back to the blog

Partnership 4 min read 2026-09-22

AAA games on telco edge infrastructure: four weeks of results

168 live sessions across 20 days of continuous telemetry, zero packet loss, and one specific thing that needs fixing before the next phase. A trial with StonesThro, Zadara and Cornerstone, the UK's largest mobile tower company: the former Vodafone and O2 joint venture, now owned by Vantage Towers and Virgin Media O2.

Jeff Outlaw Jeff Outlaw COO
A map of the United Kingdom rendered as a field of green dots, with one node lit up and a connection arcing away from it, beside the logos of Cornerstone, StonesThro and Zadara

Cloud gaming only works when players never notice the infrastructure. If they feel lag, you’ve lost them.

So StonesThro, Zadara, Cornerstone and YOM didn’t argue on paper about whether distributed micro-edge compute can carry AAA gaming. We ran it, with real games and live player sessions.

the setup

YOM supplied the gaming workloads and player sessions. StonesThro, the UK sovereign micro-edge specialist, delivered compute and connectivity with two partners: Zadara for the managed edge cloud, and Cornerstone for the sites.

Cornerstone is the UK’s largest mobile tower infrastructure business. It began as the Vodafone and O2 network-sharing joint venture; Vodafone’s half has since passed to Vantage Towers, so today it is owned by Vantage Towers and Virgin Media O2.

It runs the towers and rooftops that carry both operators’ mobile networks: 16,500+ sites across the UK. Compute at those sites processes the game a few kilometres from the player, not in a distant data centre.

The objective was straightforward: determine whether telco edge infrastructure could support real-time, high-performance gaming at scale and identify the operational boundaries before moving into a larger deployment. StonesThro published its own account of the trial on LinkedIn.

the results

Across 20 days of continuous telemetry:

  • 168 live gaming sessions
  • Zero packet loss across measured sessions
  • Peak GPU memory use of 16.6 GB out of 48 GB, proving each node has room to serve significantly more players

what we learned

The network held up. Zero packet loss is a solid result, and it held across every session we measured.

What needs work is the computing environment behind each gaming session, specifically how consistently frames were delivered. Average frame rates held up well, but the slowest moments dropped to around 60% of the typical rate, and it is that variation rather than the average that a player feels. The cause sits in how CPU is allocated to each session, on both sides of the stack.

Finding that now, not in production, is the whole point of a trial. Running live traffic across the estate tells you how the whole stack behaves under real conditions, and where it bends.

why telco towers

A tower site is the one place in the country that already has power, fibre backhaul and a lease, and sits within a few kilometres of everyone. Nobody has to build it. YOM only has to put a node in it.

For the YOM network that changes three things:

  • Latency. The game renders on the footprint that carries the player’s mobile signal, so its round trip is a few kilometres, not a few hundred.
  • Coverage. 16,500 sites is national reach without a single new data centre.
  • Density. Each node ran at a third of its GPU memory, so the same site serves more players before anyone adds hardware.

And the estate is not gaming-specific. The node that streams a AAA title tonight runs AI inference for the same neighbourhood tomorrow, on the same power and the same fibre.

why it matters

The real question is whether distributed infrastructure can deliver console-quality play, reliably, at national scale, close to wherever players are.

For YOM, telco sites are one half of the network. Gaming PCs in homes give it density: hardware already bought, plugged in and cooled, in every city. Tower sites give it certainty: pre-permitted, pre-powered, spread across the country, never switched off, and with no interconnect queue to wait in. A session routes to the nearest qualified node of either kind, and the tower is the capacity that is still there when a home is switched off. That is what this trial put under real players.

This trial produced real operational data, a measured view of network performance, a specific list of fixes, evidence of spare GPU capacity, and a clear roadmap for the next phase: more games, more players, and more sites… followed by AI workloads.

Distributed cloud gaming isn’t just an idea on a slide. It’s already running. Don’t take our word for it: read StonesThro’s write-up of the same four weeks.

#edge infrastructure#cloud gaming#telco towers#partnerships#trial results
share:

the network under all of this

real-time work has to happen close by

About fifty kilometres from the person waiting for it. The hardware that close to people already exists, and most of it is switched off.