Nine days before one of its first operational deadlines, the United States’ Genesis Mission is moving from a declaration of intent towards a large-scale scientific, political and economic infrastructure. Yet the most revealing development this week is not another supercomputer or investment figure. It is a call for contributions.

On 7 August, the US Department of Energy (DOE) launched the Genesis Open Models Initiative. Its first model, Genesis-Science-1, is being developed with US company Arcee and is intended to become an open-weight model specialised in scientific research. DOE is inviting universities, laboratories, companies, scientific organisations and research teams to contribute data, code, working environments, evaluations, reinforcement-learning tasks, rubrics and expert knowledge. The first window for pretraining contributions closes on 14 August. The post-training window closes on 25 August.

The scene is revealing. A national mission created to accelerate science through artificial intelligence now needs the scientific community itself to teach the machine what it should learn, how a scientific task should be recognised and what counts as an acceptable result.

That is where the question moves beyond engineering.

From promise to infrastructure

Genesis Mission was established by Executive Order 14363, signed on 24 November 2025. Its stated aim is to build an integrated platform connecting supercomputing, experimental facilities, scientific datasets and AI systems in order to accelerate US research. The administration has set itself the goal of doubling the productivity and impact of American science and engineering within a decade.

The order also established a series of deadlines. One falls on 21 August 2026. By then, the Secretary of Energy is directed to seek to demonstrate, subject to applicable law and available appropriations, an initial operating capability of the American Science and Security Platform for at least one national science and technology challenge.

The wording matters. The order does not guarantee that such a capability will exist by that date. It requires DOE to seek to demonstrate it.

At the head of this reorganisation is Darío Gil, a Spanish engineer born in Murcia and raised in Madrid, a former director of IBM Research and the current Under Secretary for Science at the Department of Energy. DOE also identifies him as Director of the Genesis Mission. His position between corporate research and the federal scientific apparatus is particularly significant for a project attempting to assemble government, national laboratories, universities and major technology companies.

In his public letter to the scientific community, Gil presents science, engineering and technology as a currency of strategic power. He describes the present moment as a competition the United States must win and places Genesis within a national undertaking historically comparable to the Manhattan Project and Apollo. This is not a neutral description of the international context, but the political frame Gil himself uses to explain the mission.

As that point approaches, Genesis has expanded institutionally. On 22 July, the White House announced more than $5 billion in federal commitments linked to the mission and said that more than fifteen federal agencies would contribute funding, facilities, datasets or research programmes. On the same day, DOE reported more than $800 million in commitments from members of the Genesis Mission Consortium, including compute resources, access to models, cloud infrastructure, scientific expertise, research partnerships and direct funding.

DOE also announced 278 projects selected following what it described as the largest response to a funding opportunity in the department’s history. Of those selections, 168 are led by universities, 87 by DOE and National Nuclear Security Administration national laboratories, 19 by companies and four by non-profit organisations. Altogether, 342 institutions are participating.

There is still, however, an important distinction between selection and actual funding. DOE itself states that these selections initiate award negotiations and do not in themselves constitute a commitment to provide funding. That gap between announcement, negotiation and execution is one of the issues worth watching.

The international dimension is already beginning to organise itself around this platform. On 4 June, DOE announced Japan as Genesis Mission’s first international partner through a planned $1 billion scientific partnership over five years, $500 million from each country, subject to future appropriations. The agreement brings together US national laboratories and Japanese institutions, and Japan cannot simply be described as a subordinated actor. It does, however, raise a question about the asymmetry of the architecture: how are other countries’ capabilities integrated into an infrastructure whose institutional centre, declared objectives and criteria of success remain defined from the United States?

A new form of Big Science

Genesis can be read as an updated form of Big Science. Not simply because it is a very large scientific project, but because it shifts the unit through which knowledge is produced.

The familiar image of a relatively autonomous laboratory gives way to an assemblage of federal agencies, universities, national laboratories, technology companies, supercomputers, data infrastructures, foundation models, experimental facilities and industrial supply chains.

The American Science and Security Platform is intended to connect precisely these elements. AI appears here not merely as a tool used by researchers but as a layer designed to coordinate data, simulations, instruments and distributed workflows.

That changes the question of who does science.

If a result depends simultaneously on public datasets accumulated over decades, models developed by private companies, federal computing capacity, domain experts designing evaluations and agents carrying out parts of the experimental process, the attribution of authorship, responsibility and merit becomes less straightforward.

This is not a concern being imposed from outside the project. Science: A New Golden Age, published by the White House Office of Science and Technology Policy on 21 July, argues that today’s funding structures, publication systems and credit mechanisms were built for a world of human-paced discovery. Among its proposals is experimentation with “AI-native” scientific institutions, faster and more open forms of publication and more granular ways of attributing contributions.

Technological acceleration is therefore accompanied by the possibility of reorganising scientific careers themselves.

Before training the model, science has to be made legible

Genesis-Science-1 makes that process visible at a much more concrete level.

The call does not ask only for polished scientific papers. It seeks scientific text, code, documentation, structured collections, expert demonstrations, research software, complete workflow environments, reinforcement-learning tasks, held-out evaluations, scoring rubrics, tests and verifiers.

It is a significant list.

For a model to participate in research, part of scientific practice must first be transformed into objects that the system can process. Procedures, criteria, mistakes, decisions and forms of validation need to become explicit enough to function as training or evaluation material.

That requires choices.

Which parts of scientific practice can become data?

Which errors are worth preserving because they teach something?

Which procedure adequately represents a discipline?

Who decides that an evaluation measures a scientifically meaningful capability?

And what happens to tacit, situated or difficult-to-formalise knowledge?

The contribution programme indirectly acknowledges this problem by expressing particular interest in workflows that are poorly represented by standard AI evaluations and require direct collaboration with the people who actually perform the work.

The question, then, is not simply how much scientific knowledge a model can ingest. It is also what transformations that knowledge must undergo before becoming legible to the machine.

There is also a political distinction that needs to remain visible. Genesis-Science-1 may be built through broad contributions and may eventually release its weights, but that does not resolve who governs the infrastructure selecting those contributions, allocating compute, defining evaluation priorities, establishing institutional partnerships and setting the programme’s objectives. Open weights do not necessarily mean open governance.

Who contributes and who receives credit

The contribution process adds another layer. Participating organisations must describe what materials they hold, which experts can support them and the terms under which those materials could be used.

Applications pass through a review process covering scientific fit, rights and handling, expert and evaluation readiness, technical integration and final programme selection. Selected contributors will receive early evaluation access and credit in the technical report and release materials.

An emerging economy of scientific credit is visible here.

Data, working environments, evaluations and even useful failures can become contributions to a shared infrastructure. Yet institutions do not possess equal repositories, equal capacity to document them or equal access to the networks through which decisions are made about what enters the model.

The apparently open category of the “scientific community” contains very different positions.

The question is therefore not only who will be able to use Genesis-Science-1 once it exists. It is also who can participate in defining what the model will recognise as science.

Science, sovereignty and power

Genesis explicitly links scientific discovery to national security, energy policy, industrial competitiveness and US technological autonomy. One of its priority areas is the critical-minerals supply chain, presented as a problem of reducing dependencies and building a self-reliant foundation for the industries of the future.

The programme proposes using AI to integrate geophysical information, materials science, process optimisation and economic modelling in order to locate resources, develop alternatives and recover scarce elements from waste.

There is an interesting continuity between that operation and the construction of scientific models themselves.

In one case, industrial waste and mineral deposits are reconsidered as strategic reserves. In the other, decades of data, simulations, code, laboratory records and expert knowledge are reorganised as resources for training AI systems.

The materials are very different, but in both cases the mission works through a politics of identifying, extracting, standardising and circulating resources considered strategically valuable.

AI infrastructure does not remove the materiality of science. It multiplies it.

It needs minerals, energy, data centres, instruments, skilled people, repositories, usage rights and accumulated knowledge. The promise of acceleration rests on that network of conditions.

Genesis can therefore be understood not only as scientific infrastructure but also as infrastructure for geopolitical power. Its importance lies in the capacity to concentrate compute, scientific datasets, models, experimental facilities, standards, supply chains and institutional alliances within a single architecture. That concentration can create centres and peripheries of scientific capacity even without direct territorial control.

This is an anthropological and political interpretation of the architecture documented by the programme, not a label used by DOE. It does not mean that every international partnership is domination, nor that cooperation with Japan has only one meaning. It does require us to ask who owns and governs the infrastructure, who sets its priorities, who can define the scientific problems, who has the compute required to participate and what autonomy remains for institutions contributing knowledge.

The test will not be purely technological

DOE estimates that meeting the goals of Genesis will require training around 100,000 people working in science and engineering over the next decade with dual competence in AI and a scientific or engineering discipline.

This offers another way of understanding the mission. Genesis is not simply introducing a new tool into existing laboratories. It is attempting to alter infrastructure, training, professional careers, credit systems and relations between government, academia and industry so that a particular form of AI-assisted science can operate at national scale.

On 21 August we should know more about the platform’s initial operating capability.

But even if DOE succeeds in demonstrating it, the more important question will remain open.

An infrastructure can accelerate calculations, automate experiments and connect repositories. Deciding which questions are worth asking, what counts as evidence, which forms of knowledge can be formalised, who participates in those decisions and how recognition is distributed remains a social question.

Genesis Mission presents itself as a project for teaching machines to do science.

Perhaps we are also watching the reverse process: the reorganisation of science so that it can work with machines.

The United States is not only trying to teach machines to do science. It is building a national infrastructure from which decisions can be made about which science is accelerated, which knowledge becomes computable and who controls the means required to produce it.

When the infrastructure for doing science also becomes an infrastructure of power, who decides which forms of knowledge deserve to be accelerated — and who will have the capacity to do so?

References