Inside Basecamp Research: the AI startup converting evolution into training data

London-based AI biology startup Basecamp Research has secured $140 million in new funding, with backing from Nvidia and Anthropic's Anthology Fund, as it scales up models trained on genetic material harvested from rainforests, oceans, and hot springs. The company's EDEN models are already designing antibiotic candidates that have cleared early animal tests, and its next goal is ambitious: developing therapies that edit genes directly inside the human body. The raise underscores how biology, not language, may be the ultimate scaling challenge for AI — and why Basecamp argues the world's public genome databases fall dramatically short.

According to the company, investor S32 led the round. Nvidia, Anthropic's Anthology Fund, the NATO Innovation Fund, and Redalpine also participated, among others.

Where the money is going

Founded in 2020, Basecamp plans to use the capital to continue developing its EDEN biological AI models and to advance its own therapy candidates toward clinical development. Cell therapies come first — treatments that genetically modify a patient's cells so they can fight diseases such as cancer. Basecamp wants to perform that modification directly inside the body rather than in a lab.

To get there, the company turns to microorganisms from habitats scientists have barely studied, with the models learning patterns from their genomes that could have medical applications.

"Given the complexity of biology, we just need orders of magnitude more data," Basecamp CTO Philip Lorenz told THE DECODER. He stressed that pairing massive data collection with smaller, targeted experiments is what matters most.

There is still significant development work ahead. So far, Basecamp has lab results and mouse data — nothing that yet proves the approach can yield safe, effective therapies in humans.

Why public genome databases aren't enough

The Wall Street Journal recently reported that pharma companies are pouring billions into AI, yet solid evidence of meaningfully higher clinical success rates remains scarce. Lorenz doesn't see that as a knock on the technology. Scaling delivered rapid gains for language models; biology is progressing too, but its hardest open questions — like accelerating clinical trials — are far tougher. As a problem, he says, biology is "way, way, way more complicated and way larger."

The data landscape is also fundamentally different. Language models learn from vast text corpora, while pharma companies often work with small, siloed datasets from individual studies.

To illustrate the gap, Lorenz cites an Epoch AI estimate placing the upper limit of all existing words at roughly five quadrillion tokens. Basecamp estimates Earth's nucleotides at 10 to the power of 37. "If you took a stack of cards, like poker cards, of 10 to the power of 37 cards, that stack would surround the observable universe a million times," Lorenz says. You don't need a dataset that large, he adds, but you do need one far bigger than today's.

Public databases also skew heavily toward a handful of organisms. According to Basecamp's research paper on its BaseData database, about 68 percent of the sequence volume in the Sequence Read Archive comes from just five species — humans alone make up roughly 54 percent. That makes sense for medical research, but for a model meant to capture the breadth of biological processes, it's a bottleneck. "If you were to train an LLM only on newspaper articles from 1975, it would be a really, really bad model," Lorenz said in a Microsoft case study. "That is kind of where we are in biology."

So Basecamp collects its own samples with local research partners, drawing from rainforest soil, volcanic ground, and the deep sea off Antarctica. Its network now spans more than 30 countries and all seven continents; Microsoft counts 208 participating organizations across 31 countries.

Scaling toward a quadrillion tokens

At the time of the interview, Lorenz said the dataset held about 15 trillion tokens — already comparable to the text datasets behind models like Claude or GPT, except a token here is a single DNA building block. According to Microsoft, the first EDEN generation trained on 9.7 trillion DNA building blocks from more than a million newly sequenced species, with compute reportedly matching GPT-4's — a figure OpenAI has never officially disclosed.

Over the next year and a half, the dataset is expected to grow roughly a hundredfold, surpassing one quadrillion tokens. The framework is the Trillion Gene Atlas announced in March, in which Basecamp, Anthropic, Nvidia, PacBio, and Ultima Genomics plan to assemble genetic data on the scale of one trillion genes. Basecamp's latest funding announcement already calls the atlas the training foundation for EDEN, though it doesn't say how far the expansion has progressed.

What evolution teaches drug design

Lorenz sees evolution — the force behind all biological processes — as the bridge between environmental samples and medicine. Whether a hot spring bacterium adapts to heat or a cancer cell spreads, similar selective pressures apply. Human and vertebrate genomes are largely sequenced; the rest of biodiversity barely is. Many drugs, after all, originally came from plants, bacteria, and fungi.

Competing microorganisms offer a concrete example. Some of the most powerful therapeutic molecules emerge from "biological warfare" between organisms, Lorenz says. When one bacterial species tries to dominate an ecosystem, it may develop substances that inhibit or kill rivals — chemistry that can serve as antibiotics in medicine. In the Microsoft case study, he also names phages, viruses that infect bacteria and inject their DNA, whose evolutionary arms race with bacteria produced tools now used for gene editing.

Basecamp wants its models to go beyond rediscovering known compounds and instead design new candidates from learned patterns. The company records the chemical, physical, and ecological conditions at each collection site alongside the genetic material, and prioritizes long, continuous DNA stretches, which preserve information about which genes sit next to each other and may work together — context that short fragments lose.

An AI-designed antibiotic passes its first animal test

Lorenz points to EDEN's antibiotic designs as early evidence of medical value. "We prompt on a pathogen and then the model designs an antibiotic that kills it," he says.

In a not-yet-peer-reviewed paper on the EDEN model family, the authors report on a small, curated set of antimicrobial peptides — short protein chains that can attack bacteria. According to the paper, 97 percent of tested candidates showed activity in the lab. That rate applies only to the tested selection, not to anything the model might generate going forward.

Basecamp didn't stop at test tubes. A June announcement describes tests with a candidate called EDEN-7: in mice infected with multidrug-resistant bacteria, it reportedly performed about as well as a last-resort antibiotic. The company says the model generated the candidate directly, with no rounds of tweaking beforehand, working with researchers at the University of Pennsylvania led by Cesar de la Fuente.

Lorenz places heavy weight on the step from design to experiment: what matters is whether molecules actually work. Basecamp has also had EDEN generate a synthetic gut microbiome of about 9,000 bacterial species, per Microsoft. To verify the results, the team brought in UC Berkeley microbiome researcher Jill Banfield, who set strict criteria and co-authored the paper. "It is good to ask a skeptical human and not just AI," Lorenz said.

The therapy bet: inserting genes inside the body

For its own therapy work, Basecamp is deploying EDEN to design biological tools that insert larger DNA stretches at chosen spots in the human genome. These include large serine recombinases, enzymes that can rejoin DNA, engineered to insert therapeutically useful genetic information at precise locations.

The approach targets a core problem in gene therapy: many inherited diseases stem from different mutations depending on the patient. Rather than fixing each one, the enzymes would insert a healthy copy of the gene regardless of the specific mutation. These insertion tools originate from the same phage-bacteria conflict, and must be reprogrammed to function in the human genome.

In the EDEN paper, the authors report several working enzyme candidates for every disease-relevant target region studied, with half of the generated serine recombinases active in human cells. One possible application is CAR-T cells — genetically modified immune cells that recognize and attack cancer. In a January announcement, Basecamp said primary human T cells modified this way cleared more than 90 percent of tumor cells in the lab.

With the new funding, the company wants to develop this into a treatment that works directly inside the body, which would simplify cell therapy manufacturing that can currently cost hundreds of thousands of dollars per patient, according to the announcement.

But a genetic edit that works in the lab doesn't answer how components reach the right cells safely. Arc Institute co-founder Patrick Hsu raised this in a May 2025 Financial Times report, naming delivery, potential toxic effects, and cells' defense responses as open challenges.

"Our ambition is to make biology programmable. You prompt on a disease and out comes a molecule that will address that," Lorenz told Microsoft. He acknowledges the company is still far from that goal: "These molecules are great, they are active, they work. But for them to go into the clinic, there is a lot more that needs to happen."

Basecamp's published pipeline illustrates the long road. As of September 2026, none of its six programs has moved past lead optimization, the stage where promising candidates are refined. Preclinical studies come next, followed by clinical trials. Four programs sit in lead optimization, including in vivo CAR-T cell therapies for blood cancer and an unnamed autoimmune disease, and a liver gene therapy for the inherited metabolic disorder phenylketonuria (PKU).

Meta description: Basecamp Research raised $140 million from Nvidia, Anthropic's Anthology Fund, and others to scale EDEN, its biological AI models trained on DNA from nature.

Tags: Basecamp Research, EDEN AI, biological AI, drug discovery, gene therapy

Featured image: Abstract visualization of glowing DNA strands merging with neural network patterns against a dark, deep-ocean-inspired background.