Big Science Without Big Labs: AI Agents are Ready to Deliver 2025’s Medical Breakthroughs

The medical team is a mix of bioengineers with backgrounds in biology and neuroscience. There’s no clinic, and their laboratory is a data center full of NVIDIA HGX H100 GPUs. 

Welcome to Indie Science

FutureHouse is part of a new wave of research orgs redefining how science gets done. At the heart of their work is AI agents, some bought and some built in-house, on-demand compute, and outsourced wet labs.

“Even if we had all the information we needed in order to understand how the brain works, we wouldn't necessarily know it. Because no one has enough time to go and read all the scientific literature,” explains founder Sam Rodriguez. “ And even if they could, they wouldn't be able to go and assemble it into one kind of comprehensive whole … You wouldn’t be able to remember when you got to the end, what you read at the beginning.”

That was the catalyst for building an agile research stack that would become FutureHouse’s AI scientist, Robin. The AI scientist is composed of an unified system made of three specialized agents: Crow, Falcon, and Finch. The integration of these separate systems into a single workflow will continue to be used to propose and validate novel treatments and therapies for human diseases. In essence, rather than spending millions setting up physical laboratories and in-house teams, the multi-agent system does it with:

• AI agents which design and refine hypotheses.
• Contract research orgs (CROs) to handle physical experiments and data collection.
• Cloud GPUs (like the NVIDIA H100s they run on Voltage Park) for powering large-scale training and inference.

The result is faster discoveries backed by large corpus medical texts, and constant iteration that doesn’t rely on funding rounds.

This big, lean, bold bet on AI + bioscience is working.

In May, FutureHouse announced Robin had successfully gone through a hypothesis and investigative iteration that led it to determine that an existing medicine may also be used to treat a serious eye disease that can lead to blindness.

Related: Demonstrating end-to-end scientific discovery with Robin

AI Compute as a Catalyst

Of course, this model only works if AI compute is flexible and affordable. That’s where Voltage Park comes in.

Voltage Park gives them supercomputer-grade GPUs without the red tape.

FutureHouse runs on NVIDIA H100s, the gold standard for AI training and inference. But instead of signing year-long leases or provisioning idle capacity, they spin up what they need, only when they need it.

“We’re running a serious lab without owning a lab. That only works because we can rent H100s by the hour, scale to hundreds of GPUs overnight, and shut it all down after an experiment finishes. That’s what Voltage Park enables.”
— Sam Rodriques, CEO & Co-Founder of FutureHouse

Think Bigger, Start Smaller

What FutureHouse shows is this: in 2025, you no longer need a big lab to do big science.

If you have an idea worth chasing, the right team, and the right agents, GPUs, and CROs you can punch above your weight.

And if you're building something bold in the bio/AI space, your GPU stack shouldn’t hold you back from making the next big breakthrough - sans the lab.

‍