Backblaze says AI traffic and neoclouds could shape future networks

Cloud storage firm Backblaze says that a sharp rise in AI-driven data traffic to neocloud operators may signal a shift from internet-style traffic patterns to large, high-bandwidth flows characteristic of large-scale model training and inference work.

Backblaze, known for its regular reports on the performance of its storage hardware, has now begun publishing a quarterly Network Stats report covering traffic movement into and across its infrastructure.

For Q4 2025, which is only the second period it has issued a network report, the firm observed a large amount of traffic between itself, neoclouds, and traditional hyperscalers from June through November.

Neocloud operators, or GPU-as-a-service providers, specialize in offering AI training infrastructure and little else, while traditional hyperscalers such as AWS and Azure provide a range of services covering most workloads.

Backblaze tracks network traffic by type, differentiating between internet service provider (ISP) data and that associated with hyperscalers, for example.