Picking the Right Challenges for Genesis Mission

Overview

Genesis Mission, launched by the Trump administration via an executive order last November, aims to “double the productivity and impact of American science and engineering within a decade.” More concretely, it is an initiative led by the Department of Energy (DOE) that aspires to incorporate AI into the Department’s research, development, and operations by building an AI-ready technology and data platform and using that platform to solve 26 energy-related challenges. Example challenges include Accelerating Delivery of Fusion Energy, Predicting US Water for Energy, and Safeguarding Nuclear Materials from Proliferation Threats. 

The initiative aims to accomplish its goals by leaning into public-private partnership agreements, which could help the federal government learn from industry how to effectively apply AI to scientific discovery and agency operations while building on DOE’s strengths. These strengths include unique datasets, world-class instrumentation, and unparalleled talent at the national labs. Genesis Mission also provides a unique and vital pilot for applying AI not just to scientific research but also to the unique operations and workflows of scientific institutions.

DOE is preparing to announce its first awards toward the 26 challenges, and Congress is currently deliberating about whether or not to grant DOE additional appropriations for the initiative. For Genesis Mission to earn congressional support, DOE should pick fewer than 26 priority areas, clarify whether the platform will be focused primarily on general scientific use or on achieving the challenge directions, and focus on achieving concrete outcomes that it can communicate to Congress and the American public. The Genesis Mission executive order frames the initiative as an urgent national effort comparable to the Manhattan Project, but the Manhattan Project had one goal, not 26. More clearly and concretely defining the goals of Genesis Mission will make it easier for DOE to communicate its potential impact on the lives of ordinary Americans and provide quantifiable metrics that can be used to evaluate its success. If DOE can provide this clarity, then Congress should appropriate the additional funds necessary to achieve these outcomes and then hold DOE accountable as it works to achieve them.

The big ambitions of Genesis Mission 

DOE has a broad spectrum of responsibilities, including supporting energy technologies through basic and applied research, development, and permitting, and maintaining and safeguarding our nuclear stockpile through the National Nuclear Security Administration (NNSA). The 26 challenges of Genesis Mission span these areas and are largely drawn from existing research projects at the 17 DOE national labs (see the Appendix for a summary of each challenge).¹ For example, Delivering Nuclear Energy that is Faster, Safer, Cheaper builds on work at Idaho National Lab to develop digital twins — computer models that replicate real-world processes — that can speed up the reactor development process, and Scaling the Grid to Power the American Economy builds on grid modeling and simulation efforts at the National Lab of the Rockies in Colorado.

As evidenced by the above challenges, Genesis Mission is more than an AI-for-science initiative. It is an effort to incorporate AI throughout all aspects of DOE. In addition to applying AI to basic research, the challenges call for AI to be deployed to speed up operations, permitting, and other bureaucratic functions of the Department. The challenge descriptions also make clear that off-the-shelf LLMs alone will not be enough to meet the goals of Genesis Mission. Bespoke models will need to be developed for many of the challenge directions.

Across the challenges, Genesis Mission aspires to use AI to:

• Deliver more scientific breakthroughs
  • in quantum computing, sensing, and communication, including the Quantum Genesis initiative; 
  • in materials science;
  • in the physical laws that govern the cosmos;
  • in predicting water availability;
  • by making more and better autonomous labs; and
  • by optimizing particle accelerators.
• Reduce costs in
  • energy (by delivering more energy from nuclear power plants, accelerating the timeline of fusion, better characterizing underground sources of oil, gas, and heat, and modernizing the energy grid);
  • nuclear cleanup and stockpile stewardship; and
  • housing construction and operation.
• Improve national security by
  • better detecting and responding to nuclear threats;
  • deploying more and better nuclear weapons;
  • improving nuclear deterrence and proliferation capabilities; and
  • reducing America’s dependence on foreign sources of critical minerals.
• Restore US manufacturing prowess in biotech and microelectronics by better bridging the “valley of death,” which too often prevents scientific discoveries from entering the market. 
• Make better data centers, faster.
• Digitize old nuclear records.

Expand

Genesis Mission also plans to build out an AI-for-science technology platform that supports the 26 challenges and serves as a resource for the broader scientific community. This effort would build on DOE’s long history of developing datasets and providing supercompute for the public good. (As one example, Brookhaven National Lab began the repository of 3D protein structures that would eventually become the Protein Data Bank, which was used to train the Nobel Prize-winning natural law model, AlphaFold.) DOE also gives scientists access to high-performance computing (HPC) for training models and running simulations on its flagship supercomputers through a number of programs. The new platform developed as part of Genesis Mission would consist of datasets, AI models, HPC, and autonomous labs, and be integrated into the American Science Cloud, a to-be-built cloud environment that is also among the initiative’s goals. 

DOE has given out $320 million in awards under Genesis Mission for development of the platform, and a request for applications for an additional $294 million for teams to work on the challenges recently closed, with awardees to be announced soon. Most of this funding comes out of the regular DOE annual appropriations, with ongoing efforts previously funded by the Office of Science, which has an annual budget of around $8 billion, or out of Laboratory Directed Research and Development funds. Some of this funding also comes from the One Big Beautiful Bill, which gave DOE $150 million for civilian AI-for-science work and $115 million for NNSA. 

Though President Trump did not directly request money for Genesis Mission in his fiscal year 2027 budget request to Congress, the budget did request $1.2 billion for a new Office of Artificial Intelligence and Quantum at DOE. This new office would support Genesis Mission, including the construction of new AI supercomputers. Congress is currently considering this request. The House has so far signaled that it will not include this extra money, while the Senate is still deliberating.² 

A successful Genesis Mission will demonstrate how to improve federal R&D with AI

The DOE national labs have a long history of generating new scientific breakthroughs that have benefited the lives of millions. If implemented well, Genesis Mission can build on this history of success, while also engaging with the datasets and unique strengths of other federal agencies, to:

• Ensure that publicly funded research and, by extension, the public benefit from industry advances in developing and applying AI to scientific research and development. Private companies currently lead in these areas and are often better-resourced in compute, talent, and dollars than the federal research enterprise.³ As a result, Genesis Mission places a heavy emphasis on public-private partnerships. Proposals submitted to the initiative must come from some combination of national lab staff, academics, and industry researchers. DOE has also established a Genesis Mission Consortium for industry and academic stakeholders. Members are expected “to contribute computing power, AI tokens, technical expertise, or in-kind support” that will help advance the Genesis Mission and, in return, will benefit from access to DOE data, expertise, and infrastructure. So far, at least 24 organizations, including OpenAI, Anthropic, and NVIDIA, have signed memorandums of understanding with DOE. If Genesis Mission structures these public-private partnerships to ensure that federally funded scientists can leverage private resources and learn from industry experts, rather than simply giving funds, data, or scientific expertise to industry, the federal research enterprise would be better positioned to stay at the bleeding edge of AI-for-science, rather than being left behind.
• Demonstrate how AI can improve the federal scientific enterprise. Other federal science agencies, including NSF, NIH, NASA, and NIST, stand to learn from Genesis Mission, and DOE can also learn from these agencies, which have been building useful AI tools for science for some time now. Areas of possible symbiosis include how to generate and host AI-ready datasets and how to effectively utilize academic, federal, and industry partners to build new AI models to accelerate discovery. Genesis Mission can also demonstrate how to evaluate whether new AI models are a better fit than frontier LLMs for a given use case, and what research directions are better served by autonomous and cloud labs than by human researchers. Relatedly, Genesis Mission challenges linked to NASA, NSF, or NIST datasets are also being considered.
• Teach the broader federal enterprise how to apply AI to process integration, digitization, and operations. Lessons in applying AI to DOE operations could be extended to other agencies. For example, the Department of the Interior could apply the lessons learned from the Delivering Nuclear Energy that is Faster, Safer, Cheaper challenge to use AI to speed up the development process for new offshore oil or gas rigs or wind turbines while also reducing cost. 

DOE should streamline its goals and messaging

Building the technology to support specific challenges at DOE is a different task from building out a general-use AI-for-science platform, and the optimal program for the former may not be the optimal program for the latter, though they may overlap. Supporting the challenges requires focus on the most directly related programs and facilities at the corresponding national labs. For example, Transforming Nuclear Cleanup and Restoration, which aims to train foundation models on nuclear cleanup operations, describes drawing on the expertise and datasets of Savannah River National Lab and leveraging the Equinox supercomputer at Argonne National Lab. In contrast, building turnkey AI-for-science infrastructure will require carefully integrating and expanding the full AI-for-science technology stack across DOE. An AI-for-materials science program, for example, could integrate expertise and data from across the national labs, autonomous labs and compute from Berkeley, Oak Ridge, and Argonne National Labs, data services from Jefferson Lab, etc. into the American Science Cloud. While both approaches may be effective paths for driving innovation and impact, taking on both at once will be a big lift for the Department, especially without additional resources. The Office of the Under Secretary for Science should decide whether its top priority is the challenge directions — and, if so, which ones — or the platform, and then streamline Genesis Mission with clear goals and structures for science and technology that are tailored to that primary mission.

Further, the 26 very different challenge directions on top of the platform make it difficult for DOE to clearly articulate the focus and possible transformative outcomes of Genesis Mission. Downselecting from the 26 challenges and the varied technological goals of the platform to instead focus on a handful of outcomes would help DOE communicate a clear message to Congress and to the public that could help win future support. Focusing on a subset of priorities will also make it more likely that each priority has the resources and attention it needs to succeed.⁴

When downselecting challenges, DOE should lean into the government’s strengths and offer quantifiable outcomes

If DOE decides to prioritize specific challenge areas, the Office of the Under Secretary for Science should:

• Define clear outcomes that will demonstrate the program’s impact to Congress and the public and that DOE can use to evaluate the initiative’s success. For example, the chosen areas can be framed as grand challenges — ambitious but achievable goals that require sustained coordination and collaboration to target difficult problems with major societal impact. DOE already has experience with grand challenges, and the Quantum Genesis initiative could be interpreted as a grand challenge with its call to “create and deploy the world’s first scientifically relevant, fault-tolerant quantum computer.” Some of the challenges also already hint at this approach, such as the Delivering Nuclear Energy that is Faster, Safer, Cheaper challenge, which targets a 2x schedule acceleration and more than 50% operational cost reductions for nuclear reactors. Explicit goals like this help to communicate the value of Genesis Mission.
  
  When considering possible outcomes, the Under Secretary’s Office should ask questions such as: Is the goal to reduce electricity costs, and if so, by how much? Is the goal to reduce the cost to taxpayers of nuclear cleanup and, again, by how much? Is the goal to drive new materials discovery and, if so, how will those materials change the lives of ordinary Americans? Is the goal to increase national security in defined ways that can be reported to members of Congress in classified briefings? Some of these goals are buried within the current challenge statements, but not in all cases.
  
  At a minimum, DOE should clarify which set of concepts Genesis Mission is focusing on. The recent Genesis Mission request for applications listed 21 priority areas that have incomplete overlap with the 26 challenges. 
  
  
• Focus on directions where the federal government adds unique value: basic science, classified research, and national coordination challenges like the electrical grid. Though DOE is establishing a new Office of AI and Quantum to oversee Genesis Mission, it should make sure not to solely focus on general AI-for-science tools or on quantum computing. Industry and academia are making rapid progress in developing AI-for-science tools, and, though quantum computing is often conflated with AI, it is a unique field of science deserving of continued support but still in its relative infancy with limited real-world applications to date.

DOE will face political challenges by picking a reduced handful of priority areas. Different national labs work on different parts of DOE’s mission, and, in downselecting, some national labs may therefore benefit more than others. This asymmetry would likely create friction within DOE and in Congress, where senators and representatives may push back on Genesis Mission funding if it does not benefit their state or district. If done carefully, however, pain to individual national labs can be minimized. Ongoing work and existing grants at the labs should continue, and all labs should be allowed to compete for the chosen directions, as they do for other DOE grants. This will allow natural winners to emerge over time and give the lab ecosystem time to adapt.

For the platform, DOE should be selective about which components to develop

If DOE decides to prioritize the platform, the Office of the Under Secretary for Science should be clear-sighted when building out its different elements. If the platform is intended as a turnkey AI-for-science tool, how will the different elements of the platform be integrated across the DOE ecosystem? If the platform is more like an extension of existing DOE user facilities, which elements should DOE focus on? Specific considerations for each element include:

• Datasets: Generating scientific datasets is a clear area of past and present strength and should be continued, especially with DOE’s unique slate of large-scale scientific equipment — particle accelerators and detectors, light sources, and so on.
• New AI models: Industry has so far led the way in developing scientific AI models, and so Genesis Mission should think carefully about the value add of any new model before supporting it. The value of a new model from Genesis Mission could include making any model open source and open weight, the opportunity to train scientists in modern AI methods through new model development, and the ability to leverage unique scientific knowledge bases from the national labs.
• Autonomous labs: Private companies are increasingly moving into this space, and so DOE should recognize where it can push the frontier and where it should let industry lead. While the national labs pioneered autonomous labs for materials science at Lawrence Berkeley, Argonne, and Oak Ridge National Labs, startups like Lila Sciences, Periodic Labs, and Radical AI have now raised significant capital to continue to develop this area. Pacific Northwest National Laboratory’s recent partnership with Ginkgo Bioworks to build an autonomous lab for anaerobic microbial experimentation suggests that DOE is already thinking about what new areas of science to automate.
• Computing: Not all compute is created equal. Computing efforts at DOE can roughly be grouped into three different areas: conventional supercomputing for scientific modeling and simulations, HPC for training and running AI models, and quantum computing. While DOE has been a longtime leader in conventional supercomputing, China recently took the lead, building the world’s most powerful microprocessor-driven computer. DOE also launched the HPC revolution that underlies AI by embracing GPU-based computing for parallelizing scientific computations in partnership with NVIDIA, though more recent GPU computing for AI has been driven by industry investments of over $100 billion in new data centers. DOE similarly is a longtime sponsor of quantum computing, where US industry now leads. As evidenced by the Quantum Genesis initiative announcement and Argonne National Lab’s recent partnership with NVIDIA and Oracle to add two new DOE supercomputers, DOE clearly aspires to maintain the US’s global leadership in these types of computing. However, as compute spending from industry is so much greater than what the federal government can bring to bear, DOE should think carefully about its role and whether it is better positioned to act as a coordinating and derisking body rather than as a standalone leader.

Important structural decisions at DOE can further maximize impact

Regardless of whether or where Genesis Mission ultimately decides to focus, decisions about how to structure programs and where to allocate resources within them can affect their success. To maximize the impact of Genesis Mission, the Under Secretary’s Office should: 

• Continue to support basic research and innovative ideas that bubble up from the national labs and academia rather than putting all of its chips into the Genesis Mission. Many of today’s advances in biology, such as CRISPR or AlphaFold, stem from small, early-stage investments from DOE, and investments like these should not all be subsumed to the Genesis Mission challenge areas. Further, AI will not accelerate all areas of science equally, as some fields still lack the data or instrumentation needed for AI-mediated design and discovery, and these areas will still need support. 
• Get discoveries out of the lab and into the real world, overcoming the so-called valley of death. The valley of death is the difficult transition period faced when trying to turn early-stage research into commercially viable products. Inadequate funding and lack of market readiness often lead early-stage startups to fail. As the priority areas surface new technologies, existing mechanisms, such as ARPA-E’s SCALEUP program, should be leveraged to support commercialization. SCALEUP provides nondilutive grants to startups working on transformative technologies. Genesis Mission investments need to be balanced against DOE’s capacity to transition technologies out into the marketplace.
• Choose funding mechanisms that match the programs’ goals. The Office of Science has utilized other transactions (OT) agreements to structure awards. OTs can be a very effective mechanism for moving quickly and structuring flexible, milestone-based contracts, but they are just one out of many possible funding mechanisms. The Office should continue to be thoughtful about choosing the funding mechanisms that are most effectively aligned with their programmatic goals. The goal of one priority area may be better served by a different combination of funding approaches than another.
• Build the human infrastructure that is necessary to set up successful programs and public-private partnerships. Crosscutting Genesis Mission goals are likely to need coordination from within the Department rather than just by the stakeholders, and strong program managers will be needed to guide these areas. Effective public-private partnerships will also require full-time staff who are knowledgeable about the wide range of possible legal agreements and can creatively structure the agreements to meet the needs of both the government and industry. 
• Support the technology necessary for successful public-private partnerships. This could include, for example, developing systems that allow different users access to different subsets of data, metadata, or models in order to gate off sensitive intellectual property while still allowing multiple stakeholders to work together on a project.
• Evaluate how success relates to organizational structure. Phase II awards for the recent Genesis Mission request for proposals are meant for large teams and require a triple-helix partnership between national labs, industry, and nonprofits, including academia. Any of these entities can serve as the lead. After the duration of the award, DOE should evaluate the effectiveness of the different partnership structures to determine whether certain management and organization philosophies lead to better or worse overall outcomes.

Congress should require clarity on the concrete goals of Genesis Mission — and then support it

Applying AI to ongoing work at the national labs has historically garnered bipartisan support. While Genesis Mission is an initiative of the Trump administration, the Biden administration also proposed an AI-for-science effort called Frontiers in AI for Science, Security, and Technology (FASST) initiative, though that effort never got beyond the planning stage. 

However, the increasingly large federal deficit puts pressure on any new initiatives to offer clear and compelling outcomes to justify new congressional spending. When DOE can clearly articulate its scientific and technology goals, with quantifiable metrics and a clear spend plan, then Congress can evaluate whether Genesis Mission passes this bar, and, if so, it should increase DOE appropriations to support the initiative. 

Quantifiable metrics and explicit outcomes will also help congressional committees perform directed oversight as Genesis Mission progresses, and Congress should request progress reports toward these concrete outcomes in appropriations report language. Careful oversight can make sure the initiative stays on track, holding DOE accountable for its promises.

Conclusion

Genesis Mission can bring out the best in industry, academia, and the national labs to unleash AI-accelerated discovery — but only if it focuses and outlines clear goals. An overly broad Genesis Mission that fails to deliver will lower overall long-term support for AI at DOE, especially at a time when AI polls poorly with the public. A handful of successful demonstrations will make a better case for sustained support, allowing DOE to expand AI across more priority areas and ultimately bring more resources to the national labs, all while demonstrating tangible wins to Congress and to Americans. If the Office of the Under Secretary of Energy for Science can distill the many diverging Genesis Mission directions into clear, quantifiable outcomes, then Congress should follow through on bipartisan support for AI-for-science at DOE and fund the initiative.

Appendix

The 26 challenges of the Genesis Mission can be roughly grouped as follows, though there is substantial overlap across categories. 

Basic science breakthroughs

Challenges that aim to develop fundamental understanding of natural laws and make experimental tools and facilities more capable and autonomous.

Unifying Physics from Quarks to the Cosmos Develops AI models that simultaneously learn from particle collisions, nuclear decays, and cosmological surveys to internalize the Standard Model and thereby accelerate discovery in high-energy and nuclear physics.

Discovering Quantum Algorithms with AI Uses AI to automate and optimize the design and translation of new algorithms for quantum computing. 

Realizing Quantum Systems for Discovery Applies AI to understand and control quantum systems for quantum computing, sensing, and communication.

Designing Materials with Predictable Functionality Builds AI frameworks that couple materials prediction, synthesis, characterization, and analysis in closed-loop to design materials that exhibit desired property specifications.

Predicting US Water for Energy Develops AI models trained on atmospheric and terrestrial observations that can predict cloud physics, surface and subsurface water flows, and the broader hydrologic cycle.

Achieving AI-Driven Autonomous Laboratories Integrates AI into experimental workflows and data analysis to more rapidly implement complex experimental designs over large parameter spaces.

Enhancing Particle Accelerators for Discovery Uses AI to optimize accelerator design and operation.

Increasing Experimental Capacity at Nuclear Research Facilities Proposes an AI “facility operating system” that uses agentic workflows to plan, schedule, and steer experiments in real time. 

Cost reduction and efficiency

Challenges that seek to reduce costs and increase efficiency by streamlining exploration, design, construction, permitting, and operations

Reimagining Construction and Operation of Buildings Applies AI to automated design, physics-based modeling, rapid permitting, and optimized maintenance to lower construction and operations costs.

Delivering Nuclear Energy that is Faster, Safer, Cheaper Uses AI to accelerate reactor design, licensing, manufacturing, construction, and operation in human-in-the-loop workflows, targeting at least 2x schedule acceleration and more than 50% operational cost reductions

Accelerating Delivery of Fusion Energy Builds AI models to accelerate fusion infrastructure development and shorten innovation cycles.

Unleashing Subsurface Strategic Energy Assets Develops AI that can predict the behavior of subsurface reservoirs for extracting unconventional oil and gas, geothermal, and coal bed methane.

Scaling the Grid to Power the American Economy Applies AI to improve grid planning, interconnection, operations, and security, targeting 20–100x faster decision-making and at least a 10% improvement in electricity cost and reliability.

Transforming Nuclear Cleanup and Restoration Trains an AI foundation model on decades of cleanup operations data to reduce liability

Streamlining Production, Removing Red Tape, and Ensuring Safety in the Nuclear Enterprise Deploys AI to automate safety analyses, documentation, and risk-aware work planning in high-hazard facilities, targeting 50% reductions in planning and documentation time.

National Security

Challenges centered on nuclear security monitoring and operations.

Accelerating Nuclear Threat Assessment, Preparedness, and Response Deploys an AI system that fuses radiation sensors, simulation, and intelligence reporting for real-time decision support for nuclear and radiological events, aiming to shrink detection-to-response times from days to hours.

Safeguarding Nuclear Materials from Proliferation Threats Builds AI models that fuse satellite imagery, sensing, and open-source and government data to detect proliferation anomalies in near real time and generate evidence packages.

Strengthening Deterrence Through Attribution of Nuclear and Radiological Signatures Fields forensic AI that combines imaging, spectroscopy, morphology, and inverse modeling to rapidly trace the origin of nuclear materials.

Integrating Design and Production Operations for Nuclear Deterrence Creates a “nuclear security enterprise twin” to shrink the iteration time between design and production for new weapons systems.

Accelerating Materials Discovery, Production, and Qualification for Strategic Deterrence Links material design, automated testing, and qualification in an AI-driven workflow to accelerate the development and certification of mission-critical nuclear and non-nuclear materials.

Securing America’s Critical Minerals Supply Uses AI to identify alternative materials that reduce dependence on foreign sources and to accelerate mineral development timelines.

Manufacturing

Challenges that aim to improve manufacturing processes.

Scaling the Biotechnology Revolution Integrates genomics, multi-omics, and imaging data into AI models to accelerate the industrial scale-up of biofuels, biochemicals, and bioproducts.

Reenvisioning Advanced Manufacturing and Industrial Productivity Applies AI to cross the “valley of death” between scientific discovery and commercially viable products.

Recentering Microelectronics in America Builds an AI-driven full-stack co-design ecosystem linking materials, devices, and manufacturing workflows to accelerate new microelectronics research, development, and manufacturing.

Data Center Technology Development and Deployment

A challenge that aims to develop and deploy improved data center technology.

Securing US Leadership in Data Centers Uses AI and machine learning to derisk new data center technologies and their integration with the grid.

Digitizing Historic Data

A challenge that seeks to convert analog, legacy records into usable datasets.

Harnessing America’s Historic Nuclear Data and Research Builds an AI pipeline to digitize eight decades of analog nuclear records.