1 of 23

Overview of NSF’s Office of Advanced Cyberinfrastructure

Katie Antypas

Director, Office of Advanced Cyberinfrastructure

​

Confab25

April 9, 2025

1

2 of 23

NSF Office of Advanced Cyberinfrastructure (OAC)

  • Advanced Computing. Provide state-of-the-art advanced computing and services for the broad research and education community.
  • Software and Data Infrastructure. Support development and deployment of shared data and software resources, tools, and services. Support scientific communities to use and share data.
  • Networking and Cybersecurity. Invest in campus and regional connectivity and communities to enable advanced computational science activities. Support the transition of cybersecurity research to practice.
  • Learning and Workforce Development. Foster a national research workforce for creating, operating and using advanced computing, data and networking infrastructure for discovery and innovation.
  • Strategic Investments. Lead and engage on special opportunities, cross-cutting agency and national initiatives

2

Transforming science and engineering research and education�through an integrated cyberinfrastructure ecosystem

Katie Antypas

Office Director

Amy Walton

Deputy Office Director

  • Over 20,000 researchers and students supported by OAC investments
  • In FY24 440 awards, investing ~$245M

3 of 23

OAC’s unique role with partners across NSF Directorates to advance research and education

3

Cyberinfrastruture for all S&E Domains

.

  • OAC supports all NSF Directorates
  • Provides coordinated access to infrastructure
  • Participates in NITRD and other national priority task forces and subcommittees

4 of 23

Challenges:�Significant ecosystem disruption across many fronts

4

Nyx simulation of Lyman alpha forest

Data

Growth of data from sensors, detectors and instruments creates new user requirements

Post-Moore

Specialized hardware and accelerators

AI

Rapid advance and integration of AI into scientific applications and workflows

Business Models

New business models and entrants for computing and data infrastructure

Scaling out

Cross-agency initiatives, global competition, and need to broaden access to resources

5 of 23

Opportunities and Strategic Priorities

5

Enabling discovery through integrations of data, software and the infrastructure ecosystem

Growing and developing communities, workforce and partnerships

Investments in new technology adoption and scaling advanced infrastructure

Discovery through Integration

Building Workforce, Communities and Partnerships

Advancing Infrastructure

Data

Post-Moore

AI

Business Models

Scaling out

Challenges

6 of 23

Discovery through Integration

6

Integrates and coordinates access to traditional HPC

Adds integration to private sector resources, AI models, services, collaborations, datasets

Guidance, support, expertise to build community gateways leveraging ACCESS and other resources

Data caching and sharing service

Data discovery services

Integration services for data

Integration of national scale resources

Integration services and aggregation of campus and regional computing

7 of 23

Building Workforce, Communities and Partnerships

7

Cybersecurity framework

Facility data lifecycle

Building a Computational and Data-Intensive Research Workforce & Network in the Mid-Atlantic Region

CyberMAGICS Workshop: 2118061

8 of 23

Advancing Infrastructure

8

GPU and FPGA support for the Galaxy framework

National Deep Inference Fabric

A transparent research computing fabric to allow scientists to probe and alter internals of trillion-parameter AI models:

​

9 of 23

NAIRR Pilot ��

9

Building new partnerships and communities

Providing novel infrastructure and services

Integrating cyberinfrastructure

10 of 23

National AI Research Resource (NAIRR) - A vision to drive US AI innovation, discovery, and competitiveness

10

Envisioned NAIRR Architecture

National goals

  • Accelerate AI and AI-powered discovery and innovation.
  • Expand the US AI R&D workforce and train the next generation of AI researchers and educators.
  • Increase integration and use of world-class public and commercial AI resources.
  • Advance public trust in AI.

Computing and software

User support and expertise

Datasets and Models

Educational materials and training tools

Central Portal

About NAIRR & NAIRR Pilot 

NAIRR Task Force Report

nairrpilot.org

NAIRR Pilot

11 of 23

  • AI2: Allen Institute for AI
  • AMD
  • Amazon Web Services
  • Anthropic
  • Cerebras
  • Databricks
  • Datavant
  • EleutherAI
  • Google
  • Groq
  • Hewlett Packard Enterprise
  • Hugging Face
  • IBM

​

​

  • Intel
  • Meta
  • Microsoft
  • MLCommons
  • NVIDIA
  • Omidyar Networks
  • OpenAI
  • OpenMined
  • Palantir
  • Regenstrief Institute
  • SambaNova Systems
  • Vocareum
  • Weights & Biases

​

​

26 industry and non-profit partners are contributing in kind

state-of-the-art resources

Research Resources

Computing

Datasets

Educational/training opportunities

Models, software, platforms

Collaborations

DARPA

DOD

DOE

DoEd

FDA

NASA

NIH

NIST

NOAA

NSF

USDA

USGS

USPTO

VA

NAIRR Pilot

12 of 23

Pilot Year 1: We built the key foundations needed for a full NAIRR

12

Expand AI research & education pathways 

Portal deployment, request process, on boarding, resource matching, support, training

Establish strong govt-industry partnership

A new multi-lateral model based on voluntary in-kind contributions to bring the most advanced resources to US researchers

14 agencies

26 industry and non-profit partners

Portal

Establish single access point for US users

Integrate & deliver contributed resources

Back-end operations to exploit and track all contributed resources: compute, data, models, expertise

80% of partner resources are engaged and being utilized

600+ requests to date

350 approved projects underway in 42 states

Technical Demo Projects, NAIRR Secure, NAIRR Classroom, community workshops & outreach, data set contributions

Initiated over 20 funded projects and pilots �--> Focus of Y2

Results

NAIRR Pilot

13 of 23

Training the next generation of AI researchers and leaders across the country

13

300+ Research projects supported across 43 states

Researchers must be US based. Researchers come with funding support from 11 different agencies including NSF, NIH, NOAA, DOE, DOD, DARPA, ONR, ARO, NASA, USDA and VA as well as foundation and non-profit funding.

NAIRR Pilot Allocations by Science Category

(A100 GPU hours equiv)

NAIRR Pilot

14 of 23

Research community requires resources of many scales

14

As of 3/31/25

Largest computing awards in collaboration with industry and DOE

Other projects: NAIRR classroom, model access, novel architecture hardware

(GPU Hour bins)

Large

Small

​

#of Projects

Agency Supported

266

Industry Supported

172

Note: Some projects are granted access to multiple resources

NAIRR Pilot

15 of 23

Some NAIRR Pilot project highlights

15

Effective Backdoor Mitigation Depends on the Pre-training Objective PI: Blimes, U of Washington

Training Vision Language Models for Agricultural Resilience

PI: Ganapathysubramanian, Iowa St.

Investigations of compressed language models for accuracy and robustness

PI: Srikumar, U. Utah

Rapid Assessment of Wildland Fire Position and Plume Dynamics using Coordinated Multi-Unmanned Aircraft System Sensing

PI: Scherer, Carnegie Mellon University

NAIRR Pilot

16 of 23

Results of early Request for Information: Researcher challenges and barriers

  • Access to, lack of large datasets for a particular domain can limit development and validation of AI models
  • Data preparation, cleaning, wrangling is a significant effort
  • Large scale non-public/sensitive datasets while ensuring privacy and compliance
  • Data storage, movement and networking

​

​

16

NAIRR Pilot

17 of 23

NAIRR Pilot Dataset Opportunity and �Data Infrastructure Investments

  • AI USE CASES
  • USER COMMUNITY
  • METADATA AND DOCUMENTATION
  • USER SUPPORT AND TRAINING
  • DATA POLICY

17

National Data Platform

PI. Ilkay Altintas

Advancing the Open Science Data Federation Platform

PI. Bockelman

NSF 25-018

Submissions must discuss:

Under review

NAIRR Pilot

18 of 23

Building a NAIRR community to share best practices, provide critical feedback, and onboard new students

18

We supported an inaugural NAIRR Pilot convening in February:

    • Connect and exchange ideas to enhance ability of Pilot to drive innovative AI & science research and reach new and emerging AI research and education communities
    • Build collaborations and partnerships
    • Gather lessons learned, challenges and gaps for improving the pilot and planning for a full scale NAIRR.

​

NAIRR Pilot

19 of 23

Leadership Class Computing Facility

19

20 of 23

20

Leadership Class Computing Facility Awarded

Distributed Science Centers

Leadership Capability

Partners

The LCCF, led by the Texas Advanced Computing Center (TACC)

Will support largest supercomputing capability at an academic institution

    • 5 Distributed Sites
    • 27 academic partners

Current Status

    • July 2024 award with expected operational deployment in FY 2027
    • Phase 1 Vista system

LCCF provides the computational and data analytics ecosystem that provisions large-scale capabilities for S&E research to enable discoveries that would not be possible otherwise.

Total Facility Funding: ~$457M

21 of 23

Questions

21

22 of 23

Backup

22

23 of 23

NAIRR Pilot Insights and Lessons From Year 1

23

The public-private partnership model is VERY advantageous.

  • Delivers a much richer array of resources and expertise compared to a public-only model.

High demand for the array of NAIRR Pilot AI resources.

  • Many research and education topics, use cases, and project sizes
  • Ways of accessing compute may need to be faster and more interactive; consortia may be needed for the largest scales
  • Demand for secure AI resources for sensitive data and uses cases.

Intensive coordination is key to connect partners, resources, and users

  • Constant contact, clear channels, and nimbleness have been key
  • Significant appetite for workshops and training on using resources.

Access to trusted data and their integration are central challenges

    • Needs are very varied across projects and disciplines
    • AI data infrastructure capabilities must be co-designed with communities and with specific use cases in mind.

Important to measure performance, impact, and success along the way.

    • A full NAIRR will need both near-term and longitudinal assessments
    • Must identify and relieve pain points for users across NAIRR steps.

NAIRR Pilot