Backblaze, a well-known cloud storage firm, has identified significant changes in network traffic flow patterns attributed to the rise of AI-driven data movement towards neocloud providers. These changes suggest a transition from traditional internet traffic dynamics to those characterized by large bandwidth requirements essential for AI model training and operations.

The company, which consistently reports on its storage equipment performance, has expanded its insights to include network data through a newly introduced quarterly Network Stats report. In the fourth quarter of 2025, notably the second issuance of its kind, Backblaze reported an uptick of traffic between itself, emerging neocloud entities, and conventional hyperscalers such as AWS and Azure from June to November.

Neocloud platforms, reminiscent of GPU-as-a-service, cater specifically to AI training facilities as opposed to providing the broad range of services that traditional hyperscale providers offer. This movement has revealed an evolving network structure where hyperscaler and internet service provider (ISP) traffic maintains historical norms. However, a significant surge was observed in migration and neocloud traffic. Migration involves data input into Backblaze’s infrastructure via dedicated fiber channels, bypassing the public internet. Notably, migration traffic increased starting August, tapering off in October.

Concurrently, neocloud traffic spiked from July through November, with a peak in October. Backblaze correlates this pattern with the AI development cycle, involving data consolidation, training exports, and the iterative experimentation process.

Despite a reduction in migration and neocloud surges, baseline traffic levels have heightened across Backblaze’s networks. The firm notes significant activity concentration in its US-West region for general ISP traffic, while AI and neocloud engagements are primarily within the US-East area, particularly Northern Virginia known for intensive AI computational capabilities.

This shift, highlighted by localized IP interactions, reflects AI’s impact on network realignment, creating dense connections between data storage and compute services, potentially guiding future efforts in network, storage, and compute integration.

Backblaze plans to continue tracking and reporting these network phenomena quarterly, exploring aspects such as IPv4 to IPv6 transitions and cross-platform connectivity.