Backblaze, Inc. (BLZE) Discusses Quarterly Network Traffic Highlights and New Geo Data Insights Transcript – Image for illustrative purposes only (Image credits: Unsplash)
Cloud engineers building AI infrastructure now grapple with unpredictable bursts of massive data transfers rather than steady streams, according to fresh analysis from Backblaze. The company’s Q1 2026 Network Stats report, discussed in a recent webinar, maps these shifts through new geographic heatmaps and traffic breakdowns. These insights highlight how AI workloads concentrate power in specific regions, forcing providers to rethink capacity planning.[1][2]
Neocloud Traffic Cools in Winter, Then Rebounds
Backblaze observed a notable dip in neocloud and hyperscaler traffic during the winter months of early 2026. Combined, these categories fell to 25.5 percent of total network traffic in the first quarter, down from 36.4 percent in the prior period.[2] Neocloud volumes, often tied to emerging AI-focused clouds, showed particular variability with lows in January and February followed by an uptick in March.
CDN and regional ISP traffic filled the gap during this lull. CDN’s share climbed to 32 percent from about 20 percent, while ISP regional traffic rose to 27.8 percent from 21.5 percent. Overall network baseline activity increased quarter-over-quarter, pointing to sustained demand beneath the fluctuations.[1]
Geographic Heatmaps Spotlight AI Hotspots
Newly added geographic data transformed Backblaze’s analysis, enabling heatmaps that pinpoint traffic concentrations. In the United States, neocloud activity clustered heavily in California and the Ashburn-Reston corridor in Virginia, areas dense with GPU clusters and data centers.[2] These regions dominated high-magnitude flows, where bits transferred per IP address remained elevated even amid lower volumes.
Beyond the U.S., emerging global nodes appeared in Finland, Brazil, France, and Canada for neocloud traffic. CDN patterns proved more distributed, with the Netherlands leading outside the U.S. due to its connectivity hub at AMS-IX. Singapore stood out for content delivery, while Germany hosted steady traffic.[1]
- U.S. neocloud: California leads, with Virginia close behind.
- Global neocloud extensions: Finland, Brazil, France, Canada.
- CDN ex-U.S.: Netherlands, Singapore prominent.
- High magnitude persists in U.S. East, spreading to West and EU-Central.
Elephant Flows Define AI’s Network Signature
Webinar speakers Brent Nowak and Stephanie Doyle introduced “elephant flows” as a hallmark of neocloud traffic – large, high-bandwidth transfers between few endpoints that aggregate into intense bursts.[3] Unlike the small “mice flows” of typical web traffic, these demand specialized handling, such as 100G or 400G ports and private network interfaces.
Magnitude metrics underscored this intensity: neocloud transfers sustained high bits per IP despite winter slowdowns, reflecting GPU clusters moving vast datasets in short, powerful sessions. Unique IP counts revealed diversity in U.S. West ISP traffic but concentration elsewhere, guiding infrastructure upgrades like added capacity in high-impact zones.[1]
Neocloud and hyperscaler patterns contrasted with steadier CDN and hosting flows, which spread across more endpoints and proved easier to balance.
Infrastructure Planners Shift to Extremes
Executives framed these findings as a call to action for the industry. “The data tells a clear story: AI is reshaping global infrastructure investment,” stated Gleb Budman, Backblaze CEO. “GPU clusters are concentrating not only in traditional high-intensity regions such as Northern Virginia and California, but also in Finland, Brazil, and beyond.”[2]
Dan Spraggins, Backblaze’s SVP of Engineering, added that providers must move “from planning for averages to engineering for extremes.” The company has tuned its stack accordingly, supporting fluid data movement for volatile AI pipelines. For more details, see the full report on the Backblaze blog.[1]
During the May 4 webinar Q&A, questions touched on Finland’s rise and flow visibility limits, confirming the data’s network-level focus without application specifics.[3]
These patterns signal broader challenges for data center operators worldwide. As AI training cycles evolve, winter lulls may recur, but the baseline rise suggests enduring growth. Engineers in California or Virginia hubs, and now emerging spots like Finland, will feel the pressure to scale for those elephantine surges first.