Backblaze Publishes Q4 2025 Network Stats: Data Reveals the Rising Dominance of Neoclouds and AI Workflows

Core Insights - Backblaze's Q4 2025 Network Stats report highlights a significant increase in AI-driven data traffic to neoclouds, indicating a shift towards AI-native network behavior optimized for large-scale model training and inference [1] Group 1: Network Behavior Changes - AI is rapidly reshaping network behavior, with data traffic patterns showing a concentration in the US-East region, particularly around neocloud compute hubs in Northern Virginia, New York, and Atlanta [1] - The report indicates that large datasets are increasingly moving in short, sustained bursts, contrasting with previous diffuse internet traffic patterns [1] Group 2: Traffic and Migration Trends - US-West remains crucial for consumer traffic, leading in total traffic volume due to lower latency connections to consumer networks [1] - Migration traffic saw a spike from August through October, driven by the onboarding of large datasets via private fiber [1] Group 3: AI Traffic Characteristics - High-magnitude AI flows are becoming the norm, with a validated shift from many-to-many internet traffic to sustained, high-throughput connections between specialized storage and compute systems [1] - The report emphasizes the importance of low-latency connectivity for model training workloads, as AI traffic concentrates in specific regions [1]