Rethinking Datacenter Refresh
We optimize when datacenter components are replaced to reduce cost and carbon emissions.
Our Work
Our Projects
Original research, open infrastructure, and aligned projects we support, all working toward AI that can scale sustainably.
AlgoPerf
We benchmark neural-network training algorithms to identify faster, more efficient ways to train models.
LLMargin
We are improving energy accounting for language-model serving by connecting GPU measurements to individual requests.
Carbon Connect
We develop efficient algorithms and datacenter practices to reduce the environmental impact of computing.
Parallex
We connect scientific workloads with idle computers to make research computing more accessible.
Have a Project in Mind?
Bring us a research question, an infrastructure challenge, or an ambitious project where open engineering can make a difference.