An NVIDIA-Backed Lab Pushed Antibiotic AI Into Claude — And A Rival Copied It Same Day
NVIDIA's Cash and Compute Land at Basecamp
NVIDIA's venture arm backed Basecamp Research in a pre-Series C round, handing the London-area biology lab both cash and the compute cluster to train its EDEN foundation models, which Basecamp later pushed into Anthropic's Claude Science, as rival NVIDIA-backed biotech AI labs launch their own foundation model platforms in a biological AI infrastructure land-grab.
NVentures, NVIDIA's in-house fund, closed the deal in January 2026 after years of technical collaboration between the two companies (press release). Basecamp had already raised about $85 million from investors including Singular, True Ventures, Hummingbird Ventures, and former Unilever CEO Paul Polman. Huang named Basecamp in a £2 billion UK AI industry pledge last year, and NVIDIA's commitment rode that broader promise. The pre-Series C lands before the startup's next priced round, giving NVIDIA a seat as the models move from research to pharma partnerships.
Basecamp calls the lead system aiPGI. It places large therapeutic DNA segments at precise genomic locations, a job CRISPR can't do without snapping DNA. The training devoured compute: the largest EDEN model ran 1.95 septillion calculations across 1,008 NVIDIA Hopper GPUs, accelerated by BioNeMo libraries, matching GPT-4-class scale. It fed on more than ten trillion evolutionary DNA tokens drawn from over one million newly discovered species. Gowers said the models cost tens of millions of pounds and his team built them directly with NVIDIA's GPUs.
NVIDIA tied compute to this lab for its data moat. Gowers argued in 2024 that top pharma trains biology models blind to most of nature. The team's investor blog says only 1% of nature's proteins are known. Basecamp collects primary samples from volcanoes and hot springs through partnerships with more than 152 organizations in 28 countries. That dataset — not just the GPUs — is what NVIDIA's investment accelerates.
From model to therapy
EDEN is no chatbot. Basecamp designs therapeutics with pharma partners using the models. In lab validation, the models built active insertion proteins for 100% of tested disease sites, and across more than 10,000 genomic locations they cleared over nine in ten tumor cells in CAR-T assays and confirmed activity against nearly all drug-resistant pathogens.
"What we are showing in this paper is the ability to do large edits, large insertions, which opens the door to tens of thousands of genetic diseases that today wouldn't be treatable by CRISPR." — Glen Gowers, tech.eu 2026-01-12
Gowers said the system can take a pathogen name and output a killing peptide with near-perfect lab accuracy. The paper, co-authored by NVIDIA, Microsoft, and academic partners, debuted at NVIDIA's JP Morgan Healthcare Conference presentation. NVIDIA's BioNeMo supplied the libraries; the chipmaker dug into the research, not a passive check.
Basecamp's founders — biology PhDs Glen Gowers and Oliver Vince — met at Oxford and once sequenced DNA on an Icelandic ice cap. NVentures' pre-Series C cements their stance as a frontier AI lab for evolution. Basecamp now holds NVIDIA compute and a validated gene-insertion model family; rivals can't match that stack without similar backing.
Inside the Claude Science Pivot
On June 30, 2026, Basecamp pushed its EDEN antibiotic and vaccine models into Claude Science, Anthropic's AI workbench for life sciences. The connector puts a 28-billion-parameter genomic model (trained on Basecamp's BaseData, sourced from over one million species) behind a conversational box a researcher opens in Claude Desktop or Claude Science. BaseData holds more than ten billion novel genes.
The integration is no thin branding exercise. Anthropic bet that frontier models are smart enough; clunky workflows stall science. Opus 4.8 coordinates the workflow, linking to over 60 databases and local lab systems. The core idea decouples the user interface and execution from the underlying foundation model.
Pharma can't ship proprietary molecules or patient records to outside clouds without violating HIPAA and leaking IP. Claude Science ships only the text prompt; raw data stays in air-gapped lab servers. Jonah Cool, Anthropic's head of life sciences partnerships, said EDEN gives researchers "a new way to explore and prioritise treatments for some of the most dangerous pathogens on Earth."
The connector exposes two EDEN capabilities. EDEN-AMP generates novel antimicrobial peptide candidates against drug-resistant strains. EDEN-Immunogenicity predicts whether a protein-coding antigen triggers an immune response. Both carry a research-use-only label; Basecamp's docs say outputs are predictions, not experimental results, meant to guide lab work. On the independent ImmunoDB set, the vaccine target model scored an AUROC of 0.85. For antibiotic calls, the system handles one strain per request, and the predicted minimum inhibitory concentration applies only to that strain.
The tool promises speed, not instant cures. Penn collaborators confirmed the model's high hit rate against WHO priority pathogens in lab tests (discover-pharma.com, July 2, 2026). Drug-resistant infections kill nearly 5 million people yearly, yet major pharma has retreated from antibiotic development; the hit rate matters because bench screening is slow and expensive.
A researcher's move is plain: describe a target or pathogen, and Claude runs a prioritization workflow against the genetic sequence, returning a shortlist of high-performing antibiotic or vaccine candidates in minutes. HPCwire noted this compresses weeks of per-pathogen research into one conversation. For vaccines, the same path prioritizes protein targets from a genome, cutting weeks of bench analysis.
Gowers said: "Microbes have been producing antibiotics and evolving resistance to each other for billions of years. EDEN learned from that history, and now, through Claude, researchers all over the world can design successful new antibiotics in minutes, not years." His claim rests on zero-shot mouse data and the Penn hit rate, not approved therapeutics.
Anthropic absorbed Coefficient Bio for $400 million in April 2026, pulling computational biologists in-house. A July presentation said the company launched its own drug discovery program to show public markets that AI can cut cost per candidate, moving from chat to physical drugs. Anthropic is becoming a vertically integrated tech-bio entity, and Claude Science is where it proves clinical utility to enterprise biopharma.
Basecamp aims to scale BaseData 100-fold over two years through the Trillion Gene Atlas, a partnership with Anthropic, NVIDIA, PacBio and Ultima Genomics. The larger corpus will feed EDEN and, by extension, the Claude workflow. Today's connector is the first visible seam of that infrastructure play, not its final form.
Must Bio PhDs Yield to Mathematicians?
Basecamp Research posted a Senior Research Scientist, Mathematical Foundations role in London on June 26, 2026, one of seven open jobs on its board. The ad reads like a frontier AI lab's wish list, not a biotech listing. It demands a PhD in math, physics, or similar quant field, a strong publication record at NeurIPS, ICML, or ICLR, and hands-on training of billion-parameter models with long context and distributed custom architectures.
The high bar shows where Basecamp spends its NVIDIA cash. Five months before the listing, it closed the NVIDIA-backed financing round. The posting says its research group "operates more like a frontier research lab than a typical biotech team," building new theory for biological AI from scratch.
The scientist joins the AI Research team under its head, pushing math limits on what foundation models learn from metagenomic data. Basecamp trains those models on BaseData, its ethically sourced global corpus of newly discovered organisms. EDEN — built with NVIDIA and powering the aiPGI platform — springs from that stack, turning disease biology into therapeutic candidates already effective in mice.
Is this hire new due to the model push? Yes. The listing wants someone to "bring mathematical structure to hard problems" and see work "translated into models that generate genuinely novel biology", a mandate absent when Basecamp focused on wet-lab discovery. NVIDIA's cash landed in January; by June the lab recruited a mathematician to extend the models behind aiPGI and Claude. Not coincidence.
The handful of openings suggests deliberate expansion, not a mass hire. The listing describes existing colleagues covering genomics, sequence modeling, and large-scale deep learning, plus biologists, bioinformaticians, and engineers across the wider company. The Senior Research Scientist role is the explicit math anchor in that mix. Basecamp's global sampling network, the same partnerships that fed EDEN, gives models genetic diversity public databases lack.
Domain expertise in biology is welcome but secondary. "What we really care about is that you find it exciting to work with data from unexplored ecosystems encoding billions of years of evolutionary optimization," the posting says. The best candidates treat biological data as a source of new scientific questions, not just apply off-the-shelf ML. This flips the old biotech hiring order: math first, biology second.
Basecamp promises twice-yearly promotions and coaching through steep growth. The lure for a mathematician who can train billion-parameter models on unique DNA: publish at top venues and watch theory become a mouse-tested protein. That's the new bar, set by EDEN and NVIDIA's check.
Rivals Race to Match EDEN's Claude Coup
Basecamp's EDEN integration into Anthropic's Claude Science set a mark that peer NVIDIA-backed biotech AI labs answered within months. Finance.via.news reported that five companies, Natera, Owkin, Boltz Lab, Edison Scientific, and Basecamp itself, launched foundation model platforms recently, all on NVIDIA's stack. The move answers a window for platform lock-in in biological AI, where switching costs climb once researchers train on a given workflow.
NVIDIA expanded BioNeMo, its open drug-discovery platform, on Jan. 12, 2026. Same day, it announced Eli Lilly's AI lab on a DGX SuperPOD and a Thermo Fisher autonomous lab tie-up. Ai.via.news reported these moves pitch BioNeMo as the standard layer between AI and lab ops. Yet startups ride the same GPUs but claim distinct territory.
Owkin moved early. On Jan. 12, 2026, Owkin said it advanced OwkinZero, a frontier biology model trained on expansive patient data, with NVIDIA. Owkin argued that building the first Biological Artificial Super Intelligence needs not just powerful models but the right agentic infrastructure to scale them. The model targets discovery from clinical datasets, a different angle from Basecamp's protein engineering.
Boltz Lab struck closest. NVIDIA's January release named Boltz PBC as launcher of a molecular design platform for structure prediction. Across June 2026, Boltz paired with Takeda (June 18) and GSK (July 2), then on June 30 wired its API into Claude Science — the same conduit Basecamp opened, eroding any brief exclusivity.
Natera built on proprietary cancer genomes, per NVIDIA's January release. Edison Scientific launched Kosmos, an autonomous AI scientist that runs six months of work overnight. Both join the pile of NVIDIA-stacked models hitting enterprise use within months, a pace ai.via.news said dwarfs old pharma tech cycles that took years.
Basecamp's EDEN scales from 100 million to its largest configuration, among the largest biology models. Owkin hides its size, betting on patient data. Boltz focuses on structure, Natera on cancer, Edison on autonomous discovery. All ride NVIDIA but diverge in scientific lane.
The rush is financial. Traditional drug discovery eats 10–15 years and over $2 billion per approved drug, finance.via.news reported. Patent cliffs in 2026–2028 squeeze pharma R&D, making AI a necessity. NVIDIA VP Kimberly Powell said Jan. 12 that "Biology and drug discovery are reaching their transformer moments." Her platform supplies plumbing; startups want the model layer where researchers form habits.
Once a lab ties its data and staff to a platform, switching costs lock in, ai.via.news wrote. Each launch becomes a land grab for mindshare among scarce math-biology hybrids. By late June, Boltz had already replicated the Claude Science connector, ending Basecamp's brief exclusivity. The race hinges on who locks the researcher's daily routine first, not who published the biggest parameter count.
Working in frontier tech? Zero G Talent tracks the openings: see every open Anthropic role, browse frontier tech jobs, the companies hiring, and the people building the field.