Why J&J is paying a Dutch startup to design proteins with AI

For decades, creating new proteins has been biology's version of throwing darts blindfolded. Scientists modify a gene, grow it in cells, test it in the lab, and hope it works. Most attempts fail, and those that succeed take years and cost millions of dollars.

Here's the problem: biology is incredibly complex. A single protein might have 300 amino acids arranged in trillions of possible combinations. Traditional methods rely on intuition, incremental changes, and endless rounds of trial and error. The average cost to bring a new drug to market now exceeds $2.6 billion, and most of that money is spent on experiments that go nowhere.

Cradle Bio believes they've found a way out.

Cradle's core insight is simple: if biology is an information system, then AI should be able to learn its grammar.

DNA is code. Proteins are the output: molecules that do actual work in your body (bind to cancer cells, break down toxins, regulate metabolism). Change the DNA sequence, and you change what the protein does. The question is: which changes make it better?

Large Language Models like ChatGPT learned to write by reading billions of sentences. Cradle does the same with proteins. Their AI trains on billions of protein sequences to learn patterns: what makes a protein stable, what makes it bind tighter, and what makes it break down at high temperatures.

Once it knows the rules, it can suggest better designs.

Here's how it works in practice: A scientist uploads their protein data to Cradle's platform. The AI generates 50-100 candidate sequences optimized for specific properties (longer-lasting, more stable, better binding). The scientist tests them in the lab. Results go back into Cradle. The AI learns what worked and what didn't. Next round, predictions get better.

This feedback loop is the key. Most AI models are static: train them once, done. Cradle's models get smarter every time you use them. It's like having a lab partner that remembers every experiment you've ever run.

The results: 70% of Cradle's AI-designed proteins work in the lab, compared to under 5% with traditional methods. That's not incremental improvement. It's a 14x jump in success rate.

Cradle was founded in 2021 by Stef van Grieken (ex-Google) and a team of machine learning engineers and veteran bioengineers. Dual headquarters in Amsterdam and Zurich. They exited stealth in 2022 with a radical premise: make protein design as easy as writing code.

The business model is deliberately simple. No royalties. No IP disputes. No revenue-sharing deals. Just SaaS. You pay a subscription fee, you use the platform, you keep everything you build.

This matters because most biotech partnerships are messy. Drug companies hate royalty deals because they eat into margins and create ongoing negotiations. Cradle's model is clean: pay for the tool, keep the output. It's Figma for proteins.

The traction speaks for itself:

Use cases span industries. Johnson & Johnson is designing antibodies. Novozymes is engineering industrial enzymes. Novo Nordisk (the company behind Ozempic) is developing proteins for the treatment of metabolic diseases. Ginkgo Bioworks, the synthetic biology giant, partnered with Cradle in a large-scale deal announced in 2024.

Cradle also runs its own wet lab in Amsterdam. This isn't just software. They validate predictions through real experiments, generating proprietary training data that improves their models' performance out of the box. Think of it as Tesla running test tracks to train self-driving AI.

The company has grown to 40 people across ML engineering, biotech research, and commercial teams. They're hiring aggressively (especially in sales and operations) to scale from 21 customers to their stated goal of "a million scientists."

Cradle raised over $100 million in funding since 2022:

IVP is notable here. They backed Netflix, Slack, Twitter, and Coinbase at inflection points. Their bet on Cradle signals a belief that AI protein design becomes foundational infrastructure, not a niche tool, but something every biotech eventually needs.

Alex Lim, General Partner at IVP, put it plainly: "Biology is one of the domains where generative AI can have the biggest positive impact. Cradle is leading the way with its pioneering approach to protein design as a digital service."

The investor thesis is simple: if Cradle's 5-10x speedup holds across industries, it compresses decades of R&D timelines. Drug development that used to take 5 years could take 18 months. Industrial enzymes that took $20M to optimize could cost $2M. That's not incremental. It's structural.

Cradle is positioning itself as a platform play, not a drug company. They don't own the molecules their customers create. They own the infrastructure that those customers use to make them. The comparison to AWS isn't accidental. They want to be the cloud provider for protein engineering.

The funding will expand three areas:

Cradle's Chief Commercial Officer, Sam Partovi, joined in 2024 from the life sciences software world—his mandate: scale from 21 customers to thousands.

Biotech has a fundamental speed problem. Clinical trials take 10+ years. Most drugs fail. Even the ones that succeed cost billions to develop. The traditional model (guess, test, fail, repeat) is too slow for the problems we need to solve.

AI can't fix biology's complexity. Cells are messy. Proteins fold in unpredictable ways. No model will perfectly predict how a drug behaves in humans.

But AI can eliminate bad experiments before they happen. If you can predict with 70% accuracy which proteins will work, you skip the 95% that won't. You compress years of wasted lab time into months of focused iteration.

This has downstream effects across industries:

Pharma: Faster antibody design means cheaper cancer drugs, faster vaccine development, and better treatments for rare diseases.
Industrial biotech: Enzymes that work at higher temperatures or lower pH levels mean cleaner manufacturing processes and less chemical waste.
Food and agriculture: Proteins that replace animal-derived ingredients (like egg whites or dairy) get cheaper and scale faster.
Materials: Engineered silk proteins, bio-based plastics, and sustainable textiles all depend on the optimization of natural proteins.

If Cradle scales to "a million scientists" (their stated goal), protein design becomes democratized. It's not just Pfizer and Genentech. It's scrappy biotech startups, academic labs, food companies, and materials researchers. The barrier to entry drops from "you need a $50M lab and a decade of expertise" to "you need a laptop and a subscription."

That's the bet. Not that Cradle will discover the next blockbuster drug. But they'll build the tools everyone else uses to find theirs.

Cradle is still early. They're 3 years old, 40 people, 21 customers. The molecules being designed on their platform haven't reached the market yet. The real test comes when those drugs enter clinical trials, when those enzymes are deployed at an industrial scale, and when those materials are manufactured at volume.

But the early signals are strong. The science works. The customers are real. The funding is there. And the problem they're solving (how to engineer biology with the speed and predictability of software) is one of the defining challenges of the next decade.

If they're right, biology stops being an artisanal craft and becomes an engineering discipline. And that changes everything.

If this was useful, forward it to someone building in biotech. More stories like this every week at Thinking Folds.