Networking & Connectivity

Data has to move between chips, servers, and data centres at very high speed. These companies build the pipes and roads that make that movement possible.

Training a frontier model is a coordination problem as much as a compute problem. Tens of thousands of GPUs have to act as one machine, exchanging gradients and weights hundreds of times a second. That only works if the network is fast enough not to be the bottleneck. The companies in this layer are the unsung heroes of the AI build out. Without them, the chips might as well be doorstops.