Inside the DOE's Genesis Mission: 278 Projects Building AI as Scientific Infrastructure

Genesis Mission: AI as National Scientific Infrastructure

Since the White House unveiled the Genesis Mission in November 2025, the program has aimed to accelerate breakthroughs in energy, scientific discovery, and national security by combining AI, supercomputing, quantum systems, and advanced scientific instruments. In July 2026, the U.S. Department of Energy (DOE) moved into large-scale execution, selecting 278 Phase I projects from over 5,000 applications across 342 institutions — including DOE and NNSA national laboratories, universities, companies, and nonprofits. The projects span energy, materials, biotechnology, quantum science, particle physics, and advanced manufacturing, placing AI at specific bottlenecks within each scientific domain.

Case 1: Accelerating Monte Carlo Event Generation for the LHC

At the University of Oregon, physicist Stephanie Majewski leads a team awarded $662,000 for the MANGO project ( Monte Carlo Acceleration via Normalizing Flows using GPU Optimization ). The bottleneck: the Large Hadron Collider (LHC) produces massive collision data, but physicists must first simulate billions of "what-if" collision events using Monte Carlo generators to compare with real detector data. This simulation step consumes enormous compute resources.

MANGO redesigns event generation using normalizing flows — a generative AI technique — optimized for GPUs, targeting both speed and accuracy gains. Crucially, the approach aims to be hardware-agnostic, running on both GPU and CPU environments without vendor lock-in. For particle physics, faster simulation means more events, more theory tests, and more compute for the hardest physics problems.

Figure 1: Schematic of LHC experiment simulation.

Related link: https://barefield.ua.edu/2026/08/12/ai-meets-particle-physics-in-ua-led-department-of-energy-project/

Case 2: Discovering Cross-Scale Laws in Plasma Dynamics

Emory University physicists Justin Burton and Ilya Nemenman tackle a deeper question: can AI uncover physical laws from complex plasma motion that humans have not yet derived? Their project spans scales from laboratory dusty plasmas (~ 2,000 particles ) to astrophysical extremes like supernovae and gamma-ray bursts.

Traditional methods treat each scale separately. The Emory team aims to train a single AI tool for cross-scale plasma fluid dynamics, seeking a higher-level law akin to the ideal gas law but for complex plasmas. This builds on prior work where AI revealed previously unnoticed phenomena in dusty plasma experiments, earning the 2025 Cozzarelli Prize from the National Academy of Sciences.

Figure 2: Hubble image of the Crab Nebula, an astrophysical plasma environment.

Figure 3: Burton featured in the NAS Cozzarelli Prize video.

Related link: https://news.emory.edu/stories/2026/07/emory-scientists-selected-us-genesis-mission-awards-speed-discovery-through-ai

Case 3: AI Agents for High-Energy Physics Simulation and Analysis

The University of Alabama (UA) leads a project using AI agents to automate multi-step high-energy physics workflows. Unlike traditional AI tools that wait for a scientist to pose each task, agentic frameworks like HEPTAPOD let AI organize sequences of simulation, theory calculation, data processing, and analysis — calling specialized computational tools autonomously.

High-energy physics is a natural fit: every stage (simulation, theory, data handling, analysis) already has complex software and mature compute infrastructure. UA's team, collaborating with CERN's Compact Muon Solenoid (CMS) experiment , built a 2025 pilot applying large language models to particle theory tasks. That prototype now becomes the backbone of the Genesis project, scaling to automate CMS's massive data streams. The methods are designed to transfer to other data-intensive disciplines, boosting national research productivity.

Figure 4: The CMS experiment's massive solenoid magnet at CERN.

UA leads the joint project with Fermilab; Google and NVIDIA provide additional support.

Related link: https://research.uoregon.edu/news/uo-physicist-earns-genesis-mission-award

Phase I Goals and Infrastructure

The first phase aims to identify promising paths toward transformative scientific capabilities and lay groundwork for future investment and scaling. Teams will design and demonstrate AI-integrated research workflows while rigorously evaluating whether these methods accelerate discovery, improve prediction, enhance experiments, or yield new scientific insights. Awardees gain access to mission platforms including industry-partner AI frameworks, advanced models and software, and DOE national laboratory high-performance computing resources.

Code example

来源：ScienceAI
本文
约2500字
，建议阅读
7
分钟
粒子模拟、等离子体与科研智能体，勾勒AI for Science新图景。