AI adoption is increasing across the pharma value chain, and the slides on AI are relentless, in sales pitches and diligence calls. One is not wrong to assume that an abundance of AI, particularly in pharma drug discovery, should lead or should have already led to bumper revenues at CROs.
The bull case: the revenues are around the corner
During IQVIA’s Q2 earnings call on 28 July 2026, management stated that large pharma clients say AI in discovery will increase demand for CRO services, and are asking IQVIA to gear up. Some large pharma predict they will double their study portfolios and are asking for thousands of FTEs. The drivers include adjacent-therapy molecules where clients lack expertise, capacity without headcount, cost, global footprint, regulatory knowledge, site networks and specialist patient recruitment. This is the sharpest articulation of the thesis anywhere on the tape, but it is management paraphrasing unnamed clients.
The ‘Not Yet’ case
During ICON’s Q2 call on 30 July 2026, an RBC analyst asked CEO Barry Balfe outright whether ICON is in customer dialogue about a coming deluge of molecules and the capacity to absorb it. Balfe rejected the framing and said that reading discovery advances as capacity-determining would be naive, that those advances take years to reach the clinic, and that it will be a while before discovery produces transformational change in development operations.
Molecule volume is real. Molecule volume is not the binding constraint
WuXi AppTec added 839 molecules to its small-molecule development and manufacturing pipeline during 2025, taking the total under contract to 3,452 at year end, and added a further 328 in Q1 2026 alone. The volume premise checks out. The question was never whether molecules multiply. It is whether molecules, rather than capital and development capacity, gate trial starts. Every bull model implicitly assumes molecule count is the input variable. It isn’t.
Molecules vs capital
Discovery activity today reaches a Phase I award perhaps 18 to 36 months out, Phase III well beyond. Trials are funded by sponsors, not molecules. Biotech venture funding recovered in H1 2026 but stayed concentrated, with roughly 76% of tracked capital in $100m+ megarounds and early-stage activity at multi-year lows. A molecule at a company that cannot raise does not become a trial.
The abundance lag
Ben Liu (co-founder and CEO, Formation Bio) in “Uneven Frontiers” (a16z / his Substack, 17 June 2026) and on The Heart of Healthcare podcast (13 July 2026):
- Discovered candidates have roughly doubled over a decade while FDA novel approvals stay flat at ~50 per year.
- Drugs per attractive target have risen from two or three to seven or eight, with 100+ programmes each against PD-1 and GLP-1 in 2025.
- Pre-Phase 2 asset valuations have fallen as the supply of plausible early assets rose.
- The steepest value inflection is Phase 1 readout to Phase 2 readout, roughly a 4x step, with preclinical upfronts in the tens of millions versus hundreds of millions to billions after strong Phase 2.
In a Forbes article (April 2026), he recounts that when he pitched his own Parkinson’s candidates, pharma told him they have more drugs than they can afford to develop, and that a discovered drug isn’t worth that much. He expects discovery to be commoditised by China and AI. Liu is an interested party: Formation Bio’s model presumes exactly this.
AI-derived molecules show Phase 2 success around 40%, in line with the historical benchmark, so the attrition funnel into Phase III is unchanged.
Clinical development, not discovery, is the industry’s bottleneck. More molecules chasing fixed development capacity is a buyer’s market in assets, not a seller’s market in CRO services: the marginal molecule doesn’t become a trial; it becomes an asset that never gets developed or gets licensed cheaper.
When can we see the revenues?
The nearest-term opportunity is nonclinical testing. Many AI-native discovery companies are asset-light on tox and nonclinical capability, and every nominated candidate needs that work from someone. Charles River’s TuneLab arrangement with Lilly, supplying standardised nonclinical testing to AI/ML-enabled biotechs, is the cleanest contractual example in the market today.
Revenue realization timeline: nonclinical and discovery chemistry now, Phase I/IIa awards in 2027, and any Phase III effect in 2028 or later, contingent on biotech capital formation broadening.
The AI-discovery narrative is a valuation premium available today to discovery-chemistry, CRDMO and nonclinical-testing assets. It is not yet available to clinical CROs. Tellingly, the listed names closest to discovery are the weakest cohort in the sector: Evotec and Syngene have both cut guidance, Evotec citing continued softness in early drug discovery.
Sources: