Google DeepMind and DOE launch Genesis: AI tools for all 17 national laboratories
The Genesis mission gives every U.S. national lab access to AI co-scientist, AlphaEvolve, AlphaGenome, and WeatherNext.
Whether you're looking for an angel investor, a growth advisor, or just want to connect — I'm always open to great ideas.
Get in TouchAI, startups & growth insights. No spam.
TL;DR: The U.S. Department of Energy and Google DeepMind have partnered on the Genesis mission, a national initiative to put frontier AI tools into the hands of scientists at all 17 national laboratories. The initial rollout includes AI co-scientist on Google Cloud, with AlphaEvolve, AlphaGenome, and WeatherNext expanding access in early 2026. Twenty-four organizations, including Microsoft, NVIDIA, OpenAI, Anthropic, and xAI, have signed collaboration agreements with DOE to support the effort.
The Genesis mission is a White House initiative launched in late 2025 to use artificial intelligence to accelerate the pace of American scientific research. The stated goal is ambitious: double U.S. research productivity within a decade. The mechanism is practical: give the country's best-equipped research institutions direct access to the best available AI tools.
The Department of Energy operates 17 national laboratories across the United States. These are not small operations. They include Los Alamos National Laboratory, where the atomic bomb was developed. Lawrence Livermore, where the National Ignition Facility achieved fusion ignition in 2022. Oak Ridge, which houses the Frontier supercomputer, currently one of the fastest machines on the planet. Argonne, Sandia, Brookhaven, Pacific Northwest, and ten others.
Together, these laboratories employ over 70,000 scientists, engineers, and support staff. They work on problems ranging from nuclear weapons maintenance to climate modeling, from materials science to genomics, from fusion energy to quantum computing. They represent one of the largest concentrations of scientific talent and infrastructure anywhere in the world.
The Genesis mission connects these laboratories to frontier AI systems developed by the private sector. Google DeepMind's contribution is the most specific and detailed commitment announced so far: a phased rollout of five AI tools and platforms to all 17 labs.
"We stand at an inflection point where the convergence of advanced AI and scientific research promises to unlock a new golden age of discovery." -- Demis Hassabis, CEO of Google DeepMind
U.S. Secretary of Energy Chris Wright framed the initiative in historical terms. "Throughout history, from the Manhattan Project to the Apollo mission, our nation's brightest minds and industries have answered the call when their nation needed them," Wright said in the DOE announcement. "Under President Trump's leadership, the Genesis Mission will unleash the full power of our National Laboratories, supercomputers, and data resources to ensure that America is the global leader in artificial intelligence and to usher in a new golden era of American discovery."
Google DeepMind's commitment includes five distinct AI tools and platforms, rolled out in two phases.
Phase one, which began in December 2025, provides access to AI co-scientist on Google Cloud. This is the tool that is already live and available to researchers at all 17 laboratories.
Phase two, planned for early 2026, expands access to four additional tools: AlphaEvolve, AlphaGenome, WeatherNext, and Gemini for Government (built on Gemini 3, Google's most capable model at the time of announcement).
Here is a summary of what each tool does and where it fits in the research workflow:
| Tool | Type | Primary application | Availability |
|---|---|---|---|
| AI co-scientist | Multi-agent research collaborator | Hypothesis generation, literature synthesis | Live (Dec 2025) |
| AlphaEvolve | Coding agent for algorithm design | Materials science, drug discovery, energy | Early 2026 |
| AlphaGenome | Genomics prediction model | Non-coding DNA analysis, disease research | Early 2026 |
| WeatherNext | Weather forecasting model family | Climate modeling, cyclone prediction | Early 2026 |
| Gemini for Government | General-purpose LLM (Gemini 3) | Reasoning, multimodal analysis, research support | Early 2026 |
The distinction between these tools matters. AI co-scientist and Gemini for Government are general-purpose. They can be applied to nearly any scientific domain. AlphaEvolve, AlphaGenome, and WeatherNext are domain-specific. They are built for particular problem types and are optimized accordingly.
AI co-scientist is not a single model. It is a multi-agent system built on top of Gemini that acts as a virtual scientific collaborator. The system is designed to help researchers do three things: synthesize large volumes of existing research, generate novel hypotheses, and draft research proposals.
The way it works is closer to a research team than a chatbot. Multiple AI agents collaborate within the system, each handling different aspects of the scientific reasoning process. One agent might focus on literature review, another on identifying gaps in existing research, and a third on proposing experiments that could fill those gaps.
Google DeepMind has already demonstrated concrete results from this approach. In published case studies, AI co-scientist proposed novel drug repurposing candidates for liver fibrosis. These were not abstract suggestions. The candidates were subsequently validated through laboratory experiments. In another case, the system predicted complex antimicrobial resistance mechanisms that matched experimental results before those experiments were even published.
That last point deserves emphasis. The AI system predicted a biological mechanism that was later confirmed by human researchers working independently. It did not just summarize known science. It generated a prediction that turned out to be correct.
For the national laboratories, the most immediate application is likely literature synthesis. A scientist at Oak Ridge working on advanced materials, for example, might need to review thousands of papers on crystallography, polymer chemistry, and computational modeling before designing an experiment. AI co-scientist can compress weeks of reading into hours and surface connections between papers that a human might miss.
The system runs on Google Cloud and is trained using Google's Tensor Processing Units (TPUs). This means the national laboratories do not need to provision their own hardware for this specific tool. The compute runs in Google's infrastructure, with appropriate security controls for government use.
AlphaEvolve is a Gemini-powered coding agent designed to discover and improve algorithms. Unlike a standard code generator that writes programs from human specifications, AlphaEvolve uses an evolutionary approach. It generates candidate algorithms, evaluates them against objective criteria, and iteratively improves them across multiple generations.
The technical architecture involves orchestrating multiple LLMs in an autonomous pipeline. Each LLM proposes changes to an algorithm's code. An evaluation system scores the results. The best-performing variants survive and become the starting point for the next round of modifications. This cycle repeats until the algorithm converges on a solution that meets the specified performance criteria.
The applications for national laboratories are broad. In materials science, AlphaEvolve could design algorithms for predicting crystal structures, optimizing alloy compositions, or simulating molecular interactions. In energy research, it could tackle optimization problems in grid management, battery chemistry, or fusion reactor design. In drug discovery, it could improve the algorithms used to screen potential drug candidates against protein targets.
What makes AlphaEvolve particularly relevant for the DOE labs is that many of these institutions already have massive datasets and simulation capabilities but are bottlenecked by the algorithms they use to process that data. A better algorithm running on existing supercomputers could unlock discoveries that more compute alone cannot.
Google DeepMind has described AlphaEvolve's early results as showing "incredible promise" in computing and mathematics, with the expectation that it could be "transformative across many more areas such as material science, drug discovery and energy."
AlphaGenome addresses one of the most persistent challenges in genomics: understanding what the non-coding portions of DNA actually do.
Here is the scale of the problem. Only about 1-2% of the human genome directly codes for proteins. The remaining 98% was once dismissed as "junk DNA," but scientists now understand that much of it plays critical regulatory roles. These non-coding regions determine when genes turn on and off, how strongly they are expressed, and how they interact with each other. Mutations in these regions are linked to a wide range of diseases, from cancer to neurological disorders.
The challenge is figuring out what each non-coding region does. Traditional experimental methods are slow and expensive. You cannot simply knock out a region and see what happens, because the effects might be subtle, context-dependent, or only manifest under specific conditions.
AlphaGenome is trained directly on raw DNA sequences and predicts 11 types of biological signals that help determine how genes are used inside cells. These include whether a gene is turned on or off, where gene activity begins, how genetic messages are edited (spliced), how tightly DNA is packed (chromatin accessibility), which regulatory proteins bind to specific DNA locations, and how distant regions of the genome interact with one another.
The model's capabilities were published in Nature, which is not a venue known for publishing incremental work. The peer review process for that journal is among the most rigorous in science.
Google DeepMind released AlphaGenome's source code and model weights for noncommercial use. Since its launch, nearly 3,000 scientists from 160 countries have started using it. Research applications include cancer biology, neurodegenerative disorders, and infectious diseases.
For DOE laboratories like Brookhaven and Lawrence Berkeley, which have significant biological research programs, AlphaGenome could accelerate work on understanding radiation effects on DNA, developing biofuels, or engineering microorganisms for environmental remediation.
"AlphaGenome can help scientists better understand the non-coding part of DNA, speeding up research on genome biology and improving disease understanding." -- Google DeepMind
WeatherNext is Google DeepMind's family of weather forecasting models. The latest version, WeatherNext 2, generates forecasts 8 times faster than its predecessor and with resolution down to 1-hour intervals.
The performance numbers are stark. WeatherNext 2 surpasses the previous WeatherNext model on 99.9% of variables and lead times from 0 to 15 days. Variables include temperature, wind speed and direction, precipitation, pressure, and humidity. The model can predict the path, intensity, structure, and size of cyclones up to 15 days in advance.
The speed advantage over traditional physics-based forecasting is what matters most for operational use. WeatherNext 2 can produce hundreds of possible weather outcomes from a single starting point, with each prediction taking less than a minute on a single TPU. Equivalent physics-based simulations would take hours on a supercomputer.
This capability is powered by a new modeling approach called a Functional Generative Network (FGN). The technical innovation is that the model injects controlled noise directly into its processing, which allows it to capture the inherent uncertainty in weather prediction. It is trained on individual variable predictions but learns to forecast complex, interconnected weather systems that depend on how all the individual pieces interact.
For the DOE laboratories, WeatherNext has direct applications in climate modeling (a major research area at several labs), renewable energy forecasting (predicting solar and wind output), and national security scenarios involving extreme weather events. Laboratories like the National Renewable Energy Laboratory and Pacific Northwest National Laboratory have research programs that could immediately benefit from more accurate, faster weather prediction.
WeatherNext 2 forecast data is already accessible through Google Earth Engine and BigQuery, and through an early access program on Google Cloud's Vertex AI.
The fifth component is Gemini for Government, which brings Google's most capable general-purpose model into a government-accredited security environment.
Gemini for Government is built on Gemini 3, which Google describes as its most intelligent model with "state-of-the-art reasoning capabilities and multimodal understanding." This means it can process text, images, code, and other data types within a single system. For researchers, this translates to the ability to analyze experimental data, read and interpret scientific figures, write and debug code, and engage in complex reasoning about research problems.
The security infrastructure matters because DOE laboratories handle classified and sensitive research. You cannot run a consumer-grade AI service in a facility that works on nuclear weapons design or national security research. Gemini for Government runs on Google's AI-optimized and FedRAMP-accredited commercial cloud, which provides the compliance framework that government agencies require.
This is not unique to Google. Other cloud providers offer similar government-accredited environments. But the combination of Gemini 3's capabilities with government-grade security is what makes this relevant to the Genesis mission. It means that researchers at Los Alamos or Sandia can use frontier AI for unclassified research without leaving the approved technology ecosystem.
The practical impact is removing friction. Today, a scientist who wants to use a commercial AI tool for research might need to manually ensure that no sensitive data leaves the secure environment. With Gemini for Government deployed within the accredited infrastructure, that barrier drops significantly.
Google DeepMind is not the only organization involved in Genesis. On December 18, 2025, the DOE announced that 24 organizations had signed collaboration agreements to advance the mission.
The list reads like a directory of American technology leadership. It includes cloud providers (Amazon Web Services, Google, Microsoft, Oracle), chip manufacturers (NVIDIA, Intel, AMD), AI labs (OpenAI, Anthropic, xAI), enterprise technology companies (IBM, Dell, Hewlett Packard Enterprise, Accenture, Palantir), and others.
The breadth of participation is notable. These are companies that compete fiercely in the commercial market. AWS, Google Cloud, and Microsoft Azure fight for every government contract. NVIDIA, Intel, and AMD compete on AI chip performance. OpenAI, Anthropic, and xAI are direct rivals in the frontier model race. Yet all have signed on to support the same national initiative.
The agreements take the form of Memorandums of Understanding (MOUs). These are not binding contracts with specific deliverables. They are statements of intent to collaborate. The actual scope and terms of each organization's contribution will be determined through separate, more detailed arrangements.
What the MOUs do signal is political alignment. In a moment when AI policy is intensely debated, every major American AI company has chosen to associate itself with a government-led initiative to apply AI to scientific research. This is the least controversial application of AI imaginable. No one objects to using AI to discover better battery materials or predict cyclone paths.
For the DOE, having all major players signed on provides flexibility. If a particular laboratory needs access to a specific model or platform, the MOUs provide a framework for negotiating that access without starting from scratch.
The Genesis mission did not emerge from a vacuum. Google DeepMind has been building toward this moment since AlphaFold solved the protein structure prediction problem in 2020.
AlphaFold's impact is the most concrete evidence that AI can transform scientific research. The system predicted the three-dimensional structures of virtually all known proteins, a problem that had stumped scientists for over 50 years. The freely available AlphaFold Protein Structure Database has been used by over 3 million researchers in more than 190 countries, including over 1 million users in low- and middle-income countries.
Over 30% of AlphaFold-related research is focused on understanding disease. The system has accelerated drug discovery, materials science, and fundamental biology in ways that were difficult to predict before 2020. Demis Hassabis and John Jumper of Google DeepMind shared one half of the 2024 Nobel Prize in Chemistry for the work.
AlphaFold proved a thesis: that AI systems trained on scientific data can produce results that experienced human scientists could not achieve on their own, even given unlimited time. The protein folding problem was not going to be solved by a bigger microscope or a faster centrifuge. It required a fundamentally different approach to the problem. AI provided that approach.
The Genesis mission is an attempt to replicate that success across the entire spectrum of DOE research. The tools being deployed, AI co-scientist, AlphaEvolve, AlphaGenome, WeatherNext, and Gemini, are each targeted at different problem types. But they share the same underlying assumption that AlphaFold validated: AI can accelerate scientific discovery in ways that traditional methods cannot match.
The question is whether AlphaFold was a one-off breakthrough or the first of many. The national laboratories are being positioned to find out.
"The profound scientific and societal value of this work was recognized in 2024 with the Nobel Prize in Chemistry." -- Google DeepMind
The Genesis mission is explicitly framed as a competitiveness initiative. The United States is not the only country investing in AI for science. China has its own national AI strategy, and Chinese research institutions publish more AI papers than any other country. The European Union is investing billions in AI through its Horizon Europe program. The UK has committed to becoming an "AI science superpower."
What the U.S. has that others do not is the national laboratory system. No other country has 17 government-funded research institutions with the scale, talent, and infrastructure of the DOE labs. The Frontier supercomputer at Oak Ridge can perform over one quintillion calculations per second. The National Ignition Facility at Livermore can create conditions hotter than the center of the sun. The Spallation Neutron Source at Oak Ridge fires neutrons at materials to reveal their atomic structure.
These capabilities are useless without good software. That is, in essence, what Genesis provides. The AI tools from Google DeepMind and other participating organizations are the software layer that turns raw experimental capability into accelerated discovery.
The economic logic is straightforward. If AI tools can reduce the time from hypothesis to validated result by even 50%, the return on the billions of dollars invested in lab infrastructure multiplies accordingly. A machine that runs twice as many useful experiments per year is effectively twice as valuable, even without any hardware upgrades.
Whether that logic plays out in practice will depend on adoption. Providing access to tools is not the same as integrating them into research workflows. Scientists are, rightly, cautious about using AI systems whose outputs they cannot fully verify. The trust-building process will take time, and the Genesis mission's success will ultimately be measured not by announcements but by papers published, patents filed, and problems solved.
The Genesis mission is a White House initiative launched in late 2025 to accelerate American scientific research through artificial intelligence. It connects the U.S. Department of Energy's 17 national laboratories with frontier AI tools from the private sector. The goal is to double U.S. research productivity within a decade.
All 17 DOE national laboratories participate, including Los Alamos, Sandia, Oak Ridge, Argonne, Lawrence Livermore, Lawrence Berkeley, Brookhaven, Pacific Northwest, Idaho National Laboratory, Fermilab, SLAC, the National Renewable Energy Laboratory, and others.
Five tools in two phases. Phase one (live since December 2025): AI co-scientist on Google Cloud. Phase two (early 2026): AlphaEvolve, AlphaGenome, WeatherNext, and Gemini for Government (Gemini 3).
It is a multi-agent system that synthesizes scientific literature, generates novel hypotheses, and drafts research proposals. In validated case studies, it proposed drug repurposing candidates for liver fibrosis that were confirmed through lab experiments, and predicted antimicrobial resistance mechanisms before they were experimentally discovered.
AlphaEvolve is a Gemini-powered coding agent that designs and improves algorithms using an evolutionary approach. It generates candidate algorithms, evaluates them, and iteratively improves the best performers. Applications include materials science, drug discovery, energy optimization, and computational mathematics.
WeatherNext 2 outperforms its predecessor on 99.9% of weather variables and lead times from 0 to 15 days. It can generate hundreds of forecasts from a single starting point in under a minute per prediction on a single TPU. It can predict cyclone paths up to 15 days out.
They include Amazon Web Services, Google, Microsoft, Oracle, NVIDIA, Intel, AMD, OpenAI, Anthropic, xAI, IBM, Dell, Hewlett Packard Enterprise, Accenture, Palantir, and others. The agreements are Memorandums of Understanding signaling intent to collaborate.
AlphaFold is not directly part of the Genesis mission, but its success is the proof point that made Genesis possible. AlphaFold solved the protein folding problem, won the 2024 Nobel Prize in Chemistry, and has been used by over 3 million researchers in 190+ countries. It demonstrated that AI can produce scientific breakthroughs that traditional methods cannot.
The AI tools are provided through collaboration agreements with the private sector, not purchased with taxpayer dollars through traditional procurement. The DOE laboratories already receive public funding for research operations. Genesis adds AI capabilities on top of existing infrastructure, with the expectation of accelerating the return on that public investment.
There is no specific timeline. Scientific research, even AI-accelerated research, takes time to design, execute, peer-review, and publish. The earliest results from AI co-scientist usage at the labs could appear in preprints within months, but validated, peer-reviewed publications typically take 12 to 18 months from experiment to print.
Google's March 2026 Pixel Feature Drop expands Circle to Search globally, adds Gemini real-time visual understanding, AR overlays, and shopping integration to Pixel 7a and above.
Gemini 3.1 Pro scores 69.2 percent on the MCP Atlas benchmark, leading Claude and GPT-5.2 by 10 points with adjustable reasoning depth on demand.
Boston Dynamics partners with Google DeepMind to integrate frontier AI into the next-generation Atlas humanoid robot, accelerating the physical AI revolution.