Governments in the United States and the European Union are mobilizing large-scale initiatives to integrate artificial intelligence into scientific discovery, yet the two regions are taking distinct routes to the same goal: accelerate research by marrying AI methods with high-performance computing.
Different architectures, shared ambition
At an ISC 2026 session in Hamburg, technologists outlined how political and institutional structures are shaping national approaches. In the United States, the Department of Energy has orchestrated a broad effort — known as the Genesis Mission — that assigns components of the programme to different national laboratories and builds dedicated compute capacity to run those workloads.
Oak Ridge National Laboratory, for example, is preparing a new supercomputer named Lux to support Genesis Mission tasks and is responsible for core infrastructure under an American Science Cloud (AmSC) concept. Argonne National Laboratory has been tasked with advancing AI model development through the Transformational AI Models Consortium (ModCo). The U.S. approach is framed as a "whole of government" effort that explicitly links federal labs with academia, industry and hyperscalers to speed discovery workflows.
"We are trying a whole of government [approach], as well as with academia and industry partners, to unify all of our research capabilities, from experimental facilities to compute facilities, integrating with hyperscalers so that we can automate discovery," said Scott Atchley, chief technology officer at Oak Ridge National Laboratory.
Europe’s federated model
Speakers representing European initiatives described a different emphasis: building federated platforms such as those coordinated by the EuroHPC Federation Platform and national partners like Finland’s CSC-IT for Science. European efforts focus on interoperable infrastructure to store, share and operate AI/ML models and large vector datasets at scale.
- U.S. model: central coordination through DOE, dedicated supercomputers (e.g., Lux), lab-assigned responsibilities, and industry/hyperscaler integration.
- EU model: federation of compute and data platforms, shared standards and tools to distribute models and vector data across borders.
- Common need: both regions emphasise that additional funding is essential to realise AI-for-science aims.
Implications and constraints
The differing architectures carry operational consequences. A centrally directed, lab-focused U.S. programme can concentrate resources quickly on priority problems and provision bespoke facilities. A federated European network aims to maximise reuse of existing systems, interoperability and cross-border collaboration, potentially easing access for a broader set of researchers.
| Feature | United States | European Union |
|---|---|---|
| Coordination | DOE-led, national labs | Federated via EuroHPC and national centres |
| Infrastructure | Dedicated supercomputers (e.g., Lux) | Shared platforms for models and vector data |
| Partnerships | Academia, industry, hyperscalers | National centres, pan-European collaboration |
Speakers at the conference underscored that despite the strategic and organisational differences, both regions are wrestling with the same practical challenge: fiscal constraints. Achieving the promised acceleration of discovery through AI will depend on whether governments are prepared to sustain and expand investment in compute resources, data infrastructure and collaborative programmes.
As governments design systems to pair experimental facilities with large-scale compute and AI models, the question for policymakers will be how to balance concentrated capability with broad accessibility — and how to fund both the hardware and the human expertise required to convert advanced systems into scientific breakthroughs.
