Machine learning and early versions of AI have been used in clinical development for decades. What is new is the combination of large genomic and health datasets, purpose-built scientific models, and far greater computing power. Together they allow researchers to make sense of data that was previously too large, too messy, or too fragmented to yield much insight. That is compressing the iterative loops that define drug discovery, and should make companies faster and cheaper to run.
But in our experience, the promise applies unevenly across the three components of a successful drug: biology, chemistry and clinical development. In places it's overblown. In many settings new models trained on disparate data sets can synthesize and hypothesize more broadly than any human, but no one is letting these same models make the key decisions. Indeed, in many cases their true predictive power is still lower in some cases than the basic statistical methods that preceded them.3 That said the frontier is clearly moving at express pace and the impact in terms of efficiency is already being felt for our portfolio.
01
Biology
Finding the Target
Before a company can pursue a drug, it needs a biological target: a protein or pathway believed to drive a disease, and one that can actually be reached with a molecule once found. Finding and validating new targets, particularly for diseases that have resisted treatment so far, is painstaking work, built on cross-referencing published studies, animal and cell experiments, and the judgment of clinicians and specialists.
An explosion of open-source datasets and models, both general-purpose language models and specialist scientific tools, now lets researchers sift through genetic and patient data, often organized into disease knowledge graphs at a scale no person could match by hand. AI's role here is mostly pattern-finding: compressing the time target identification takes, and occasionally surfacing connections that would otherwise go unnoticed.
Historically available data
Additional data sources
MODEL ORCHESTRATIONLLMs + specialist models
CLAUDE SCIENCEHighlighted source coverage
Where to act?
Target hypothesisRequires validation
COMPARE → REFINE → REPEATComputational iterationRings mark candidates · dots mark evidence
02
Chemistry
Designing the Molecule
Once a target is chosen, the task becomes designing a molecule that finds it, binds to it strongly, and produces the intended effect while minimizing side effects — in a format that can be made at scale on commercially reasonable terms. Designing a drug has traditionally meant screening large libraries of chemical compounds, or raising and refining candidate protein binders, then grinding through repeated rounds of redesign: tweaking the molecule, resynthesizing it, retesting its potency and selectivity, often for months on end.
Generative design and structure-prediction tools can now propose and screen candidate molecules computationally, including against targets long considered too difficult to drug by conventional means. This is where the bulk of early AI investment in the sector has gone, in part because of the sheer volume of chemical and structural data available on which to train models. The results of these approaches show real promise in specific areas, not least work Anthropic published last month in which they were able to combine frontier LLMs with specialist models and compute to autonomously design protein binders for 16 genetic targets, frequently outperforming the best human-derived efforts when independently tested in a wet lab.
03
Clinical
Testing in People
Finally, once a candidate molecule exists, it has to be tested in people. This is the valley of death where the failure rates rocket. The critical decisions here: which patient cohort to test, where to find those patients, and which endpoints to measure, have so far seen less structured use of AI. In our experience, that is now changing.
Big Pharma and biotech are now using models to interrogate patient records and genetic databases to find and match the right patients faster, and to build sharper ways of measuring whether a drug is actually working. A growing number of companies are using patient data from multiple sources to identify trial sites and patient groups, in an effort to cut the time it takes to enroll patients and thereby reduce time to data. We think things will go much further still in this critical area — improving decision-making and saving time and money across both how trials are designed and run.
The immediate goal across all three areas, biology, chemistry and clinical development, is to make each stage faster and improve the quality of decisions made at every step.
The more distant promise is that these tools become reliable enough — in effect such good predictors of the real world — that they can substitute for elements of lab or clinical work altogether. But this remains some way off. Regulators will take some convincing that insights that are only developed “in-silico” are robust enough to pass a patient safety hurdle. Most of what fails in drug development fails in the clinic, for reasons that have little to do with how quickly a target or molecule was found. LLM orchestration of big data sets and specialist models has clearly compressed the earliest stages of discovery. They are increasingly commoditizing bits of the discovery process. These same tools are beginning to support clinical trial design and execution. But we have not yet seen definitive proof that it can increase historically low probability of success for a drug in early clinical studies and do it with real time and cash savings.
Our AI Principles
Things are moving fast. To help structure our thinking, we use three basic principles to guide our investment decisions and our portfolio companies.
Taste over tools
AI platforms are still no substitute for the muscle memory and “taste” cultivated by true drug hunters. We have watched several companies invest heavily in what they take to be differentiated knowledge graphs and AI-enabled data platforms, built with genuine technical skill — too often with little idea how to turn the tool into a product that patients need and buyers will pay for, or elaborate the true predictive quality of the tool in a way that engenders trust. The tech-first approach only works when it is married to a team steeped in the unglamorous business of drug development — people who understand what Pharma actually buys, and can carry a molecule the distance.
Close the loop
The ultimate challenge is integrating AI tools across the full arc of drug development — from early target identification through to late commercial analysis — and then rigorously testing their impact where it matters: improving success rates in the clinic. Discovery, development and commercial functions have always aspired to inform one another; in practice the loop between them has been slow and manual. AI so far has mostly been layered onto that same sequence rather than closing the loop: a target-identification model here, a molecule-design platform there, each answering its own narrow question competently enough. That approach will not, in our view, materially shift the sector's stubbornly low probability of success on its own — though it may save time. What can have a bigger impact is a genuinely integrated architecture, one in which commercial and late-stage thinking feeds back into discovery-stage design as the programme runs, rather than being consulted once at the start and again at the end.
Stay hybrid
The thorniest question is how a biotech company keeps its edge as scientific insight becomes ubiquitous. Tech investors have long treated the absence of a proprietary dataset or equivalent moat as a warning sign, and biotech is asking itself the same question — pushing some founders toward expensive, build-it-yourself platforms that the logic doesn't always fully support. Building tools is hard, the expertise required is costly, and the build phase carries a real distraction “tax” that has so far correlated poorly with actually producing good drugs. The tools vary by category: open-source protein-design tools can be extremely impactful, deeply referenced knowledge graphs are proliferating at pace, but decision-making tools are not yet at the same point. The most successful approaches may be hybrid ones — companies jealously guard the orchestration frameworks that integrate different agents and analytical tools, and the routes by which they test those insights in labs and patients, while staying relaxed about where the underlying tech and data come from.
III
Capital Intensity and the Rise of China
While AI occupies the most column inches at present, a second, equally important structural shift is reshaping the economics of biotech company building: the globalization of drug development and innovation.
Over the past decade, China has built out its biotech infrastructure through patient regulatory reform, investing in scientific talent and massively expanding clinical trial capacity in centers with patient populations that the West can only imagine. They have done this at a pace that has produced a development system capable of running comparable clinical programs significantly faster and cheaper than its Western counterparts. By our count on our most recent visit, there are now more than 1,000 Chinese biotechs (still fewer than in the U.S., but rapidly closing the gap), many with multiple “me-better” and genuinely innovative programs in the clinic. Ambitious Chinese firms have by some counts now reached parity with the US in the number of innovative molecules in development. Centralized trial operations and population scale drive real savings — cell and gene therapy patients can be enrolled at Zhongshan Hospital in Shanghai for roughly $7,000 per patient, about a tenth of the equivalent U.S. cost.4
China’s advantages are cost, speed and scale
China vs. the US: drug R&D timelines and costs across stages.
| Measure | Drug discoveryHit to PCC | Preclinical dev.PK, Tox, CMC | INDIND to FPI | Pre-PoCPhase I–II trials | Post-PoCPhase III pivotal | NDA/BLAAccelerated review |
|---|---|---|---|---|---|---|
| Timeline | 50–70%of US time | <70%of US time | +1–2 moextra vs. US | >2×faster enrollment vs. Phase Ib | <70%of US enrollment time | +5–6 moextra vs. US |
| Cost | 20–30%of US cost | <50%of US cost | — | <50%of US cost per patient | — | — |
Faster / lower cost than USSlower / added time than US
The Licensing Wave
Pharma has taken notice.
Early doubts about the integrity of Chinese-generated data have gradually reduced, though rigorous patient by patient diligence is still being undertaken. The pace at which pharma has struck multi-target, multi-asset, multi-year deals with large, reputable Chinese biotechs is striking. Early 2026's wave of mega-deals illustrates the emerging model: AstraZeneca, Bristol Myers Squibb, Pfizer, and GSK have signed roughly $56.7B combined in licensing deals with Chinese biotechs in recent times. All four follow the same structure: pharma leverages Chinese biotechs like Hengrui and Innovent's vast platforms for the option on assets that are developed cheaply and quickly through early human data in China. In private conversations with us, pharma executives talk grimly about how these emerging Chinese players can outrun their own in-house R&D in some areas.
The effects show up clearly in licensing activity as shown below. Demand has pulled in a wider range of Western players. Big pharma and biotech are now joined by a raft of newcos — several in our own portfolio — set up by venture capital firms for the sole purpose of in-licensing Chinese assets. Competition has now moved earlier towards Phase I and preclinical stage assets. Unsurprisingly in this climate, average upfront payments for ex-China rights have followed suit: up from $52M in 2022 to $172M in early 2026.5
As the arbitrage of buying direct from China softens and geopolitical and regulatory uncertainty persists — from U.S. legislative moves against Chinese innovation and API tariffs, to tightening FDA standards on Chinese-generated data — new models of engagement between the Western and Chinese ecosystems are emerging. In our view, the old playbook of going it alone in China, sourcing early assets there and running them in the West, will become an increasingly tough road to tread for investors and biotechs. We think investors and operators will increasingly look for more integrated and carefully constructed ways to manage working with Chinese partners, for instance, using U.S. or European platforms and wet labs to generate the IP and early scientific insight, then leaning on China for preclinical and early clinical development.
For Western biotech, this is both a radical challenge and an opportunity. A company developing early clinical assets on traditional U.S. timelines and cost structures is increasingly competing against alternatives pharma can access faster and more cheaply elsewhere. Capital efficiency – always a key theme in biotech – has become even more important. For us as investors, more structured local partnerships globally — giving enduring access to Chinese innovation and early-stage science at lower cost — will, we think, become increasingly important alongside the traditional biotech model we continue to pursue in US, Europe and Japan.
IV
The Emerging Playbook
For a while the investments we made in AI-enabled platforms or in new companies in-licensing single assets from China were done as separate strategies focusing on the same goal: lowering cost and shortening the path to a credible drug. An AI-platform promised it could unlock biology no-one has drugged before or design a candidate molecule quicker than ever before. An asset-focused newco had credible early data in China and a clear plan to try and replicate it in a US trial. We viewed both as having different risk profiles. In the case of a platform play it’s often the time and spend required to build and maintain the platform and the risk that no products of value arise; for a newco developing someone else’s asset, the questions is can you locate and agree terms for a differentiated asset on terms that make a venture return for your investors.
Our thinking has now shifted. In the coming year or two as the AI tools proliferate and the scale of the China cost advantage persists, we don't think that this separation of approach holds. The companies that create durable value over the next decade will fuse them: AI-native discovery paired with increasingly global clinical execution optimized for capital efficiency and speed. Done right, this could be genuinely game-changing for both the odds of success and the cost of getting there. But you need a clear plan as to how to succeed across these various areas.
For Our Companies
In this fast-evolving world, the business model we most like to invest in combines:
Drug-hunter-led decision making
The right experience is needed to orchestrate an emerging suite of relevant tools and automations, integrating clinical and commercial strategy from day one to make the right calls.
Financial savvy
A relentless focus on lowering your costs and your cost of capital (often using partnerships with Pharma to subsidize platform plays and lean, fast trials when you have an asset in hand).
Fast enough to compete
A scientific development plan that is globally competitive on time, leveraging international opportunities where it makes sense. Taking more than 3 years from idea to the clinic probably doesn't cut it anymore.
Product focused
A clear, differentiated path to a data set that patients and Pharma are going to care deeply about.
The infrastructure to deliver these things already exists. Some companies in our portfolio and the wider ecosystem are already assembling the pieces. What's missing is the discipline to combine them with impact — most are still pulling one lever, not all of them. Fusing them is hard. It takes real technical depth on the AI side, enough drug development scar tissue in management to know when the AI is barking up the wrong tree, and the ability to run clinical trials across multiple countries, including China — all against a challenging macro backdrop for global development.
For Us as Investors
We see AI and China as genuine tailwinds for the sector — a real opportunity to raise the odds of success and cut the cost of making drugs that matter. But the disruption they bring is real, and the returns will not be shared equally. The next phase of biotech may be defined less by who can discover the most molecules and more by the firms that are best at turning the growing avalanche of molecules and ideas into medicines at scale. In a world where the challenge is moving from ideation and discovery to deciding what and how to get into the clinic, we believe the enduring qualities that will differentiate investors in this space are:
- People. Be a partner of choice to the very best drug hunters. These people are still the catalyst of our business and provide the “taste” needed to cut through the scientific noise and turn ideas into medicines.
- Pharma access. Deep connectivity and insight with Pharma at all levels, both to position the portfolio for M&A and to invest ahead of what future buyers will want.
- Build a sustained China play. An investment platform with systematic partnerships and infrastructure in China that allows you to test your best ideas at a lower cost and participate with meaning and intention across its ecosystem. Firms need to be set up for a multipolar world.
- Deep expertise in clinical execution. Putting drugs into patients is messy and unpredictable and will remain the key bottleneck in our business; firms that systematize and learn across their portfolio — building new AI tools where it helps — will have an enduring edge.
- AI signal over noise. A carefully calibrated decision-making architecture that allows you to utilize models for specific sets of analysis in ways that demonstrably improve your predictive track record and save time. This involves having a rock-solid understanding of the bits of your process where it can augment decision-making and where it can't.
The firms that fuse these edges — people, global clinical infrastructure, and intentional AI-supported investment judgement — will define the next era of biotech investing. We strive to be one of them.
Notes & Sources
- IQVIA, “Biopharma M&A Mid-Year 2026 Update,” Figure 2. H1 2026 is a half-year actual, not annualized.
- Biotech-only M&A as of June 9, 2026. Source: SR One analysis.
- Gao et al., Nature, 2026.
- Cantor Fitzgerald: China Biotech, fieldtrip notes, June 2025.
- Source: Evaluate data, as reported in Fierce Biotech, “After 230% deal size explosion, China is no longer the ‘bargain basement’ for biopharma licensing: analyst” (Feb 2026).
- Platform share of U.S. VC biotech funding by cycle, 2010–27. Source: BCIQ/LEK, via BioCentury. 2025–27 is forward outlook, not observed data.
- China vs. US drug R&D timeline and cost comparison by stage. Source: BCG, “Beyond Efficiency: China's Next Leap in Biopharma Innovation,” 2026; BCG survey and analysis.