Basecamp Research trains AI on 4 billion years of evolution to build general biological intelligence, powered by Microsoft Azure

Right now, somewhere in a Costa Rican rainforest, bacteria are fighting for survival. One species evolves an antibiotic to kill its competitor and claim all the resources. Viruses called phages hijack bacterial cells, inserting their own DNA so the bacteria replicate viral genes instead of their own. It is brutal, elegant, and it has been happening for 4 billion years.

This microscopic combat is exactly what Basecamp Research, a frontier AI lab founded in 2020, is hunting for. Not to study it, but to harness it. Those same evolutionary mechanisms that bacteria use to fight one another and phages use to insert DNA are the foundation for designing programmable therapeutics: antibiotics that kill drug-resistant superbugs, gene-editing systems that reverse genetic defects, and treatments for diseases that currently have no cure.

"Most therapeutically relevant molecule types—we are not finding those by sequencing the human genome or monkeys or mammals," says Phil Lorenz, CTO of Basecamp Research. "Those are often associated with fungi or with microbes or with plants. The vastness of life on Earth that is hidden in rainforests or in the deep ocean."

The problem is scale. Of all the biological diversity on Earth, humanity has studied roughly five drops compared to the entire Atlantic Ocean. Public biological databases, while growing, have plateaued. 70% of sequence data comes from just 10 species.

For Basecamp, this data scarcity was not just an academic problem. It was a fundamental barrier to training AI models capable of designing therapeutic molecules. 

"If you were to train LLM only on newspaper articles from 1975, it would be a really, really bad model," Lorenz says. "That is kind of where we are in biology."

Three ingredients for a biological breakthrough

Basecamp set out to fix biology's data wall by collecting genomic data from extreme environments across five continents, discovering more than one million previously unsequenced species. Over five years, field teams collected samples from volcanoes in Iceland, rainforests in Costa Rica, deep ocean vents in Antarctica, and everywhere in between. All samples were collected under consent and benefit-sharing agreements with the 208 organizations across 31 countries where the samples originated.

But collecting samples was only the beginning. 

"To really be successful here, we need those three ingredients to come together, which is someone scaling the data. That's us. But someone who really knows how to scale models and scale compute is Microsoft. On the scaling model side of things: How do we get the right infrastructure in place to do that? And on the compute side of things, how do we do this compute at a really big scale for the type of data that we have here," says Glen Gowers, Basecamp co-founder and CEO. 

Basecamp partnered with Microsoft to build its genomic data processing pipeline on Azure AI Infrastructure, including Azure Kubernetes Service, Azure Container Registry, Azure Blob Storage, Azure Files, and Azure Machine Studio.

That first ingredient, scaled data, required Basecamp to transform raw DNA sequence reads into complete, labeled genomes with protein-level annotations, taxonomic information, and biological context. The processing pipeline would require hundreds of millions of CPU hours to assemble and annotate genomes. And nature does not operate on a predictable schedule. Sample batches arrive irregularly from field teams scattered across five continents. Sometimes the team receives a few samples over several days. Sometimes 100 samples arrive at once.

"We can't tell nature or the people collecting samples, 'It would be nice if you could have 10 samples a day,'" Lorenz says. "Azure was great with the flexibility we could leverage with respect to scale and timing."

Combining expertise to move faster

Each Azure service played a specific role in meeting Basecamp's infrastructure demands at scale. Azure Kubernetes Service enabled dynamic scaling based on incoming sample volume rather than maintaining static infrastructure that would create bottlenecks during peak periods or waste resources during slower periods. Azure Container Registry streamlined deployment of annotation tools and workflows across the distributed computing environment, while Azure Blob Storage provided scalable long-term storage for the exponentially growing dataset.

"Ten times more data also means ten times more annotation and labeling pipeline and the compute that is required for that," Lorenz notes.

By the time Basecamp was ready to train EDEN, their foundation models for therapeutic design, they had assembled 9.7 trillion nucleotides of biological sequence data from BaseData. . The model training itself, conducted in collaboration with Microsoft researchers, used compute resources comparable to those of GPT-4. Advances in AI infrastructure have enabled us to scale our data a lot quicker than we otherwise might have.," Gowers says. "And so now we can combine all three of those scaling relationships to really prove that these models can get extremely intelligent across extremely diverse and generalized tasks a lot sooner than we might have otherwise assumed."

Azure Machine Learning Studio provided the infrastructure for training these large-scale models, enabling Basecamp to leverage the evolutionary signals captured in their dataset. 

"Evolution underlies all biological processes,” Lorenz says. “A lot of really powerful evolution actually doesn't happen within a single genome. It happens between species: bacteria fighting each other, phages inserting DNA into bacteria. That's where these therapeutically-relevant molecules like peptides for antibiotics or gene editing systems arise."

Three breakthroughs in three weeks

Basecamp designed EDEN to validate three therapeutic applications: gene insertion, antimicrobial peptide design, and microbiome generation. Success on even one would have been significant.

Within about three weeks in November 2025, the team saw successful lab results for all three.

"I had started planning this EDEN project saying if one of those three challenges works, that's great," Lorenz recalls. "But for all the applications, we got successful lab confirmation. That was a good moment."

For gene insertion, EDEN achieved world-first AI-programmable design, creating active proteins for 100% of tested disease-relevant target sites in the human genome. In lab validation, the integration of cancer-fighting DNA into primary human T cells achieved over 90% tumor cell clearance.

The breakthrough solves a critical scaling challenge in gene medicine. Most genetic diseases involve different mutations per patient, which traditionally requires designing custom treatments for each individual. Basecamp's approach uses proteins that insert healthy gene copies regardless of specific mutations, creating a scalable therapeutic solution.

The challenge: these insertion proteins evolved for bacterial warfare, not human therapeutics. 

"We had to train a large AI model so it generalizes and sees all of these proteins and insertion sites," Lorenz explains. "Then, because the model has generalized, it can be prompted on a sequence from the human genome and make these proteins programmable."

EDEN also designed antimicrobial peptides with 97% lab-confirmed success rates, including candidates highly potent against multidrug-resistant bacteria. The system leveraged evolutionary patterns from the bacterial warfare captured across millions of newly sequenced species.

In perhaps the most ambitious validation, EDEN generated an entire synthetic microbiome comprising 9,000 bacterial species that represent a complete digestive system. Generation happened at the billion-nucleotide scale with 99% biological accuracy.

To validate the results, Basecamp consulted Jill Banfield at UC Berkeley, who pioneered microbiome research. 

"She is an icon in the field and a very skeptical scientist," Lorenz says. "We asked what would make her believe someone generated an AI synthetic microbiome. She gave us tough metrics to hit. We convinced her. She is a co-author now on the paper. It is good to ask a skeptical human and not just AI."

Scaling biological models in ways never done before

The January 2026 announcement of EDEN's results, published in research co-authored by Microsoft, and leading academics, marked a breakthrough in programmable therapeutic design. The achievement required bringing together Basecamp's unprecedented data collection and, Microsoft's infrastructure for for training models at massive scale.

"We are the only company now to have scaled this data, gotten past this data wall," Gowers says. "And now with those three things together, a combination of these three companies and three technologies that we have all built can now start scaling these biological models in a way that we just could not before."

But proving that AI can design active therapeutic molecules is different from proving those molecules are ready for patients. Basecamp's next phase involves integrating clinical data to upgrade EDEN's capabilities around safety and toxicology.

"These molecules are great, they are active, they work," Lorenz says. "But for them to go into the clinic, there is a lot more that needs to happen. We want to upgrade our AI models to consider those questions too. Not just design toward the right activity, which is hard enough, but also the next therapeutically-relevant layer."

The company is maintaining its 10X annual database growth target, which means continued scaling of Azure infrastructure for genome assembly, annotation, and orchestration. The potential applications stretch across gene therapies for thousands of currently untreatable diseases, antibiotics that defeat resistant bacteria, engineered microbiomes, sustainable chemicals, and beyond.

"Our ambition is to make biology programmable,” Lorenz says. “You prompt on a disease and out comes a molecule that will address that."

With EDEN trained on evolutionary warfare from rainforests to ocean depths, and Azure AI Infrastructure enabling the exponential data scaling that breakthrough required, that ambition is becoming achievable. From Costa Rican rainforests to Antarctic ocean vents, bacteria have perfected their molecular tools over billions of years. Now, through the combination of Basecamp's data collection and Microsoft's infrastructure, humanity can learn that language and put it to work.

Watch and learn how innovators are leveraging technology to unlock new possibilities for a better future in this Catalyst Series episode featuring Basecamp Research.

Discover more about Basecamp on LinkedIn, X/Twitter, and YouTube.