NVIDIA's cloud-based AI infrastructure contributions to the NSF's NAIRR pilot have enabled over 700 research projects sp

The National Science Foundation's National Artificial Intelligence Research Resource pilot program has driven innovative research across the United States for over 700 projects during the past two years, spanning protein prediction and infectious disease outbreak management. NVIDIA contributed to this initiative through a cloud-based resource providing researchers dedicated access to a minimum of four NVIDIA DGX nodes for at least a month, along with technical support for onboarding and project assistance. With NVIDIA's AI infrastructure support and DGX reference architecture delivering dedicated resources, researchers have collapsed workflow timelines and uncovered groundbreaking technologies reshaping healthcare, agriculture and energy industries.

Polymathic AI, a coalition of international scientists from the Flatiron Institute, Cambridge University and Lawrence Berkeley National Lab, leveraged NVIDIA GPUs and NVLink interconnect technology to strengthen physical fluidlike simulations through a large-scale dataset called the Well. The team developed Walrus, a foundation model for fluidlike behavior, made publicly available along with its data, code and pretrained weights. This approach addresses current limitations in scale and pretraining diversity while exploring scaling laws to accelerate development of more powerful foundation models for scientific applications.

Researchers at the University of Michigan, led by Professor Venkat Viswanathan in the Department of Aerospace engineering, developed a model-fusion framework combining domain-specific molecular AI with general-purpose large language models. The family of molecular foundation models, MIST or Molecular Insight SMILES Transformers, is designed for discovery across chemical space. MIST models were pretrained on large unlabeled molecular datasets using a novel tokenizer called Smirk to better capture nuclear, electronic, geometric, isotopic and stereochemical information. Developed on a 40-GPU NVIDIA DGX cluster plus 200,000 additional NVIDIA GPU hours on ALCF's Polaris cluster, MIST has been fine-tuned on more than 400 structure-property relationships and matches or exceeds state-of-the-art performance across electrochemistry, quantum chemistry and physiology benchmarks. Fusing MIST with general-purpose LLMs makes quantum-chemical calculations more broadly accessible and accelerates energy storage and conversion system design.

Boston University's Hariri Institute for Computing and Center on Emerging Infectious Diseases trained a large language model using NVIDIA accelerated compute to support BEACON, the Biothreats Emergence, Analysis and Communications Network. This LLM, trained on a large corpus of infectious disease documents and epidemic-prone pathogen information, analyzes online posts of emerging disease outbreaks on a global scale. BEACON integrates signals from the global disease-tracking platform HealthMap, news and social media feeds, subject-matter experts and community communications to generate outbreak reports. Ioannis Paschalidis, director of Boston University's Hariri Institute, noted that the pipeline reduced outbreak report composition from several hours to roughly two minutes. Internationally deployed doctors, government organizations and academic researchers already use the BEACON model to quickly identify and treat infectious diseases.