Cradle Bio

Phase A — Understand the business

Lens 1 · Company Overview

Cradle is an AI-powered protein-engineering software company. It sells biologists a SaaS platform that uses generative machine-learning models — "techniques derived from large language models to generate amino-acid sequence variants predicted to exhibit user-selected properties (such as stability or binding)" — so that a scientist who has a working protein but wants it more thermostable, higher-binding, less immunogenic, or more expressible can get a ranked list of candidate sequences to test, instead of grinding through trial-and-error rounds in the lab. The pitch is a 1.2×–12× speed-up in early R&D and up to 90% cost reduction across therapeutics, diagnostics, agriculture, chemicals and food.

Founded May 2021 (incorporated weeks before Index Ventures first met the team in Nov 2021). HQ Amsterdam, second office Zurich, with a US presence built out in 2025. Five co-founders: Stef van Grieken (CEO), Jelle Prins, Elise de Reus, Eli Bixby, Harmen van Rossum — a deliberate "ML-from-big-tech + biology" pairing (see Lens 9).

How it makes money — and the strategic self-limitation that defines the whole company. Cradle is pure SaaS — a software subscription, explicitly not a co-development or drug-discovery shop. Van Grieken is unambiguous: competitors "split into those doing close partnerships to co-develop a drug or process, and those, like Cradle, strictly providing a software service". The model deliberately "avoids royalties, revenue share, or IP issues" — the customer keeps 100% of the molecule, the IP, and the upside; Cradle keeps a software fee. The thesis behind it: "We believe that AI in drug discovery and development will ultimately be a commodity and any team should have access to it". This is the single most important fact in the dossier and it cuts both ways — it is why pharma will let Cradle near its most sensitive programs (no IP entanglement, data stays private), and it is why Cradle captures none of the multi-billion-dollar value of the drugs it helps create. It is the Snowflake-of-proteins bet, not the Generate-Biomedicines bet.

Contract structure / payment terms: not publicly disclosed — n/a on price points, seat counts, or contract length. The model is enterprise SaaS (annual subscription, land-and-expand into pharma R&D teams); the hiring of a CCO from Veeva/Benchling/Medidata (Lens 9) signals an enterprise-license motion, not transactional/PLG.

Lens 2 · Supply Chain

A software company's "supply chain" is its compute, its data, and its talent — plus, uniquely for Cradle, a wet lab. Mapping upstream → Cradle → end customer with named stakeholders:

Upstream inputs:

• Cloud compute / training infra — Google Cloud. Cradle is a public Google Cloud customer; the platform trains and serves on GCP. This is both an input dependency and a founder-network artifact (van Grieken is ex-Google).
• Foundational training data — public protein corpora + proprietary wet-lab data. Models are "trained on billions of protein sequences as well as data generated in their own wet laboratory". The public side (UniProt-scale sequence data, and increasingly open structure predictions from AlphaFold/ESM) is available to every competitor — the moat is the proprietary layer.
• The Amsterdam wet lab — the differentiator. Cradle runs its own lab to "A/B test on many different types of proteins as well as develop 'Foundational Datasets'" and to "regularly train and fine-tune models from internal datasets". Two of the three Series B funding priorities feed this: expanding the wet lab for additional protein modalities, and growing the ML/eng team. Lab cost/size: n/a.
• Customer experimental data (the flywheel input). When a customer runs Cradle's suggested sequences in their lab and reports back which worked, that closed loop refines the customer-specific model. Critically, Cradle's design keeps "proprietary data completely private" per customer — so the flywheel is per-tenant, not a shared global model. (This is a moat nuance and a moat weakness — see Lens 3/13.)

Cradle (the company): generative + property-prediction models + the platform UI + the wet-lab data factory + an enterprise GTM org (~40+ → doubled in 2025).

Downstream / end customers (named): Johnson & Johnson (Innovative Medicine), AbbVie, Novo Nordisk, Grifols, Novonesis (the merged Novozymes + Chr. Hansen), Twist Bioscience. End markets: biologic drugs (antibodies, peptides incl. next-gen GLP-1), industrial enzymes (Novonesis), diagnostics, agriculture, food, bio-based materials.

Chokepoints / single-source dependencies: (1) Google Cloud as compute landlord — a margin and dependency lever. (2) The wet lab is single-site (Amsterdam) — a concentration risk for the proprietary-data engine. (3) Talent — the entire moat is a few dozen rare ML-×-protein engineers; in the most competitive hiring market in tech-bio. Names along the chain or it didn't happen — done: GCP, J&J, AbbVie, Novo Nordisk, Grifols, Novonesis, Twist.

Lens 3 · Competitive Advantages (moats)

The honest verdict: Cradle has a real but shallow-and-narrowing moat, and it has voluntarily forfeited the deepest one available.

What is genuinely defensible:

• Proprietary wet-lab "foundational datasets." The public models (AlphaFold3, ESM/ESMFold2, Boltz, OpenFold3) are converging and increasingly free (Lens 13). The differentiated input is labeled functional data on the specific properties customers care about (stability, expression, developability) — which public structure predictors do not give you. Cradle's lab generates this. This is the strongest leg.
• Product / workflow / "copilot for the bench." Cradle is built as everyday software for a working scientist, not a research artifact. The 8+/10 satisfaction score and the land into 6 of the top-25 pharma suggest real product-market fit on usability, which in enterprise SaaS is a durable switching-cost source once embedded in R&D workflow.
• The privacy / no-IP-entanglement wedge. Because Cradle takes no royalty and no IP, and walls each customer's data, pharma will plug it into sensitive programs (e.g. Novo's GLP-1 work) that they would never share with a co-development rival. This is a genuine commercial advantage — trust as a moat.
• Switching costs (forming). Per-tenant fine-tuned models mean a customer's accumulated experimental data is locked to Cradle's platform; ripping it out resets the flywheel. Real, but early.

Bargaining power — who needs whom more: weak for Cradle today. Its customers are the largest pharma and enzyme houses on earth (J&J, AbbVie, Novo, Novonesis); Cradle is a sub-$100M-revenue (almost certainly far sub) vendor. Those customers can build in-house (many have AI groups), back open-source (Novo is literally co-funding OpenFold3 — Lens 13), or multi-source across Cradle/Profluent/Generate. Cradle's leverage is switching cost + the privacy wedge, not indispensability.

The forfeited moat: by refusing royalties/IP, Cradle gave up equity in outcomes — the thing that turned Recursion, Generate, and the co-development players into multi-billion-dollar stories. Cradle's ceiling is "great vertical SaaS business," not "owns a piece of the next Wegovy." Whether that is discipline or under-ambition is the central debate (Lens 12).

Lens 4 · Segments

No segment-level financials are disclosed — Cradle is private and reports nothing. segments.csv is empty; every figure here is qualitative ``. There is no revenue/EBITDA/earnings split to source — n/a for all segment numbers.

What can be characterized (operational mix, not financial):

• By end-market vertical: therapeutics (the largest and fastest — antibodies, peptides, next-gen GLP-1), diagnostics, agriculture, chemicals/industrial enzymes (Novonesis), and food. Therapeutics is clearly the center of gravity — "six of the top-25 big pharma" and the Novo GLP-1 work are the headline logos.
• By geography: Europe-anchored (Amsterdam/Zurich) with a 2025 US build-out described as the growth priority — the US is where the pharma R&D budgets and the protein-AI talent war are, so the US ramp is the segment trend that matters.
• Trend: 50+ projects and 31 molecules in development as of late 2024–2025, up from "9 partners onboarded in the past year" at the Series A in Nov 2023 — i.e. roughly 21+ customers by the Series B. Direction is clearly accelerating on logos and projects; the financial slope is unobservable.

---

Phase B — Measure performance

+private overlay: Phase B swaps the SEC-grounded earnings lenses for funding/valuation trajectory, traction/unit-economics signals, and cap-table quality. All ``, unaudited.

Lens 5 → Funding & valuation trajectory (replaces Earnings Result)

The round history, seed → latest:

• Total raised: ~$103M / "over $100M".
• Valuation: never publicly disclosed at any round — n/a. (A $73M Series B led by a tier-1 crossover-adjacent firm like IVP in late-2024 techbio would typically imply a low-to-mid-hundreds-of-millions post, but I will not fabricate a multiple; treat as unknown.)
• Cadence: a clean ~12-month seed→A→B rhythm with the same lead (Index) doubling down twice before IVP led the B — a strong insider-conviction signal.
• Burn signal: the explicit use-of-funds is build-the-wet-lab + hire ML/eng + scale GTM — i.e. classic growth-stage cash consumption, not a path-to-profit narrative. No runway/default-alive disclosure — n/a. With ~$103M raised and headcount doubling through 2025 off a ~40-person base, this is a company spending into growth, which is appropriate at Series B but means the next raise (a Series C) is the real test.

Lens 6 → Founder & ecosystem signal (replaces Earnings Calls sentiment)

No earnings calls exist. The analog — founder interviews/podcasts and ecosystem tone:

• Van Grieken is a frequent, articulate public communicator (Swisspreneur ep530, BiopharmaTrend, Venture Europe, Index "catching up", Founders Forum "Rising Stars") and consistently hammers one disciplined message: AI-for-bio will commoditize; Cradle's job is to be the universal software layer, not to own drugs. The message has not wobbled across two years — a positive consistency signal.
• Ecosystem validation is strong and improving: Wired "Hottest Startups in Amsterdam 2024," Bloomberg "25 European Startups to Watch 2025," MT/Sprout "Startup of the Year 2024". The quality of the angel/investor base (Recursion's Gibson, Index×3, IVP) is the loudest endorsement.
• Tone shift to watch: the Dec-2025 "six of top-25 pharma / doubled headcount / US expansion" press release reads as a commercial-scaling narrative, a notch more enterprise-GTM-flavored than the 2023–24 "frontier research" tone — consistent with the CCO hire. Healthy maturation, not a red flag.

Lens 7 → Cap table & peer / mechanism comps (replaces Comps)

Cap-table quality: Index Ventures (seed + A lead, B participant — triple commitment), IVP (Series B lead — a growth-stage, IPO-track investor; Uber/Slack/Robinhood pedigree per Sifted), Kindred Capital (all three rounds), plus operator-angels Chris Gibson and Tom Glocer. No disclosed crossover-fund entry yet (no Fidelity / T. Rowe / Coatue mark) — which is itself the tell: Cradle is not yet IPO-proximate; IVP at Series B is one step early of the crossover signal. Secondary marks: n/a — not disclosed.

Peer comparison — by competitive model, not by P/E (all private; multiples mostly n/a):

The comp table makes the strategic fork vivid: Generate raised ~7× Cradle's capital because it sells the drug (and the upside), not the software. Cradle's capital efficiency is the flip side of its capped ceiling. Multiples (EV/Sales etc.): n/a for every private name — none discloses revenue. Do not infer one.

Lens 8 → Funding & product catalysts (replaces Stock-Price Catalysts)

No stock, so the "what moves the name" history is funding/product/logo events:

• Nov 2022 — €5.5M seed (Index) → company exists, frontier-credible.
• Nov 2023 — $24M Series A + the J&J / Novozymes / Twist partnerships disclosed; Recursion's Gibson angels in → commercial validation.
• Nov 2024 — $73M Series B (IVP) + CCO hire (Sam Partovi) → scale-up inflection; this is the clearest value step-up to date.
• 2025 — headcount doubled, US build-out, "6 of top-25 pharma," Novo Nordisk next-gen-GLP-1 work surfaced → land-and-expand into the most valuable drug category on earth.
• Pattern: Cradle's "stock" reacts to logo quality and round size, exactly as a private enterprise-SaaS name should. The next real catalyst is a Series C and/or a marquee renewal-and-expansion proving net revenue retention — or, bearishly, a down/flat round if the protein-AI hype cools (Lens 13).

---

Phase C — Judge people & books

Lens 9 · Management

Stef van Grieken (Co-founder & CEO). The single biggest asset. Seven years at Google, including Group Product Manager / leadership in Google Research & Machine Intelligence (Google Brain) overseeing infra/developer/applied-AI portfolios, plus a Google X moonshot co-lead and earlier work on Google Maps / Android Auto. Earlier in his career he was associated with Wave (the Google Wave/communications lineage often cited in his bio). This is a genuinely rare profile: a product-and-applied-ML operator from the single most important AI lab of the 2010s, not an academic. Skin in the game: founder-CEO since 2021, four years in, presumably a large equity holder (exact % n/a — not disclosed for a private co.).

Co-founders: Jelle Prins (a well-known Dutch product designer — early Uber design pedigree is frequently cited in NL tech circles; treat the Uber link as ``), Elise de Reus, Eli Bixby (ML/eng, ex-Google-adjacent), Harmen van Rossum. The founding logic — big-tech ML talent fused with biologists — is exactly the combination this category rewards and is hard to assemble.

Senior hire signal — Sam Partovi (CCO, the "40th hire," 2024). Two decades scaling life-sciences cloud platforms at Veeva, Benchling, and Medidata. This is a category-defining enterprise-SaaS GTM résumé — you hire this person when you intend to run a Veeva-style enterprise land-and-expand into pharma, and it materially de-risks the "can a research-led founder build a sales machine?" question.

Capital-allocation history: too early / private to judge on ROE/ROIC. What is visible is disciplined capital efficiency — ~$103M raised to reach 6-of-top-25-pharma is lean versus Generate's ~$693M. The deliberate no-royalty model is a capital-allocation philosophy (own software margins, not drug lottery tickets).

Red flags: none material found. No related-party deals, no promotional/over-claiming behavior surfaced, no governance smells in public sources. The only "soft" flag is strategic, not ethical — the no-IP model may be leaving enormous value on the table (the bear/skeptic's core charge).

Archetype: founder-led, mission-driven, technically-credible — the right archetype for a frontier-but-commercial tools company at this stage. Van Grieken is a builder-communicator, not a hype-merchant.

Lens 10 · Forensic Red Flags

Cradle is private and unaudited — there is no income statement, balance sheet, or cash-flow statement to forensically examine. The classic checklist (revenue recognition, receivables vs. revenue, SBC flattering non-GAAP, goodwill/intangibles, lease/related-party games) is n/a — no audited financials disclosed. The honest forensic posture toward a name like this is: you are trusting management's selective disclosures (logo counts, speed-up multiples, "6 of top-25") with no auditor and no GAAP behind them. Specifically un-verifiable and worth flagging:

• The "1.2×–12× speed-up" and "up to 90% cost reduction" are self-reported marketing ranges — the 12× is a best-case cherry, the 1.2× is the floor; the typical result is undisclosed. Treat as directional, not audited.
• "31 molecules in development" / "50+ projects" — Cradle's economics are unconnected to whether those molecules succeed (no royalty), so these are usage-vanity metrics, not value-creation metrics.
• Revenue / ARR / NRR / gross margin / burn / runway: all n/a. The most important number for a SaaS business — net revenue retention — is completely dark.

Regulatory findings (required sub-section).

• SEC (EDGAR EFTS — LR + AAER): none possible. Per regulatory/regulatory-findings.md (generated 2026-06-24), Cradle Bio has no CIK — it is private and not required to file with the SEC; total SEC findings = 0.
• Non-SEC enforcement (FTC / DOJ / FDA / CFPB / EU) — web search run: "Cradle Bio" (FTC OR DOJ OR FDA OR CFPB OR "consent decree" OR settlement OR fine OR penalty) enforcement and a broader lawsuit/controversy/layoffs search returned no material hits. As a B2B software vendor that does not itself develop or market a drug, Cradle has minimal direct FDA/clinical-liability surface.
• Item 3 Legal Proceedings: no 10-K exists (private) — n/a.
• Conclusion: No material regulatory or legal findings — verified via SEC EDGAR EFTS (LR + AAER, zero records, no CIK), web search across FTC/DOJ/FDA/CFPB + general controversy/litigation, as of 2026-06-24. Unaudited per public sources. (The real risk here is not enforcement — it is the absence of any audited disclosure at all.)

---

Phase D — Project & stress-test

Lens 11 → IPO-readiness & path-to-tradeable (replaces Forward Projection)

There is no research/private-watch.json entry for cradle-bio to anchor stage/ipo_readiness/catalyst — so I assess from first principles:

• Stage: growth / early-late — Series B done Nov-2024 (~$103M total), scaling GTM, no crossover fund on the cap table yet.
• IPO-readiness: ~2 / 5 (on the 1=seed … 5=S-1-imminent scale). Rationale: a software business can IPO far smaller than a clinical biotech, but Cradle is almost certainly still sub-scale on ARR for a public software listing, has no disclosed crossover round, and the natural next step is a Series C, not an S-1. It is not IPO-proximate in 2026.
• Milestones that would unlock an S-1 (the be-early watch-list): (1) a Series C led by or including a crossover fund (Fidelity/T. Rowe/Coatue/Altimeter) — the single clearest IPO-proximity tell; (2) ARR crossing ~$50–100M with disclosed net revenue retention >120%; (3) the US enterprise motion proving it can sell American pharma at scale; (4) a category-defining renewal/expansion at a J&J/AbbVie/Novo that converts a pilot into a multi-year seven-figure deal.
• More-likely exit than IPO: strategic acquisition. The natural acquirers are the platform/data and life-science-software majors — Schrödinger, Benchling/Recursion, Danaher/IDT, Thermo Fisher, ServiceNow-for-science aspirants, or a hyperscaler (Google given the existing GCP + alumni ties; Microsoft; AWS) — or a large pharma buying the capability in-house (the Novo relationship is double-edged here). For a pure-software, no-IP, deeply-embedded R&D tool, M&A is the base-rate outcome.
• Investability TODAY: not directly investable — private, no secondary access surfaced, no priced valuation. This is a watch / IPO-pipeline name, banked for a Series C crossover signal or an S-1.

(No forecast.ts create logged — there is no EPS line or binary clinical readout that Cradle owns to score; per skill rules, the --watchlist/unattended loop skips the forecast create step, and a private no-pipeline tools vendor has no scoreable base-case number. Correct to skip.)

Recommended private-watch.json seed (for Connor to add outside this loop — NOT written by this run per wave boundaries):

"cradle-bio": { "beat": "ai-bio", "stage": "growth", "ipo_readiness": 2,
 "lead_investors": "IVP, Index Ventures, Kindred Capital",
 "catalyst": "Series C / crossover entry; ARR + NRR disclosure; US pharma land-and-expand; possible strategic acquisition (Schrödinger/Benchling/Google)",
 "dossier": "companies/cradle-bio/deep-dive-2026-06-24.md" }

Lens 12 · Bull vs Bear

Bull case. Cradle is building the system-of-record for protein engineering — the Benchling/Veeva of biologic design — at the exact moment every large pharma is institutionalizing AI in R&D. Its wedge is trust: no IP, no royalty, walled per-tenant data, so it gets invited into the crown-jewel programs (Novo's next-gen GLP-1) that no co-development rival can touch. The proprietary wet-lab functional-data flywheel is the one thing open-source structure models don't give you, and per-tenant fine-tuned models create real switching costs. Capital-efficient (6-of-top-25-pharma on ~$103M vs Generate's ~$693M), founder is a genuine ex-Google-Brain product-ML operator, and the CCO hire signals a credible enterprise sales machine. If AI-for-bio becomes a commodity utility — as the CEO predicts — the universal, neutral software layer is a fabulous, high-margin, durable SaaS business and a clean acquisition target at a premium.

Bear case (2–3 permanent-impairment risks).

1. Commoditization from below — and your best customer is funding it. The model layer is collapsing toward free: ESMFold2 (CZI/Biohub, 1.1B structures, no commercial restrictions), AlphaFold3, Boltz-2, and OpenFold3 — which Novo Nordisk is co-funding with AWS/Columbia. When the foundational capability is open and excellent, Cradle's value compresses to UI + workflow + its (narrow) proprietary functional data — defensible, but a thin slice, and large pharma can replicate the wet-lab-data loop internally.
2. The no-IP ceiling. By design Cradle captures none of the drug upside. Its TAM is "pharma's protein-design software budget," not "a share of the next Wegovy." That caps the equity story at vertical-SaaS multiples and makes it structurally smaller than the co-development players — exactly why Generate out-raised it 7×.
3. Customer-concentration + build-vs-buy. A handful of mega-pharma logos drive the story; each has an in-house AI group and the cash to build, multi-source (Profluent/Generate/Isomorphic), or acquire. Lose one or two anchor accounts to insourcing and the growth narrative cracks.

Pre-mortem (18 months out, thesis broke): Open models reached "good enough" for most enterprise property-optimization; two anchor pharma customers pulled protein-AI in-house after building their own functional-data loops; a flat/down Series C signaled the protein-AI hype cycle had cooled; and Cradle's no-royalty model meant none of the molecules it helped design threw off any value to cushion the slowdown. It survives as a solid but unspectacular SaaS vendor, acquired by a platform major at a software multiple — not the category-definer the Series B priced.

Are multiples too high? Unobservable (private, undisclosed). But the category (protein-AI) is hype-rich; a private mark set in late-2024 enthusiasm carries down-round risk into a 2026–27 Series C if sentiment normalizes.

Contrarian view (what the market refuses to see): the crowd frames "no IP, pure SaaS" as Cradle's weakness (no drug upside). The contrarian read is it is the only durable position in a field about to be commoditized — when the models are free, the neutral, trusted, embedded software layer with the private functional-data loop is the one part that compounds, precisely because it doesn't compete with its customers for the molecule. Cradle may be right that owning the drug is the bad business and owning the workflow is the good one. The risk is that "the workflow" turns out to be a thin enough slice that pharma just builds it.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case:

• The product is a wrapper on a commoditizing frontier. Strip the marketing and Cradle = (open-ish models) + (a modest proprietary wet-lab dataset) + (a nice UI) + (enterprise sales). Two of those four are racing toward free. The most dangerous competitor bulls underestimate is not Profluent or Generate — it is open-source itself, weaponized by the customers. Novo Nordisk co-funds OpenFold3; CZI ships ESMFold2 free with 1.1B structures and no commercial restrictions. Your TAM is being given away by the people you sell to.
• Revenue concentration + no lock on outcomes. Concentration sits in a few mega-pharma logos, each of which can insource. And because Cradle takes no royalty, it has zero participation if those programs succeed — so its "31 molecules / 50+ projects" are usage metrics with no economic tail. The business is a subscription that can be cancelled, not an annuity on drug success.
• The moat may be weaker than bulls think. Per-tenant private models prevent Cradle from building one giant compounding global model — the very thing that would create an unassailable data moat. The privacy wedge that wins the deal also caps the moat. Meanwhile a large pharma can stand up its own internal functional-data loop with its own (vastly larger) experimental throughput.
• Worst capital-allocation / incentive critique: none egregious — but the strategic choice to forgo IP/royalty is, from a short-seller's chair, leaving the only real value-accrual mechanism in this industry on the table. "Capital efficient" can also read as "structurally unable to capture the upside it creates."
• What must hold for the (private) price: that protein-AI software becomes a large, durable, must-have enterprise budget line owned by a neutral vendor rather than (a) absorbed into pharma's in-house AI orgs, or (b) commoditized to near-zero by open models + cloud-native tooling. If enterprise growth disappoints by 20–30%, a 2026–27 Series C is a down round, the IPO path closes, and the outcome compresses to a sub-scale strategic acquisition.
• The single permanent-impairment scenario (and plausibility): open models reach "good enough" and the two or three anchor pharma customers internalize the workflow → Cradle's differentiation collapses to a thin UI layer and it becomes a feature, not a company. Plausibility: moderate — the privacy wedge and switching costs are a real buffer, and "good enough" in regulated biologic design is a higher bar than in consumer AI, but the direction of travel (free models, in-house pharma AI) is unambiguously against Cradle's most expansive valuation.

Lens 14 · Management Questions (ordered by information value)

1. What is your net revenue retention, and how much of ARR is expansion vs. new logos? (The one number that tells you if this is a compounding SaaS business or a pilot-churn treadmill — and it is completely undisclosed.)
2. As open models (ESMFold2, AlphaFold3, OpenFold3 — which Novo co-funds) reach parity on the capabilities you sell, what specifically remains defensible, and how do you quantify the value of your proprietary wet-lab data over the free baseline?
3. You took the no-royalty/no-IP path on principle. At what scale, if ever, would you take economic participation in customer outcomes — and if never, what is the realistic revenue ceiling of "pharma's protein-design software budget"?
4. How concentrated is revenue in your top 3–5 customers, and what is your defense when a J&J or Novo decides to build this in-house?
5. Per-tenant private models prevent a single compounding global model. Does that permanently cap your data moat, and how do you think about that trade-off versus a federated/shared-learning architecture?
6. What does the path to a Series C look like — and would you expect a crossover fund (Fidelity/T. Rowe/Coatue/Altimeter) to lead it, at a step-up from the 2024 mark?
7. Is the business default-alive on current ARR + the Series B cash, or does growth require the next raise? What is current runway?
8. What is gross margin, given the Google Cloud compute + wet-lab opex load — and how does it scale as model-serving costs fall?
9. What is the realistic acquisition landscape — and have you had inbound from platform majors (Schrödinger, Benchling, Thermo, Danaher) or hyperscalers?
10. The wet lab is single-site (Amsterdam). How critical is it to the moat, and what is the concentration/continuity risk?
11. Where have customers churned or not renewed, and what did you learn about the limits of the value proposition?
12. How do you win the talent war for ML-×-protein engineers against Isomorphic (Alphabet money), the big pharma AI orgs, and Profluent/Generate?
13. What is the actual distribution of the "1.2×–12×" speed-up across real customer projects — and what is the median, not the best case?
14. How do you keep pharma comfortable that their walled data is truly never used to improve other tenants' models — and is that auditable?
15. If AI-for-bio commoditizes exactly as you predict, why is the neutral software layer a venture-scale outcome rather than a feature that the cloud/ELN incumbents (Benchling, Veeva, GCP) absorb?

---