The challenge

Artificial intelligence depends on increasingly large computing systems. Improving the efficiency of algorithms and datacenter operations can reduce the resources required to train and operate those systems.

The work

Carbon Connect brings together research on efficient algorithms and sustainable datacenter operations. The initiative is supported by a U.S. National Science Foundation grant and is conducted in collaboration with the University of Pennsylvania.

Why it matters

Sustainable AI requires progress across the full computing stack. Research that connects algorithmic efficiency with infrastructure operations can help future systems deliver more capability with fewer resources.

Timeline

Completed milestones are highlighted; muted entries show the work still ahead.

  1. May 2024
    Complete

    NSF Expedition announced

    The National Science Foundation selected Carbon Connect as a five-year, multi-institutional Expeditions in Computing initiative.

    Read the NSF announcement (opens in a new tab)
  2. May 2024
    Complete

    Research agenda published

    The project white paper established a shared agenda spanning carbon accounting, system design, power management, hardware life cycles, and policy.

    Read the white paper (opens in a new tab)
  3. Next phase
    Future work

    Build transparent carbon accounting methods

    Develop and validate lifecycle methods that connect embodied and operational carbon across chips, systems, and datacenters.

  4. Five-year program
    Future work

    Translate research into sustainable systems

    Advance carbon-efficient design, intelligent power management, and hardware reuse while building a broader technical community.

People involved

Project leads, contributors, and collaborators named in the public project record.