The opportunity
Demand for AI training and inference is creating a market for cloud-style access to GPUs, and infrastructure owners are well placed to serve it.
The approach
SoftBank runs Infrinia on infrastructure it owns and offers AI-optimized Kubernetes and Inference itself. Infrinia is the software, not the service.
- Racks addedInfrinia finds every server
- Burned inEvery node, before it can be sold
- Benchmarked with the rackTogether with its rack
- Real training jobMust converge, bit-identical on every GPU
- Sold under SoftBank’s brandAs AI-optimized Kubernetes and Inference
