Important results from the project
Yes, the goals were met by showing that different hardware architectures can interoperate seamlessly. The project delivered NextGen Framework, an open, installable Python package for asynchronous, decentralized AI learning. Three different accelerators (Intel, AMD, NVIDIA) were integrated. The project yielded 4 open codebases, 2 curated datasets and a validated Proof of Value at AstraZeneca where search performance for drug information was improved.
Expected long term effects
In the long term, the results strengthen Swedish "Sovereign AI" and on-premise operations. The TCO analysis shows up to 18x cost advantage per million tokens compared to cloud APIs, motivating local AI investments. The framework provides highly regulated sectors with a secure method to collaborate on AI development without moving sensitive raw data, operationalizing strict eligibility requirements from EU AI Act, GDPR and DORA.
Approach and implementation
Coordinated by AI Sweden, infrastructure and product companies were brought together. An asynchronous, pull-based architecture was built around a state-driven model catalog in GitLab. Data sharing barriers arose; one company switched to an advisory role when internal data was not released. The practical verification was focused entirely on AstraZeneca´s medical case and was successfully implemented across a network of distributed and heterogeneous hardware.
External links
The project description has been provided by the project members themselves and the text has not been looked at by our editors.
