Introducing AI Platform Traffic Visibility in Kentik
Summary
Generative AI has quickly emerged as a distinct over-the-top (OTT) traffic category, yet service providers have lacked the visibility to track it. Kentik now identifies traffic from 35+ AI services across 28 providers, showing operators how it enters their networks, where it’s growing, and what it costs to deliver.
ChatGPT, Claude, Gemini, Copilot, Perplexity, and other generative AI services have rapidly moved from experiments to everyday tools. Subscribers use them to search, code, create images and video, analyze documents, and automate work. Businesses are weaving AI APIs and agents into more of their applications and services.
For internet access providers, however, this emerging class of OTT traffic has been effectively invisible. AI platform traffic often lands in an “unclassified” bucket or is folded into a generic cloud or CDN category, leaving operators unable to see what AI traffic is growing on their networks, let alone plan for it. That makes even basic questions surprisingly difficult to answer: Which AI services are growing? How are they reaching the network? Which links carry them? Are subscribers experiencing performance issues? What does that traffic cost to deliver?
Today, we’re bringing that traffic into view. AI Platforms is now a dedicated category in Kentik’s OTT Service Tracking workflow, giving network teams the same service, provider, connectivity, subscriber, and capacity context they already use to understand video, gaming, social, and other OTT traffic. This release enables infrastructure planning teams to get ahead of new AI network traffic patterns — including capacity planning for AI growth, interconnection strategy, monitoring subscriber experience, and visibility into what AI traffic costs to deliver.
AI traffic is becoming a network planning question
AI is a young OTT category, and its rankings are changing quickly. In Doug Madory’s recent analysis of the changing OTT landscape, ChatGPT’s flows per second more than doubled between late January and late April 2026. Claude grew roughly 3.1 times over the same period and overtook Gemini, while Gemini grew about 1.4 times. These shifts show why service-level visibility matters: the AI traffic mix is changing rapidly, and operators need to stay ahead of it to make informed capacity planning and network design decisions.
AI traffic patterns may also change as adoption and use cases mature. Today’s consumer internet is dominated by downstream delivery, but emerging AI use cases may shift that balance. AT&T has cited robotaxis, which generate about 20 GB of data per day, much of it upstream to the cloud, as an example of why networks may need to support more symmetrical traffic. If machine-generated workloads continue to grow, access networks may need to revisit assumptions about directionality, edge capacity, and performance.
However the agentic era develops, operators need visibility into how AI traffic is shifting across their own production networks. Industry-level data can reveal broader trends, but network planning depends on local details: how each AI service enters the network, which markets and delivery paths carry it, and where capacity constraints are beginning to emerge. Those answers come from the operator’s own telemetry.
AI Platforms joins the OTT workflows operators already use
With this release, Kentik now classifies traffic from 35+ AI services, covering the AI services your subscribers actually use, including OpenAI (ChatGPT, Sora), Anthropic (Claude), Google (Gemini, Vertex AI, AI Studio), Microsoft (Copilot, GitHub Copilot), Meta AI, Grok, Perplexity, Midjourney, DeepSeek, Mistral, Character.AI, and many more. Each one gets the full OTT treatment: mapped to category, provider, and service, with connectivity type, source CDN, and source ASN attribution.
This granular OTT visibility relies on three layers of the Kentik platform working seamlessly together:
- True Origin. Our True Origin engine correlates DNS telemetry with traffic flow data to deterministically identify the OTT service behind network flows, without deep packet inspection (DPI). It’s the same engine that already maps 1,100+ OTT services across 40+ categories and 66+ CDNs in real time, continuously learning as providers shift infrastructure.
- Automated workflows built for operators. OTT Service Tracking contains purpose-built workflows that automate what service provider teams actually do: scorecard edge capacity per service, isolate subscriber performance issues, and map the digital delivery supply chain behind every application entering the network.
- Industry-leading match rates. With OTT Intelligent Classification, Kentik statistically classifies traffic that deterministic methods can’t catch (external DNS resolvers, cached lookups, and the like), bringing total classification to 90%–95% or more and providing a much more complete view of the traffic on your network.
Together, these layers give network teams a consistent way to compare AI traffic with other OTT categories, then drill down from a trend line to the specific path, site, provider, CDN, interface, or subscriber segment behind it.
AI traffic use cases
Capacity planning for AI growth
Establish a baseline for the AI Platforms category, then follow growth by individual service, market, site, provider, and interface. The Capacity view uses the same treemap and device drill-down available for other OTT services, making it easier to spot where a fast-growing service is concentrated on a small set of edge interfaces. That helps teams prioritize upgrades based on observed demand rather than a broad market forecast.
Map delivery paths and guide interconnection strategy
For each AI service, map the inbound chain from the source ASN or CDN through the connectivity provider and entry site to your network. See whether the traffic arrives through transit, an IX, free or paid peering, or embedded infrastructure. As volume grows, this gives peering and interconnection teams the evidence to evaluate a new PNI, shift traffic to a better-performing path, or renegotiate an upstream agreement.
Monitor subscriber experience and troubleshoot complaints
Use unique destination IPs as a subscribership proxy and compare subscriber performance by site, provider, or connectivity type. If customers report slow responses from a specific AI service, operators can check whether the issue is concentrated on one market or delivery path and whether the responsible interfaces are constrained. Bitrate per destination IP is a network-delivery proxy, not an application response-time measurement, but it gives teams a practical place to start and a clearer blast radius than a generic traffic view.
Understand what AI traffic costs to deliver
With Kentik Traffic Costs, teams can connect AI service identity to the commercial model and utilization of the interfaces carrying it. Estimate the monthly delivery cost and effective cost per Mbps for the full AI Platforms category or an individual AI service, then identify which CDN, provider, connectivity type, or path is driving the spend. Broadband providers can use that evidence to compare transit and peering options, support interconnect negotiations, and keep delivery costs from rising unnoticed.
Bring AI traffic context into AI Advisor
Because the AI Platforms category is part of Kentik’s enriched traffic data, AI Advisor can use that context in conversational investigations. Operators can ask questions such as:
- “Which AI platform’s traffic grew the most over the last 30 days?”
- “How does ChatGPT reach my network, and which interfaces carrying it are closest to capacity?”
- “Summarize AI platform traffic by site and connectivity type for our weekly capacity report.”
AI Advisor will plan the investigation, query relevant telemetry and metadata, and return a data-backed summary with follow-up paths. The combination is useful for recurring reporting as well as one-off questions from engineering leaders who need the answer without rebuilding the analysis each time.
How to see AI platform traffic in Kentik
Existing OTT Service Tracking customers will find the new category already available in their portal.
- Open OTT Service Tracking. Go to Service Provider > OTT Service Tracking and select the AI Platforms category. If AI Platforms does not show up in your top categories list, you can use the Kentik global search to find it. The category page ranks AI services by traffic and breaks volume down by connectivity type, including embedded delivery, IX, free or paid peering, and off-net transit.
- Compare services and trends. Open Data Explorer from the category landing page, or build a query with OTT Service Category set to AI Platforms. Group by service and provider to compare average, 95th-percentile, maximum, and current traffic over the time range that matters to your planning cycle.
- Follow one service through the network. Select a service such as ChatGPT and open its Connectivity, Subscribers, or Capacity tabs. You can see the mix of peering and transit, the source CDN or ASN, entry sites and providers, unique destination IPs, performance proxies, and the interfaces carrying the traffic.
Start tracking AI traffic on your network today
AI platform traffic does not need to be the largest category on your network to deserve a clear view. It is already changing rapidly, and its delivery model will continue to evolve as services add richer media, APIs, agents, and machine-generated workloads. The sooner teams begin tracking these emerging patterns, the better positioned they will be to anticipate growth, adapt capacity and interconnection strategies, and address meaningful shifts before they become operational challenges.
Kentik now brings that visibility into the same network intelligence platform operators use to understand OTT demand, map the digital supply chain, protect subscriber experience, plan capacity, and manage connectivity spend.
If you’re an existing OTT Service Tracking customer, the AI Platforms category is already live in your portal. Log in and take a look at what’s on your network.
Not yet a customer? Sign up for a free trial or request a demo, and we’ll show you your AI traffic story.
