Backblaze maps neocloud traffic hotspots in AI boom

Backblaze has published its Q1 2026 Network Stats report, adding geographic heatmaps that track neocloud traffic.

The data shows heavy concentrations of neocloud activity in California and the Ashburn-Reston corridor in Virginia, while also highlighting emerging hotspots in Finland, Brazil, France and Canada.

The quarterly report examines network traffic linked to neocloud, content delivery network and internet service provider activity. The latest edition provides a more detailed geographic view of where AI-related traffic is clustering and how it shifted over the quarter.

In the US, neocloud activity remained focused on established data centre regions on the West Coast and in Northern Virginia. Outside the US, traffic patterns showed growing concentrations in parts of Europe and the Americas.

That geographic detail comes alongside changes in the overall traffic mix. CDN and ISP traffic stayed broadly steady during the quarter, while neocloud and hyperscaler traffic fell over the winter months before rebounding in March.

According to the figures, CDN traffic increased from about 20% to 32% of total network traffic over the period. ISP regional traffic rose from 21.5% to 27.8%, while combined neocloud and hyperscaler traffic fell from 36.4% in the previous quarter to 25.5%.

Lower total neocloud volume did not necessarily mean lighter workloads. Backblaze found that neocloud transfers continued to show high magnitude, measured by bits transferred per unique IP address, indicating concentrated bursts of activity from a relatively small number of GPU clusters.

Traffic shifts

The report described a pattern in which fewer jobs appeared to run during the winter period, but those that did still moved large volumes of data. This supports the view that AI infrastructure demand is defined not only by aggregate traffic totals, but also by short periods of very intense network use.

At the same time, Backblaze said its overall network baseline increased quarter to quarter. That suggests a higher underlying level of demand even when neocloud traffic fluctuates over shorter timeframes.

Gleb Budman, Chief Executive Officer of Backblaze, linked the traffic patterns to broader infrastructure investment tied to AI.

"The data tells a clear story: AI is reshaping global infrastructure investment," said Budman.

"GPU clusters are concentrating not only in traditional high-intensity regions such as Northern Virginia and California, but also in Finland, Brazil, and beyond. Backblaze's high-performance cloud storage is enabling these new AI workflows to succeed globally," he said.

Infrastructure strain

The findings add to a wider debate over how operators design networks and storage systems for AI workloads. Training and inference jobs can create uneven demand, with long quieter periods interrupted by sharp spikes in data movement between storage and compute clusters.

Backblaze said its measurements show that this burst pattern persists even when headline traffic volumes ease. That matters for network planning because operators must still provision for peak load rather than average utilisation.

Dan Spraggins, Senior Vice President of Engineering at Backblaze, said the company sees the issue as one of engineering for peaks rather than normal conditions.

"AI infrastructure requires a fundamental shift from planning for averages to engineering for extremes," said Spraggins.

"We've optimized our stack to handle the high-intensity burst patterns of GPU clusters, ensuring that massive-scale data movement remains fluid regardless of the workload's volatility," he said.

The report is the second in Backblaze's quarterly network statistics series. It combines traffic flow data, regional analysis and new heatmaps to track how neocloud activity is evolving as AI-related workloads place greater demands on network and storage infrastructure.

One of the clearest findings was the contrast between neocloud concentration and CDN distribution, with the Netherlands leading CDN traffic because of connectivity through AMS-IX.