AGI Investment Tracks · Biology & Cloud Labs

The Wet-Lab Bottleneck for AGI Biology

Analyst: biology-cloud-labs · For: Ravi · Date: 2026-05-26 · All prices/market caps at market close 2026-05-26 unless noted.
Premise. AGI is happening. Cognition is becoming abundant. Biology is the highest-value information domain humans have, and it is gated by the rate at which physical experiments can be performed: synthesize, express, perturb, measure, repeat. The companies that own the "physical compiler" for AI-designed biology are the ones whose multiples should re-rate first.

1. Thesis

A frontier model can already enumerate 10,000+ candidate proteins or 106 small molecules in seconds. None of that becomes value until atoms move: DNA must be synthesized, cells must express, assays must run, bioreactors must ferment, GMP manufacturing must release lots. The hard, slow, capital-intensive parts of this stack — long-read sequencing, oligo/gene synthesis, single-use bioprocessing, contract biomanufacturing, lab automation hardware — are the physical compiler for AGI-designed biology. They are largely an oligopoly, were brutally re-priced in the 2022–2024 bear market, and are now mid-recovery while the cognitive layer above them goes vertical.

Isomorphic Labs' May 2026 $2.1B Series B (Thrive-led, Alphabet/CapitalG/GV) is the loudest market signal: the smartest AI lab in drug design wants to spend ten figures on outsourced wet-lab cycles. They need a working physical backend. Recursion, Schrödinger, Generate, Xaira, Iambic, and the OpenAI biology efforts all share this constraint. The picks-and-shovels names sitting between AI compute and clinical biology are the leveraged trade.

2. The bottlenecks (quantitative)

LayerToday's binding constraintWhat AGI demand multiplies
DNA / gene synthesis Twist ~$0.07/bp for short oligos, $0.10–0.30/bp for genes; multi-week lead times for libraries >10kb. Twist FY26 guide ~$445M revenue (small relative to demand). Protein design at scale needs millions of synthesized constructs per project. Demand is supply-elastic.
Long-read sequencing PacBio just shipped SPRQ-Nx: sub-$300 HiFi genome, ~30% lower cost. Still 3–5x Illumina short read for the same depth. PACB FY26 guide only $165–175M. AI training data for genomics/structural variants is long-read native. PACB+ONT capacity is a floor.
Short-read sequencing Illumina dominates installed base; competitive pressure from Element AVITI, Ultima UG100, MGI on $/Gb. ILMN raised FY26 guide to $4.52–4.62B. Population-scale biobanks (UKBB, All-of-Us, China precision medicine) demand 10x throughput. Marginal capacity matters.
Lab automation hardware Tecan TTM rev CHF 882M (down 5.5% YoY, net loss); Hamilton private; Beckman inside Danaher. Lead times 6–12 months for high-end liquid handlers. Each AI-bio company building in-house automation (Recursion, Insitro, Xaira) buys here.
Single-use bioprocessing Sartorius + Repligen + Cytiva (Danaher) + MilliporeSigma own ~80% of single-use bags, filters, chromatography resin. Lead times for protein A resin still 6+ months. Every biologic, every cell therapy lot needs them. Demand grows with both AI-discovered biologics AND GLP-1 fill-finish.
CDMO biologics capacity Lonza, Samsung Biologics, Fujifilm, WuXi Biologics committed >$3B in capacity expansion. Samsung Q1 2026 revenue KRW 1.257T (~$849M). Lonza Stein facility online 2026, Samsung Songdo to 360kL by 2030. AGI-designed biologics still need GMP fermentation. Bookings are already 3–5 years out at top-tier shops.
Cloud / automated labs Emerald Cloud Lab, Strateos, Arctoris are still small, private, capacity-constrained. ECL CMU partnership real, but throughput << what AI demand wants. The most direct "cognition → experiments" pipe. Severely undersupplied.
Mass spec / proteomics Bruker (timsTOF), Thermo (Orbitrap), Waters, Sciex (Danaher). Bruker TTM $3.46B; Waters $3.77B. Lead times stretching. Protein-level validation of AI predictions = proteomics-heavy workflows.

3. Supply / demand gap model

Rough mental model for 2026–2030:

Input2026 baseline2030 demand (AGI-on)Implied multiple
Synthetic genes shipped (bp/yr, industry)~10121014+100x
Sequenced genomes/yr (industry)~5–10M100M+ (incl. cell-free dx, longitudinal)10–20x
Cloud-lab experiments/day (global)103–104106+100–1000x
Biologic GMP fermenter capacity (kL)~5,000 (global)15,000+3x (capex+regulatory bound)
Lab automation install base ($)~$10B/yr cap$30–50B/yr3–5x

Estimates — my synthesis from industry reports, not company filings. Sequencing and synthesis numbers are first-principles "AI-bio will need this" estimates, not analyst consensus. Treat as order-of-magnitude.

The honest take: CDMO capacity is rate-limited by physical buildouts (5–7 years per facility, FDA inspections, talent). Synthesis and sequencing can scale faster but the incumbents have moats. Cloud labs are the most elastic but tiny and mostly private. The closest things to "TSMC of biology" are Twist (synthesis), Sartorius+Repligen (bioprocess consumables), and Lonza (CDMO).

4. Investable public companies

TickerNameMkt Cap (live)TTM RevMult. Why bottleneckRisksPriced-in?
TWSTTwist Bioscience $4.04B$409M9.9x PS Dominant gene-length DNA synthesis. Single biggest input to AI protein design at scale. FY26 guide $442–447M, +17% YoY, Q4 EBITDA breakeven. Ansa enzymatic synthesis could disrupt 5+ yrs out. Stock already +107% YoY. Partly
RGENRepligen $6.50B$763M8.5x PS / 127 PE Protein A resin + filtration: the bottleneck input to every monoclonal antibody and biologic lot. Recovering: TTM rev +17.4%. Expensive on earnings. Tied to biotech funding cycle. Partly
SRT3.DESartorius €14.36B€3.55B~4x PS / 88 PE The European Repligen+Cytiva equivalent. Single-use bioreactor leader. Mid-term guide 8–11% organic growth. P/E still pricey after the de-rate. Pharma capex slow to recover. Partly
LONN.SWLonza CHF 34.3BCHF 6.53B5.3x PS Largest Western biologics CDMO. Stein expansion online 2026. Booked years out. Plays the GLP-1/biologic capacity squeeze and the AI-designed-biologic future. Currently unprofitable TTM (restructuring). Forward P/E 27 not cheap. Partly
DHRDanaher (Cytiva + Beckman + Sciex + IDT + Leica + Pall) $122B$24.8B5.0x PS / 33 PE The conglomerate ETF for life sciences tools: Cytiva (bioprocess), IDT (oligos), Beckman (auto), Sciex (mass spec), Leica (microscopy). Owns 4 of the 8 layers above. Mega-cap, modest growth. Diversification dilutes pure AI-bio leverage. Cheap-ish
TMOThermo Fisher $167B$45.2B3.7x PS / 25 PE The Walmart of bio tools + cryo-EM (FEI) + reagents + CRO (PPD). The default "buy the index" of bio infrastructure. Hard for the elephant to dance — growth muted vs. focused names. Reasonable
ILMNIllumina $22.0B$4.39B5.0x PS / 26 PE ~70% global short-read sequencing share. NovaSeq X drives $/Gb floor. FY26 guide raised. Element, Ultima, MGI, Singular pressuring on price. Long-read share migrating to PACB/ONT. Cheap
PACBPacific Biosciences $391M$160M2.5x PS Long-read leader; SPRQ-Nx shipping sub-$300 HiFi genome. Trading near multi-year lows. Tiny, unprofitable, capital constraints. ONT real competitor. Cheap, optionality
SDGRSchrödinger $983M$255M3.85x PS Physics-based + AI molecular design. Software business is the only AI-bio infra public-market pure-play. 23% TTM revenue growth, narrowing losses. Equity stakes in clinical programs add binary risk. Hasn't broken out commercially. Cheap
BRKRBruker $7.14B$3.46B2.1x PS Cryo-EM, mass spec, NanoString (spatial), Berkeley Lights (single-cell). Hardware moats in 3 separate biology-validation modalities. Flat revenue YoY. Indigestion from acquisitions. Cheap
WATWaters $33.3B$3.77B8.8x PS / 49 PE LC-MS gold standard. AGI biology = orders of magnitude more validation chromatography. Already richly priced. Slow growth. Mostly
AAgilent $32.5B$7.07B4.6x PS / 25 PE Mass spec + LC + reagents + CRO services + diagnostics. Solid second-tier exposure. Less pure-play than peers. Diversified across cyclical end markets. Reasonable
RVTYRevvity (ex-PerkinElmer) $10.8B$2.90B3.7x PS Reagents + dx + automation. Post-divestiture of diagnostics-China, leaner. Soft FY26 guide. Multiple downgrades. Cheap
TECHBio-Techne $7.51B$1.21B6.2x PS / 69 PE Specialty reagents (R&D Systems): the input to every academic AI-bio lab. Pricey on PE. Flat top line. Mostly
TXG10X Genomics $3.13B$639M4.9x PS Single-cell + spatial. New Atera platform. Bounced hard off lows (+194% YoY mkt cap). Competition from Bruker/NanoString. NIH funding overhang. Partly
MRVIMaravai LifeSciences $1.13B$205M5.5x PS TriLink (CleanCap, modified nucleotides): the unsung input to every mRNA therapeutic and now AI-designed gene therapy. Q1 +41% YoY. Volatile, post-COVID hangover. Highly sensitive to mRNA cycle. Cheap on recovery
TECN.SWTecan CHF 1.87BCHF 882M2.1x PS Liquid handling robots: the substrate of automated biology. Trading at 5-yr lows after revenue dip. Currently unprofitable. Demand soft, but cyclical floor. Cheap
RXRXRecursion Pharmaceuticals $1.58B$66M24x PS In-house automated wet lab (Salt Lake). Owns the closest US analog to a cloud lab. NVIDIA partnership real. Drug pipeline mostly preclinical; burning $645M/yr. Stock at all-time lows. Hold-rated. Bombed-out
ABSIAbsci $823M$1.8M447x PS (lol) AI antibody design + in-house wet lab. ABS-201 in P1/2a. Effectively a biotech, not infrastructure. Don't size as infra. Binary

Multiples: PS = price/sales (TTM), PE = price/earnings. Live data as of 2026-05-26 close from stockanalysis.com.

5. Pre-IPO / private to watch

6. Commodity / physical plays

7. People / talent concentration

8. Top 5 picks ranked

1. Twist Bioscience (TWST) — Conviction: HIGH

Closest thing to "the TSMC of biology." Effectively monopoly economics on silicon-based oligo + gene synthesis at the volumes AI biology needs. FY26 guide $442–447M, +17% YoY, Q4 EBITDA breakeven means the unit-economics fight is largely won. Up 107% YoY but at 9.9x PS still light-years cheaper than any software AI proxy. If Isomorphic, Xaira, Recursion all 10x their experimental throughput, this is one of two suppliers in the world (Twist + IDT-inside-Danaher).

What I would pay for: Demonstrated Q4 adj-EBITDA breakeven, then size up. Already a buy here.

2. Sartorius (SRT3.DE) + Repligen (RGEN) — Conviction: HIGH (as a pair)

The bioprocess consumables duopoly. Every biologic, every cell therapy, every AI-designed drug substance flows through their bags, filters, and resins. Both got crushed 70% from 2021 highs and are 18–24 months into recovery. Sartorius mid-term guide 8–11% organic. RGEN TTM rev +17%. The PE looks optically expensive because earnings are emerging from a trough — this is exactly the wrong way to value an operating-leverage recovery.

What I would pay for: Sartorius is more capacity-leveraged (broader portfolio); Repligen is more direct mAb-exposure. Own both.

3. Danaher (DHR) — Conviction: HIGH for size, MED for upside

The unfair ETF: Cytiva + IDT + Beckman + Sciex + Pall + Leica + Aldevron inside one ticker. 4 of the 8 bottleneck layers above are inside DHR. At 5x PS / 33 PE for a $122B compounder with multiple monopoly-positioned subsidiaries, the risk/return is exceptional for "I want exposure but don't want to pick the right small-cap." Use as core; complement with the focused names.

What I would pay for: Already reasonably priced. Buy a starter, add on any pharma-capex scare.

4. Schrödinger (SDGR) — Conviction: MEDIUM

The only public-market pure-play on AI-bio software/infrastructure that isn't either a biotech-with-a-pipeline or a tools conglomerate. 23% TTM rev growth, $983M mkt cap, 3.85x PS. Stock fell 44% YoY on weak biotech end-market spend and pipeline timing — but the platform is the long-term asset. Their FEP+ AI hybrid is what Isomorphic and Xaira would use if they didn't roll their own. Best risk/reward in the software-of-biology bucket below $15.

What I would pay for: Conviction grows materially if they win a major Big Pharma platform deal or one of their equity-stake pipeline assets hits.

5. Lonza (LONN.SW) — Conviction: MEDIUM-HIGH

Best Western CDMO. Capacity coming online into structural shortage (mAbs + ADCs + cell therapy + AI-designed biologics + GLP-1 secondary fill). Books years out. Currently unprofitable on restructuring, which is exactly when quality CDMOs should be bought. Geopolitical kicker: every dollar that gets pulled from WuXi by US BIOSECURE-style action lands at Samsung or Lonza.

What I would pay for: Forward P/E 27 isn't dirt cheap but it's a quality compounder at a trough margin. Size patiently.

Honorable mentions / smaller positions

What I would explicitly avoid here

9. What would change my mind


10. Sources & data

All live price/market cap data: stockanalysis.com market close 2026-05-26.

Document generated 2026-05-26. All forward-looking quantitative ratios (supply/demand model, $/bp, $/genome at scale) are analyst estimates and should be treated as order-of-magnitude. Live financial data is from public market sources at market close.