On 18 November 2026, RIPE NCC published a RIPE Atlas measurement study, “Who’s Delivering GTA 6?”, by Emile Aben with Alun Davies and James Cowie. It describes DNS steering each player to a download server while several CDNs share delivery of the preload, which began on 10 November, and says several IXPs reported all time highs on the first day; RIPE names none of them, and the only exchanges we can confirm below are DE-CIX and INEX. For peering capacity planning, the useful part is what the study leaves out: it contains no port, peering or transit capacity figures.

The scale explains why. The game is a little over 200 GB and RIPE cites 12M+ pre-orders, so one server serving everyone would face about 2.4 EB. Nothing does that; the load is spread across CDNs, cities and networks, and each network sees only its own slice.

The public numbers come from the exchanges. According to trade-press reposts of its release, DE-CIX recorded 22.30 Tbps in Frankfurt at 20:20 Central European Time (UTC+1) on 10 November, with a global high of 33.01 Tbps, then 33.03 Tbps globally on 18 November. INEX in Ireland carried 12.53 PB in a day against a six-month daily average of 9.53 PB, about 31% above that average, according to a news report citing an unnamed source. Attributing those records to the game is an inference; DE-CIX has not said how much of the traffic was GTA 6 itself.

What DNS traffic steering does to your numbers

RIPE calls the technique GeoDNS: one lookup can return different CNAMEs, each controlled by a different CDN. For PS5 the hostname is gst.prod.dl.playstation.net, for Xbox assets1.xboxlive.com. In ping.pe lookups (not Atlas), the PlayStation name resolved to Fastly in Los Angeles, Google Cloud in Tokyo, CloudFront in Istanbul and Alibaba in Shanghai.

In the Atlas sample both consoles used Akamai and Fastly, only PS5 used Google and Amazon, and only Xbox used G-Core. RIPE warns the sample leans towards Europe and technical hosts, so read it as a mix, not as market share. RIPE also does not say how the CDNs allocate load behind those answers, so the effect on any one network is a possibility to plan for, not a measured result.

Where a CDN’s traffic lands on your ports can change with a DNS answer you do not control, so size headroom for the traffic you did not choose.

Reading the IXP share of paths

RIPE found IXPs on 11.2% of PS5 IPv4 paths, 21.4% of PS5 IPv6, 19.9% of Xbox IPv4 and 29% of Xbox IPv6. Larger ISPs tend to connect directly to the CDN, smaller networks tend to use IXPs. These are probe-based shares with no volume weighting, so they say nothing about bytes or port capacity.

The latency numbers need the same caution. The median was 8 ms via an IXP against 10 ms otherwise, which RIPE itself calls correlation, not causation. One Kyiv network (AS48438) saw 220 ms without an IXP and 16 ms with one; that is one network, not a rule. RIPE’s IPv6 paths touched IXPs more often than IPv4, but that is a path count, not bytes, so measure your own IPv4 and IPv6 session load separately.

The numbers to check before IXP traffic spikes reach you

RIPE says it can only imagine what the rest of the load did on transit links or PNI. That gap is exactly where an operator has to look at its own counters, in this order:

  1. Busiest-hour utilisation per port, not the daily mean. DE-CIX says it keeps infrastructure utilization at or below 75% to leave room for spikes. That is its own operating figure, a useful benchmark rather than a standard.
  2. The failure case at peak. Calculate what each port carries with one member link of its bundle down, at the busiest hour of your busiest week.
  3. A release-day multiplier from your own history. DE-CIX reported launch-period traffic 25% above a typical week at the start of 2026; INEX, per that news report, was about 31% above its six-month daily average. Those are different baselines at different exchanges, not a combined range, so compute your ratio only against your own six-month average.
  4. IPv4 and IPv6 sessions separately. Check that your IPv6 peering sessions and their capacity are not an afterthought.
  5. Transit commit and burst. Whatever does not arrive over peering or private interconnects lands on transit. Know what your contract does when you exceed the commit.
  6. Your own aggregation layer. YouFibre reported congestion at one BNG and rebalanced traffic, while Virgin Media O2 reported record peak and daily volume. The bottleneck is often inside the network, not at the exchange port.

Then rehearse the move, not just the number: if a steering change shifts a city’s load from one CDN to another, which port takes it, and what does it cost?

Transit in a carrier-neutral datacenter

We operate our own network, and that is where the checklist above meets our own estate. None of our datacenters is owned by a carrier, so you are never buying transit from your landlord and never stuck with whoever owns the building. We sit directly on the major European internet exchanges, and we peer with AWS, Azure, GCP, Oracle and Cloudflare, so hybrid traffic takes a short, known path instead of whatever a transit provider chooses that day.

That peering is direct, at European exchange points, and reaches those clouds’ public services; a private interconnect into a customer’s own network inside AWS, Azure, Google Cloud or Oracle is possible from our sites, designed per customer. The uplink counts we can state are physical links to tier-1 carriers: 2 in Madrid, 6 in Amsterdam, 8 in Munich and 2 in Helsinki. They are not capacity figures, so ask any provider, us included, for per-port speeds and utilisation.

On Colocation we usually provide the transit ourselves, aggregated from all available transit providers, and you can also cross-connect to any carrier in the building. We additionally interconnect with 30+ network partners and run 44 points of presence across Europe and the Middle East.

Our four core routers are joined by our own Layer 2 network. It is a path between the sites, not automatic failover or replication, so if you need a workload in two sites, that is designed and bought. For reaching us from your own offices, plants or datacenter, Private Link lands at our nearest site and our Layer 2 network carries it on, encrypted from your site to ours.

What to do this week

Pull the busiest-hour graph for every peering and transit port from 10 November to today, and again after launch day, and compare it with your six-month average. If the ratio surprises you, that is your multiplier. Then work out which of your ports a DNS change at a CDN would load next.

Sources