The Wet-Lab Bottleneck
Tina He@tina
· 1
−6.7%
since Sep 30
The idea
I believe AI can now design drug candidates far faster than anyone can actually test them, so the real bottleneck in biology has shifted from ideas to experiments. This basket owns the whole testing stack: lab automation and instruments, contract testing capacity, AI design-test loop companies, per-experiment consumables, and readout tools, since all of them get paid as experiment volume rises. I lean toward names where bookings or volume can move quickly, since I think this plays out over quarters, and I keep 10% in cash to add if there are dips.
What it owns
- Labs facing throughput limits buy machines and tools to run more experiments per researcher, especially as candidate volume accelerates. · Thermo Fisher ScientificIt supplies the instruments and consumables nearly every lab experiment uses, so its growth tracks whether total testing volume is rising., DanaherIt makes the bioprocessing tools that turn a protein design into a physical molecule ready for testing and measurement., BrukerIt makes instruments that reveal what a designed protein actually looks like and whether it behaves as predicted, the ground-truth readout models cannot skip., Agilent TechnologiesIt sells analysis instruments and tools labs buy to increase throughput from existing equipment as experiment volume climbs., Waters CorporationIt provides analysis instruments labs use on every batch, so its recurring revenue directly measures whether total experiment volume is rising.27%
- Outsourced testing capacity becomes the constraint when AI-designed candidates arrive faster than internal labs can validate them. · Charles River LaboratoriesIt runs preclinical testing for outside customers, the exact capacity that becomes scarce when AI-designed candidates flood in., Medpace HoldingsIt runs clinical trials for small biotechs that lack their own trial teams, the exact outsourcing layer that expands when candidate volume grows., ICON plcIt runs clinical trials at scale for large customers, the outsourced capacity where repeatable overflow work flows when candidates multiply.22%
- Companies that operate their own automated wet labs and feed results back into design models prove the feedback loop works before it scales broadly. · Recursion PharmaceuticalsIt operates an automated wet lab that makes molecules and feeds experimental results back into AI models, the design-test-learn loop in real operation., SchrodingerIt provides design software paired with its own testing capacity, the integrated software-like drug development model described in the take., Ginkgo BioworksIt operates an outsourced lab foundry where companies send designs to be made and tested instead of building internal capacity., AbCellera BiologicsIt operates as a testing engine for antibodies, accepting designs from partners and returning experimental data on which candidates work.17%
- DNA, reagents, and supplies are consumed on every experiment, so order volume rises directly with the number of tests run. · Twist BioscienceIt synthesizes DNA on demand, the raw input for building any AI-designed protein, so order volume reflects whether candidates are actually being made., AvantorIt supplies the reagents and consumables consumed in every experiment, so its revenue scales with total test count across all labs., RepligenIt sells consumables for producing and purifying biologics, so every extra AI-designed protein built requires more of its product.16%
- Every experiment ends in a measurement, and high-quality readouts like sequencing and single-cell analysis become the training data that AI models learn from. · IlluminaIt sells sequencing that converts experiments into data, the highest-fidelity readout for training biology models., 10x GenomicsIt sells single-cell sequencing, the richest experimental data available for training and benchmarking biology models.9%
- Cash10%