Making GPUs abundant

Every scarce resource in computing becomes abundant. CPUs, storage, networking, and memory all made this transition, GPUs have not.

For GPUs, abundance will come not only from building more chips, but from using software to make better use of the chips already deployed.

Today, GPUs remain underutilized. The best solutions sit at the workload layer, batching inference requests or queueing training jobs across fleets of servers. Meanwhile, the systems layer remains largely untouched. Virtualization is now universal for CPUs, storage, networking, and memory, but does not yet exist for GPUs.

Thunder Compute exists to fill this gap. We are a systems lab with commercial focus, productionizing cutting edge GPU virtualization research. Our team blends world class systems researchers with infrastructure experts, from Citadel Securities, Aquatic, and AWS. For four years we were in stealth building research prototypes; now, we deploy our findings through our own cloud and enterprise partners to create step-change improvements in data center capacity.