Generation, delivery and storage of the electricity AI data centers run on.
Land, buildings, cooling and electrical systems that house compute.
Accelerators, custom silicon, memory and the foundries that build them.
Servers, racks, switches and the ODMs that assemble them.
Owns or controls compute and rents it out as GPU-hours.
Connects supply and demand for raw compute without owning much of it.
Abstracts the GPU away — you submit a workload, you get tokens back.
The labs whose training and inference demand drives the entire stack.