Open RAN belongs where we can name who fixes the fault

I think a challenger mobile operator should treat Open RAN as a permanent and deliberately small second source, built in parts of its network where it can name one party accountable for every fault, and should stop presenting it as the architecture its whole radio network will eventually adopt. I hold that view after running 240 open-interface sites for more than two years at Corvalta Mobile, and after watching the industry's debate swing from enthusiasm to fatigue without settling on what the evidence supports.

I am not arguing against open interfaces. The pilot was worth doing and we have kept it. I am arguing against two habits: announcing a migration percentage before the operating model exists to support it, and judging a pilot on the radio performance it achieves rather than on the time it takes to restore service when something breaks.

What the pilot measured

In 2022 we built 240 sites in a rural and small-town cluster in Emilia-Romagna, using radio units from one supplier, baseband software from a second running on commercial servers, and a systems integrator contracted to make them work together. We chose the cluster because it was large enough to carry real traffic, close enough to our Bologna operations centre for engineers to visit and far enough from any city that an extended problem would not reach most of our customers.

By the end of the first year radio performance was close to that of our conventional sites. Throughput and dropped-call rates came within a few per cent of a matched conventional cluster nearby. Energy was worse. The open sites drew about 14% more power for the same traffic, mostly because the general-purpose servers did not yet sleep as deeply at night as integrated baseband units. The suppliers' roadmaps promised to close the gap, and by mid-2026 about half of it had closed.

The commercial result came sooner. When we tendered the radio replacement programme described in my other essay, both established suppliers bidding knew that we had a working alternative in the field. I cannot prove what their prices would otherwise have been, but the bids came in about 18% below our cost model, and two procurement colleagues who have run many tenders told me that had not happened at Corvalta before. A credible second source changes behaviour long before it carries much traffic.

The fault with three owners

The finding that shaped my view came from the operations log. Over the first two years, the mean time to restore service after a fault at an open site was 2.4 times the figure for our conventional sites. Most faults were not hard to fix once found. They were hard to assign. A lost connection between radio unit and baseband could sit in the radio's firmware, the baseband software, the server's timing card or the fronthaul link, and each supplier's first response was evidence that the fault lay with someone else. The integrator was contracted to coordinate, but it could not compel a supplier to send an engineer, and our own engineers spent hours assembling logs from three portals with different time stamps.

We improved this. From mid-2023 we required every supplier to stream its logs into one store that we controlled, with synchronised clocks, and we rewrote the integrator's contract so that it held the service level for the whole site and carried penalties to match, with back-to-back terms with the component suppliers. Mean time to restore fell, but in 2025 it was still 1.6 times the conventional figure. Some of that gap reflects a young supply chain and will shrink. Some of it is structural. When a site is assembled from parts made by different companies, someone has to be paid to own the seams, and that someone needs both technical depth and contractual authority.

There is a further cost that rarely appears in pilot reports. Our operations engineers had to learn a different way of working, much closer to running a data centre than a radio network. Those who learnt quickly were in demand elsewhere in the company, and two of them moved to our cloud team within eighteen months. An operator with a small engineering bench, as challengers usually have, pays for that learning from the same people it needs for everything else.

A standing second source

My conclusion is that Corvalta should keep open-interface radio as a deliberate minority of its network, built where three conditions hold. The sites sit in layers or places where a longer restoration time affects few customers, such as rural capacity layers and indoor small cells. One party, either the integrator or ourselves, owns every fault and has the people to do so. And the cluster is large enough that the suppliers treat it as a reference customer rather than an experiment. Under those conditions we will grow from 240 sites to about 700 by 2028, mostly indoors and in rural areas. We will not set a percentage for the whole network.

Two colleagues disagree with me, from opposite directions, and their arguments deserve a fair statement. Our head of procurement believes that without a public commitment to scale, the open suppliers will not invest in our account, and the price effect will fade once the established suppliers see that our alternative is capped. He may be right, and I will watch the next tender for evidence. Our head of radio planning argues that even 700 sites will stretch a small team, and that we should hold the pilot at 240. I think the indoor programme is the right place to grow, because the operational stakes are lower there and the suppliers are improving fastest.

There is one more point I would put to any board. Open RAN is often discussed as a way to reduce dependence on a few suppliers, and it can do that. It also creates a new dependence, on the integrator or on the operator's own engineers, and that dependence is easier to overlook because it does not appear as a line in the supplier register. A board that asks about supplier concentration should ask about integration concentration too.

An operator weighing Open RAN should ask three questions before it asks about radio performance. Who, by name and by contract, owns a fault on an open site at three in the morning? How long, measured rather than promised, does it take to restore service there compared with the rest of the network? And what is the second source worth in the next tender, whether or not it ever carries much traffic? If the answers are good, build more. If they are not, keep the pilot small and keep it honest, because a working alternative in the field is worth more than an announcement of one.