Skip to content 
The Allen Institute for AI (Ai2) in Seattle has been chosen to spearhead a national effort to develop open artificial intelligence models for scientific research, backed by $152 million in funding and infrastructure from the U.S. National Science Foundation (NSF) and Nvidia. The five-year initiative will establish a new technological backbone for AI-driven science, marking a pivotal expansion of Ai2’s role in national research efforts.
Federal and Corporate Backing Marks Milestone for Ai2
The NSF will fund the project with $75 million and Nvidia will donate $77 million with hardware and software. According to Ai2 Chief Executive Ali Farhadi, the investment would be a great step forward in enhancing the installment capacity of AI matters and giving research abilities to the country. The project dubbed the Open Multimodal AI Infrastructure to Accelerate Science (OMAI) will develop a family of large multimodal AI models that are fully open.
Such systems will be trained on the scientific literature and datasets to speed-up breakthroughs in fields like materials science, biology and energy. Ai2 senior director of natural language processing research Noah Smith, a professor of computer science with the University of Washington (UW), will manage the project. According to NSF special assistant working on artificial intelligence, Tess deBlanc-Knowles, the program would make the U.S. be on the frontline on integrating AI in scientific discovery. The initiative is an element of the White House AI Action Plan to enhance the American standing in AI-based research.
Read also:U.S. Leads the Global AI Race
Creating full open AI models to support science
Ai2 has made a promise to publish the AI models of OMAI in full transparency. The models will comprise complete weight, training, code, and evaluation mappings thus enabling researchers to examine, customize, and re-train the models to their requirements. Farhadi argued that this would give confidence to scientists not only in trust, but also in its reproducibility in high-stakes work. The latter project is based on the experience of Ai2 with open AI technologies.
The institute already created high-performance open models, OLMo and Molmo and open data, Dolma. The different resources have found immensity in the research circles with scientific and language based applications. In comparison, most part of notable AI systems by privately owned companies are either proprietary or partially open. The open-source strategy taken by Ai2 aims at eliminating obstacles to collaboration and enabling more rapid advancement in the many fields of study.
Nvidia Hardware to Power Advanced Model Training
Nvidia will supply its HGX B300 systems, built on the new Blackwell Ultra architecture, alongside AI Enterprise software to optimize both training and inference processes. Nvidia CEO Jensen Huang said the collaboration aimed to make intelligence a renewable resource for America.Jack Wells, Nvidia’s director of higher education and research, said the hardware is engineered for massive datasets, enabling faster and more efficient processing to speed scientific discovery. The systems will support the complex workflows required to train large multimodal AI models tailored for scientific applications.The funding will primarily be allocated to computing resources, allowing Ai2 to scale up from its previous work and develop larger, more advanced AI systems on open foundations.
Collaboration Across U.S. Research Institutions
A number of universities will also take part in the OMAI project besides Ai2. The University of Washington will also be involved along with Hanna Hajishirzi who belongs to Ai2 as a leader. The University of Hawai I at Hilo team will have Travis Mandel as a team leader, the University of New Hampshire Samuel Carton and the University of New Mexico Sarah Dreier. The interaction of these research partners will concern not only the building of the models, data preparation, and integration of AI systems into the workflow of science.
According to Ai2, the models would contribute to assisting researchers to visualize and process data, write analytical code and identify patterns invisible to human eyes. The systems might also be used to interlink the results of various subjects of study in the sciences to come up with new findings that are interdisciplinary. Ai2 hopes to have the first significant OMAI model after about 18 months with datasets, code, and other resources being made available on an ongoing basis over the course of the project.
Positioning Washington State as a Science and AI Hub
Established in 2014 by Microsoft co-founder Paul Allen, Ai2 has established a solid reputation in research circles although it has frequently stood in the shadow of more technology-prominent organizations such as OpenAI, Anthropic, Google DeepMind, and Meta. The institute has been developing research tools, scientific search engines and open-source AI resources.
According to Washington Senator Maria Cantwell, the OMAI project will make the state a major force in the scientific innovation using artificial intelligence. She cited how both Ai2 and the University of Washington have the collective talent that advances medicine, clean energy and materials science. The effort of OMAI by Ai2 will extend its track record of providing freely accessible AI models and data and support its belief in the importance of transparency and joint research.
