Allen Institute for AI launches big computing cluster for $152M project backed by Nvidia and NSF

The Allen Institute for AI says it has brought online and started using a powerful new computing system funded by Nvidia and the National Science Foundation, the first big milestone in a $152 million project to build open AI models for scientific research.

Ai2, as the Seattle-based institute is known, was awarded the funding last August as part of the White House AI Action Plan. The project, called Open Multimodal AI Infrastructure for Science, or OMAI, aims to build AI models for fields such as materials science, biology, and energy.

Noah Smith, Ai2 senior research director and principal investigator on the project, called it a “critical step” and said in a statement that the new infrastructure represents a national investment in keeping advanced AI development accessible to the broader research community.

The announcement Thursday comes as Ai2 works to regain its footing after losing its CEO and some of its top researchers to Microsoft in March. Interim CEO Peter Clark outlined Ai2’s priorities this week, saying it’s committed to open models and longer-term research, along with applied AI efforts in areas such as scientific discovery and environmental science. 

Unlike most large-scale AI projects, Ai2 releases the full code, data, and training methods behind its models, allowing other researchers to reproduce and build on the work. 

The new system, located outside of Austin, runs on Nvidia’s Blackwell Ultra chips and is managed by Cirrascale Cloud Services.

Ai2 said research supported by the project has already produced upgrades to its Molmo and OLMo model families, including a new multimodal model capable of video understanding and a more efficient language model architecture. 

The institute said it is now focused on building unified models that handle multiple types of data, developing AI agents, and working more closely with scientific communities to ensure the models are useful for real-world research. 

Source link