Basecamp Research raised $140M to turn evolution into AI training data

Basecamp Research has raised $140 million from investors including Nvidia and Anthropic's Anthology Fund. The London-based company uses genetic material harvested from rainforests, oceans, and hot springs as training data for AI models, and applies those models to design antibiotics and tools for cell therapies.

The funding round signals where serious AI money is moving: beyond language, toward biology. In an interview with THE DECODER, Basecamp Research CTO Philip Lorenz laid out the core thesis behind the company — and the reasons biological AI lags far behind its linguistic cousin.

Why is biology a harder problem for AI than language?

Lorenz's argument, as reported in the interview, is that biology presents a far bigger problem for AI than language does. Where text corpora offer oceans of human-generated data, the genetic record of life on Earth — billions of years of evolutionary experiments encoded in DNA — remains largely untapped as a machine-learning resource.

That is the gap Basecamp Research targets. The company collects genetic material from extreme and biodiverse environments: rainforests, oceans, and hot springs. Evolution, in this framing, is itself a dataset — one that encodes solutions to molecular problems that no laboratory would stumble upon by brute force.

The commercial stakes are concrete. Models trained on this data are being pointed at two of the most expensive problems in medicine: designing new antibiotics and building tools for cell therapies. Antibiotic resistance and the high cost of engineered cell treatments are established pressures on drug development pipelines, which is why a genetic-data moat could prove valuable to pharmaceutical partners.

Who is backing the company?

The $140 million round includes two names that carry weight in AI circles:

• Nvidia — the chipmaker whose hardware underpins most frontier model training
• Anthropic's Anthology Fund — the investment vehicle tied to the Claude maker

The participation of both an infrastructure giant and a frontier AI lab suggests the bet is not merely on biotech outcomes. It is also on the thesis that nature's genetic diversity is a fundamentally different kind of training corpus — one that could matter as AI looks beyond text.

Why don't good benchmark scores guarantee good molecules?

Lorenz raised a caution that applies well beyond his own company: good scores on paper don't guarantee good molecules. Benchmark performance in computational biology, in other words, can diverge sharply from whether a designed molecule actually works.

The point matters for how the field measures progress. If evaluation methods for biological models are unreliable proxies for real-world function, then headline benchmark results — the currency of the AI industry — carry less weight in drug design than in, say, coding or question answering. Closing that gap between paper metrics and functional molecules is part of what Basecamp Research says it is building toward.

What happens next?

With $140 million in hand and backing from Nvidia and Anthropic's fund, Basecamp Research is positioned to expand its genetic data collection and its model development for antibiotics and cell therapy tools. Whether the bet pays off will depend on the test Lorenz himself sets out: not leaderboard numbers, but molecules that work.