Backblaze has launched its This fall 2025 Community Stats report, which gives perception into how visitors flows throughout its platform and what that exercise reveals about trendy AI infrastructure. That is the corporate’s second quarterly Community Stats publication, alongside its ongoing Drive Stats and Efficiency Stats collection, and focuses on network-level habits, together with ingress, egress, bandwidth focus, and regional exercise throughout buyer workloads.
Fairly than presenting theoretical fashions, the report attracts from manufacturing visitors noticed by Backblaze’s Community Engineering workforce, which displays how information flows into, out of, and throughout the platform over time. The purpose is to grasp how buyer habits and broader business modifications present up on the community layer. In This fall, AI emerged as one of many strongest drivers behind shifting visitors patterns.
A major issue behind this shift is B2 Overdrive. Launched in April 2025, it gives a direct, high-performance connection between Backblaze storage and neocloud compute environments used for mannequin coaching, inference, and experimentation. With this in place, Backblaze now has visibility into how AI workloads transfer huge datasets between storage and compute, typically briefly however intense bursts.
These workflows usually comply with a well-recognized cycle: intensive collections of pictures, movies, and metadata are ingested and consolidated, then exported for coaching and experimentation. As fashions evolve, these belongings are refreshed, up to date, and saved once more, creating recurring waves of high-volume visitors. Backblaze operates at each ends of this course of, serving because the sturdy storage layer throughout ingestion and because the high-throughput supply feeding downstream compute platforms. As soon as fashions are educated, they’re saved, served, and periodically retrained, with Backblaze persevering with to offer the underlying storage.
AI Workloads Are Driving Bigger, Extra Concentrated Site visitors Flows
Throughout This fall, Backblaze recorded heavy visitors between its platform, neocloud suppliers, and conventional hyperscalers, concentrated primarily from June by means of November. This exercise reveals important ingestion occasions adopted by intensive processing and model-related information exports. So far as networking goes, this represents a shift from diffuse, internet-style visitors to fewer endpoints carrying a lot bigger flows, a sample that aligns carefully with AI-centric infrastructure.
When visitors is grouped by community kind, Backblaze’s report reveals a number of tendencies. Content material supply networks, internet hosting suppliers, and web service suppliers remained largely inside historic ranges, per regular workloads for content material supply, internet hosting, and backups. Two classes confirmed noticeable development over the quarter.
Migration visitors elevated from August by means of October, pushed by massive information transfers over personal fiber connections inside information facilities. These hyperlinks are used to effectively onboard massive datasets with out counting on the general public web. On the identical time, neocloud visitors rose noticeably from July by means of November (peaking in October), indicating growing interplay between Backblaze storage and AI-focused compute platforms. BackBlaze says these modifications counsel the next baseline stage of community exercise is rising.
To assist body the information, Backblaze additionally outlines its regional and community classifications. US-West is the corporate’s largest and longest-running area, whereas US-East reveals the strongest proximity to the neocloud infrastructure. CA-East, positioned in Canada, is Backblaze’s latest area. Community classes embrace CDNs, conventional internet hosting suppliers, hyperscalers, regional ISPs that deal with last-mile connectivity, Tier 1 ISPs that carry long-distance visitors, neoclouds targeted on AI compute, and migration hyperlinks used for large-scale onboarding.
Heatmaps Present The place AI Site visitors Is Concentrating
Backblaze used heatmaps to visualise the place AI-related exercise is clustering throughout its areas.
Wanting first at complete visitors quantity, US-West-to-regional-ISP connections dominate. That is anticipated given the scale of Backblaze’s footprint in that area and its entry to web exchanges akin to Equinix-IX, which carry Backblaze nearer to client networks and scale back latency for on a regular basis workloads.
Extra notable is the focus of neocloud visitors in US-East, the place stream information reveals exercise clustering round cities akin to Chicago, Dallas-Houston, Denver, New York, Northern Virginia (together with the Reston and Ashburn hall), and Atlanta. This focus within the japanese United States traces up with the place a lot of at present’s AI compute capability is positioned. Shorter distances between storage and compute scale back latency, making it simpler to maneuver massive volumes of knowledge at excessive speeds. That proximity helps clarify why a lot AI-related visitors clusters in these areas.
As a result of that is the primary quarter with an entire dataset, Backblaze notes that it’s too early to attract long-term conclusions about whether or not these concentrations will shift. Monitoring these actions over future quarters is now a key focus.
Excessive-Magnitude Transfers Level to Sustained AI Pipelines
A second set of heatmaps examines “magnitude,” outlined because the variety of bits transferred per IP handle. This metric combines quantity with the variety of collaborating endpoints, offering perception into the load on particular person information flows.
Excessive-volume visitors unfold throughout many IPs is simpler to distribute and steadiness throughout the infrastructure. Excessive-volume visitors involving only some IPs is more difficult, but in addition extra revealing, because it displays massive, sustained transfers between particular programs.
With B2 Overdrive in place, Backblaze routinely helps shopper transfers starting from 100Gbps as much as 1Tbps. These high-magnitude flows are most outstanding in areas serving AI-heavy neocloud endpoints, indicating that clients are actively transferring huge datasets by means of the platform as a part of their AI pipelines.
A 3rd heatmap examines uniqueness, which BackBlaze measures because the variety of distinct IP addresses concerned in visitors. US-West reveals the best general uniqueness, attributable to its bigger variety of information facilities and broader mixture of workloads. In contrast, Neocloud visitors usually includes fewer, extra persistent endpoints. This sample aligns with AI pipelines that depend on steady, long-running connections between storage and compute, indicating a broader shift towards sustained, high-throughput relationships between specialised programs quite than many-to-many communication.
Early Quarter-Over-Quarter Information Reveals Rising Cloud-to-Cloud Motion
This fall additionally introduces Backblaze’s first quarter-over-quarter comparisons, however that is the primary full quarter of collected information since monitoring started in August 2025.
Wanting on the early quarter-over-quarter numbers, visitors going to different cloud suppliers jumped from 36.2% to 49.6%. Neocloud locations slipped barely, falling from 19.8% to 18.4%, whereas visitors to hyperscalers elevated considerably, rising from 3.5% to 18%. Backblaze notes that this information displays its personal buyer combine and solely a single quarter of sampling, so it’s far too early to deal with these shifts as broader business tendencies. That stated, they do observe an general enhance in cloud-to-cloud visitors, with the vacation spot combine altering in comparison with the earlier quarter.
Wanting Forward
Total, Backblaze views This fall 2025 as an early have a look at what an AI-native community is beginning to appear like. AI-related visitors is exhibiting up in bigger, extra concentrated flows, neocloud connections have grow to be a key a part of how information strikes, and information gravity is pulling storage, compute, and community structure nearer collectively.
The following quarter would be the first with a full quarter-over-quarter comparability, enabling a extra in-depth have a look at how these AI-driven flows evolve. Backblaze additionally plans to look at areas akin to IPv4 versus IPv6 visitors, cross-cloud connectivity tendencies, and extra focus evaluation as extra information turns into out there.
