Information Systems & Technology (IS&T) and the Office of Research Computing and Data (ORCD) are actively engaged in several projects to support generative AI capabilities that will benefit research at MIT. These projects explore emerging and optimal methods to meet the AI needs of researchers across campus, including in areas ranging from AI safety studies to the integration of AI in teaching and in basic research and beyond.
One major project, led by the ORCD team and supported by more than $15 million in external philanthropic funding, involves the deployment of a unified, cross-MIT platform dedicated to AI research and related activities. This initiative will provide MIT researchers with access to a regularly updated and robust infrastructure that integrates GPU hardware, standardized software, open training data and models, and professional support staff. The initiative also includes seed funding for research projects that contribute to the overall design and evolution of the system. To support this project, ORCD and IS&T are currently configuring 500KW of new power and cooling infrastructure at the MGHPCC zero CO2 data center. This capacity will accommodate ORCD hardware, as well as hardware procured by principal investigators who wish to participate.
Additionally, ORCD and IS&T are working with commercial providers to incorporate subsidized cloud resources within the cross-MIT platform. MIT researchers currently have access to Office 365 Copilot, and a suite of OpenAI tools will be available in the fall. An ongoing experiment with AWS provides subsidized access to the AWS Research and Engineering Studio, including Amazon Trainium hardware. Plans are also in place for a fall experiment that will offer subsidized access to Lambda Labs One Click Cluster technology.
As the generative AI hardware and software infrastructure landscape evolves in the coming months and years, ORCD and IS&T will introduce and develop more services, and the teams welcome suggestions and ideas for ways they can further support MIT research. Send your suggestions and ideas to orcd-help@mit.edu.