NVIDIA in conjunction with the Lawrence Berkeley National Lab in California, has announced they will be building a new AI supercomputer called ‘Doudna’ designed to fuel real-time scientific discovery using AI, data, and simulations all under one roof to deliver the next wave of breakthroughs from fusion energy and material science to pandemic response.
The machine is scheduled to go live in 2026 and is being built on Dell Technologies hardware powered by NVIDIA’s Vera Rubin architecture. And for a bit of a surprise, Doudna a.k.a NERSC-10 is named after CRISPR pioneer and Nobel laureate Jennifer Doudna, who actually showed up at the launch and admitted she was “surprised and delighted” to have a supercomputer named after her.
Anyway, Doudna will be curated and specially design to integrate seamlessly with national experimental facilities, such as directly stream data straight from telescopes, detectors, or genome sequencers into the system through DOE’s high-speed ESnet network for almost instant analysis and feedback.
The machine’s expected to outperform its predecessor, Perlmutter, by more than 10x in scientific output, while only using 2 to 3 times the power. That means it’s not just faster but significantly more energy-efficient too — a 3 to 5x boost in performance per watt, thanks to new chip designs and smarter system-level efficiencies.
In practical terms, Doudna is going to be a powerhouse for things like:
- Modeling fusion energy simulations in real time
- Using AI to help biologists rapidly fold proteins
- Designing next-gen superconducting materials
- Processing massive astronomical data sets on the fly
The system also supports quantum workflows, meaning it’s ready to handle tomorrow’s science, not just today’s. With NVIDIA CUDA-Q and next-gen InfiniBand networking, researchers can start developing algorithms for future quantum-HPC systems too.
NVIDIA CEO Jensen Huang has expressed the vital importance of this kind of infrastructure, calling it “the foundation of scientific discovery” and even linking it to broader economic and technological leadership. And Nick Wright, the lead architect of Doudna, put it simply: this system is about helping scientists think bigger and move faster.
Chris Wright from the Department of Energy summed it up best: this isn’t just a supercomputer sitting quietly in the corner anymore. It’s part of the experiment and embedded in the workflow. Or, as Huang called it, “a time machine for science” — one that can compress years of research into just days.
Although the full system is set to launch next year, research groups of over 20 teams are already porting their work to Doudna, using tools like PyTorch, TensorFlow, Holoscan, and CUDA-Q,
Just a bit of a side note, this also kind of proves that cutting-edge technology are most likely being provided for state-sponsored purposes before hitting the commercial ground.
