Shifting Streams and AI Surges: What Our Data Reveals About the OTT Landscape
Summary
OTT data from early 2026 shows streaming hierarchies holding steady while AI platforms reshuffled rapidly. Claude tripled in traffic, overtook Gemini, and is on pace to challenge ChatGPT by fall. Doug Madory digs into the data in this new analysis.
We’re more than a third of the way into 2026, and we thought it might be interesting to take a look at some of the trends we’re seeing in our OTT Service Tracking in two of the most frequently discussed categories: video streaming and generative AI platforms.
Every category produced at least one moment where the steady-state pattern broke. We saw Claude’s customer reach spike 47% in a single week — the same week the Pentagon formally designated Anthropic a “supply chain risk.” Our customers saw Disney+ streaming bandwidth grow substantially over the quarter, while Netflix’s edged slightly downward.
Methodology
The data in this analysis relies on aggregate traffic data from OTT Service Tracking, which combines DNS queries with NetFlow to allow service providers to understand exactly how OTT services are being delivered to their end users – an invaluable capability when trying to determine what is responsible for the latest traffic surge.
The capability is more than simple NetFlow analysis. Knowing the source and destination IPs in the NetFlow of a traffic surge isn’t enough to decompose a networking incident into the specific OTT services, ports, and CDNs involved. DNS query data is necessary to associate NetFlow traffic statistics with specific OTT services in order to answer questions such as, “What specific OTT service is causing my peering link with a certain CDN to become saturated?”
A few important caveats up front
This analysis draws on measured traffic from US regional and national ISP customers who have opted in, so the patterns here may not exactly mirror the broader national market. Subscriber growth at any one of those ISPs can also push a service’s traffic upward without any actual change in per-user consumption. Finally, there is some variation in how Kentik is deployed on these networks, which can affect the traffic statistics when aggregated. For these reasons, the relative shape of the curves and the differences between services are more reliable signals than the absolute traffic levels.
Streaming services
The left panel on the graphic below shows the absolute hierarchy of streaming flows-per-second on a log scale: YouTube and Netflix at the top, separated by a modest gap; Disney+ and Amazon Prime Video clustered in a middle tier; Hulu and Peacock at the bottom. The vertical span across all six services is roughly an order of magnitude — YouTube generates about 10x the per-second flow count of Peacock — and the rank order held remarkably steady through the entire window. None of the services overtook another; the gaps simply widened or narrowed at the margins.
The relative-growth panel on the right tells a more dynamic story. Disney+ is the clear standout, climbing steadily from January through mid-March and holding nearly 50% above its January 12 baseline through April. We chose the baseline of the week of January 12 to avoid abnormal holiday viewing patterns. Hulu (the Disney subsidiary) followed a similar trajectory, ending the period 40-50% higher. Amazon Prime Video posted modest sustained growth, and YouTube was nearly flat. Netflix is the one major service whose flow count actually declined over the period, ending below its starting level – a notable departure from where the rest of the streaming category went.
The most striking feature of the chart is Peacock’s February spike, a near-3x jump in the weeks of February 2 through February 16. The timing aligns with two major NBC properties that streamed on Peacock: Super Bowl LX on February 8 and the Milan-Cortina Winter Olympics, which ran February 6-22. Outside that window, Peacock’s flow count returned roughly to its pre-event level, suggesting the spike was event-driven viewership rather than a step-change in the subscriber base.
Switching to another one of the many metrics we track: unique destination IPs. This metric gives a different lens on the same period. Instead of counting connections, this metric tracks the number of distinct customer-network endpoints that reached each service in any given hour, a closer proxy for “how many users showed up” than flows-per-second. The absolute hierarchy on the left looks broadly similar: YouTube and Netflix at the top, Disney+ and Amazon Prime Video in the middle tier, Hulu and Peacock at the bottom. The one notable change in the middle – Disney+ pulled clearly ahead of Amazon Prime Video starting in mid-February, whereas on flows-per-second, the two were much closer together.
The relative-growth panel mostly tracks the NetFlow view, with a few telling differences. Peacock’s February spike is still visible, but smaller, suggesting the Olympics and Super Bowl drove a lot more activity from existing viewers than from entirely new ones. Disney+ and Hulu remain the strongest sustained growers, peaking almost 50% above their January 12 baselines and easing back slightly into late April. Netflix again declined, mirroring its drop in flows-per-second – even on the more forgiving “are users showing up at all” measure, Netflix lost ground over the quarter.
Is Netflix viewership declining?
Netflix is a useful illustration of why having different metrics matters. Its flows-per-second and bits-per-second both edged down over the period, while its unique destination IPs held roughly flat. That pattern means the same number of customer endpoints reached Netflix, but each one generated fewer connections and fewer bytes. Lower streaming resolution would account for the bit drop, as would AI-assisted bitrate ladder generation, a technology that lets services deliver equivalent visual quality at meaningfully lower bitrates. Neither explains the flow decline, though, since flow count is governed by HTTP segment cadence rather than bitrate. The more parsimonious explanation is reduced per-session watch time.
AI platforms
AI platforms are a much younger category than streaming video, and that shows up in the data immediately. The absolute-traffic panel on the left shows ChatGPT well ahead of both Claude and Gemini in flows-per-second through the entire window, more than doubling over the period. Gemini was the second-largest of the three for most of the period, but its growth was the slowest — it edged up gradually before leveling off in April. Claude started smallest, climbed steadily, and overtook Gemini in mid-April to finish the period as the second-largest of the three on this metric.
The relative-growth panel on the right makes Claude’s trajectory the headline. Indexed to January 26, Claude grew roughly 3.1x by late April, ahead of ChatGPT’s ~2.1x and well ahead of Gemini’s ~1.4x. We chose to baseline against the week of January 26 because of a data artifact in the week prior. The most striking feature is the single-week spike the week of March 2, when Claude’s flows briefly hit 336% of baseline before snapping back the following week. The timing matches the Pentagon’s formal designation of Anthropic as a “supply chain risk” on March 3 – a connection we’ll come back to later in this section. Outside that one-week event, Claude’s growth has been steady and accelerating: The slope from mid-March onward is steeper than ChatGPT’s.
The same exercise on unique destination IPs tells a similar but slightly slower story. Over the same post-spike window, ChatGPT’s customer reach grew ~3.1% per week compounding, while Claude’s grew ~10.3% per week – virtually the same growth-rate gap we saw on flows-per-second. Cross-validated across two independent metrics, those rates look reasonably reliable.
When might Claude overtake ChatGPT?
Setting up the question as a math problem (extrapolate the post-spike growth rates and project when the curves cross) gives a more concrete answer than you might expect, with one wrinkle: the answer is meaningfully different depending on which metric you use.
Both flows-per-second and unique destination IPs show Claude growing at roughly 3.4x ChatGPT’s compound rate (about 10.5% per week versus approximately 3.1% per week, using the eight weeks of post-spike data from March 9 through Apr 27). But Claude’s starting fraction of ChatGPT differs across the two metrics: it’s at 38% for flows-per-second and only 27% for unique destination IPs.
That difference produces a five-week gap between the two crossover estimates. On flows-per-second, Claude’s curve crosses ChatGPT’s the week of August 3, 2026, about 14 weeks out from April 27, assuming steady rates of change.
On unique destination IPs, the crossover lands around the week of September 7, 2026, about 19 weeks out—same compound growth rate; more ground to cover on IPs, so the crossover lands later. The gap also tells us something about the two user bases: ChatGPT’s reach across customer networks is wider relative to its flow count than Claude’s, suggesting that Claude users generate more flows per IP on average. Claude’s user base is more activity-intensive per customer, consistent with a mix that skews toward developer and API usage.
Please note the log scale on the y-axis in the graphic below summarizing these projections.
There’s a more imminent crossover that’s worth flagging, with one caveat. On unique destination IPs, Claude is essentially tied with Gemini as of the week of April 27, putting the two within rounding distance of parity. With Claude growing on this metric at ~10%/week and Gemini at ~1.5%, the crossover is days away under either extrapolation method: linear projects mid-May, compound puts it within the current week.
The caveat: A growing share of Gemini’s usage is delivered inside Google Search (most notably via AI Overviews) rather than through gemini.google.com or the API directly, and that embedded traffic doesn’t classify as “Gemini” in our telemetry; it appears as ordinary Google Search activity. The numbers above, therefore, reflect direct Gemini reach, not Gemini’s full footprint as an underlying model. Within that direct-reach lens, Claude is on track to become the #2 AI platform almost immediately, and the open question after that is how fast it can close the gap on ChatGPT.
Linear extrapolation tells a different story for the larger ChatGPT crossover. ChatGPT and Claude are adding roughly the same absolute traffic each week, so under a strict linear model, the gap doesn’t close at all, and Claude never overtakes ChatGPT. The two methods diverge precisely because the gap is large, and an exponential curve eventually catches a linear one. The compound case assumes Claude’s recent growth rate holds for several more months; the linear case effectively assumes it normalizes immediately. Reality is almost certainly somewhere in between.
In summary, current trends point to a Claude/ChatGPT crossover sometime in August or September 2026 if the post-spike growth rates hold, and never if growth normalizes to absolute-volume parity. Most likely scenario: Claude continues to close meaningfully through the summer, with the crossover landing in the fall or sliding into 2027 if the growth rate decelerates even slightly. The Claude vs. Gemini crossover, by contrast, is a near-certainty for May.
Conclusions
If there’s a single takeaway from looking at OTT through the first third of 2026, it’s that streaming video and generative AI are very different kinds of markets. Streaming is mature, with a remarkably stable hierarchy: YouTube and Netflix sit at the top, Disney+ and Amazon Prime Video occupy a second tier, and the rank order barely budged across the window. The interesting movement happened at the margins, with Disney+ quietly extending its lead over Amazon Prime Video, Netflix drifting downward on per-user activity, and Peacock briefly jumping the entire ranking for two weeks of NBC tentpole programming before reverting. Generative AI is the opposite: a young, fast-moving category where ranks change monthly. Claude overtook Gemini on flows in mid-April and is days away from doing the same on customer reach. The pace of change in AI is much faster than anything we see on the streaming side or any other OTT category, for that matter.
A second takeaway is methodological. Looking at flows-per-second, bits-per-second, and unique destination IPs side by side surfaced patterns that no single metric would have caught. Netflix’s bits-per-second decline could easily have been read as engagement loss; cross-checking against the flat unique-IP series, and against industry reporting on AI-assisted bitrate ladder generation, pointed instead at supply-side encoding efficiency plus reduced per-session watch time. Claude’s March 2 spike in flows looked like a one-week blip on its own; on unique destination IPs the same week registered as a permanent step-up in customer reach, which changes its interpretation entirely. The differences between metrics are often the most useful signal in the data, and they’re the part of the picture you only see by looking at all three.
What to watch over the rest of 2026: on streaming, whether Netflix’s flows and bits decline continues or stabilizes, and how durable Disney+‘s gains turn out to be, given that Disney is forecasting a 10% streaming operating margin for the year. On AI, the Claude versus Gemini crossover on customer reach should be visible in our next data drop, and the larger Claude versus ChatGPT race will resolve over the summer one way or another: either Claude’s compound growth rate holds, and the crossover lands in August or September, or it normalizes and the gap stays open into 2027. We’ll be back later in the year with another look at the data; by then, we should be able to tell whether the trends in this post were the start of a longer reshaping of the OTT landscape or just an unusually noisy quarter.
