DARPA just deployed a managed AI and HPC platform built by Parallel Works and CoreWeave, and the goal buried in the announcement is the part worth pausing on: the agency wants to identify a biological threat, and understand what it actually does, within an hour of encountering it.
The project driving this is called NODES, short for Network of Optimal Dynamic Energy Signatures. Researchers are trying to build a deep learning tool grounded in biophysics that can scan vast numbers of protein sequences and predict their biological function by picking up on subtle patterns in how proteins move. That’s a narrow, technical goal on paper, but the stakes behind it are anything but small. An hour is roughly the window a response team gets before a biological threat moves from detected to dangerous.
DARPA runs the platform through Parallel Works’ Activate control plane, which sits on top of CoreWeave’s AI cloud infrastructure. Underneath all of it are Nvidia HGX H100 systems, dedicated specifically to this work rather than shared across other customers. DARPA’s original tender, issued back in April 2026, asked for 50 dedicated H100 GPUs with another 100 held in reserve purely for redundancy, a detail that says something about how little tolerance the agency has for downtime on a project like this.
Parallel Works CEO Matthew Shaxted framed the partnership around keeping researchers focused on biology instead of infrastructure management, arguing that NODES teams shouldn’t lose time building or maintaining compute environments when they could be advancing the actual research. CoreWeave co-founder and CEO Michael Intrator echoed that, describing his company’s job simply as giving researchers the computing power to take on DARPA’s hardest problems.
What makes this partnership notable isn’t the GPU count, though that’s substantial. It’s the pairing itself. Parallel Works, spun out of Argonne National Laboratory back in 2015, brings deep HPC management experience built for exactly this kind of government and research work, while CoreWeave supplies the raw AI cloud capacity at a scale few providers can match right now. Together, they’re betting that protein-level biological threat detection is less a research problem than an infrastructure one, and that solving the infrastructure piece fast enough is what actually gets DARPA closer to that one-hour target.
