A forced equipment swap is the cheapest time to redesign a radio site

My position is easy to state and harder to carry out. When an operator has to remove a supplier's equipment from thousands of radio sites, it should not replace like for like. It should treat the programme as the one occasion in a decade when crews, cranes, permits and budget will reach every one of those sites, and use each visit to change the site's power system, its band plan and its legacy layers at the same time. That makes the programme harder to manage. It also turns a cost the operator did not choose into the largest energy and resilience improvement it is likely to make.

I write from one programme. Between January 2023 and November 2025, Corvalta Mobile replaced the radio equipment of a supplier that had come to be treated as high-risk on 4,100 of the 9,400 sites we then operated, after moving our packet core to other suppliers in 2022. I will not describe the security assessment behind the decision, which belongs to the authorities and to our board, and I will not name the supplier. My subject is what an operator does once the decision to remove equipment has been taken.

What like-for-like actually buys

The case for like-for-like replacement is respectable. It limits the number of things changing on a site at once, so when a cluster's performance drops after the swap the team knows where to look. It is easier to contract, because the new supplier quotes against a known inventory. It is easier to explain to a finance committee, because the cost of the programme matches the cost of the obligation and nothing else. Our chief financial officer, Marco Pellandi, made all three arguments in 2021, and went further. He proposed keeping the radios in service until their normal end of life around 2029, removing the supplier only from the core and from remote management access in the meantime.

He was right about the arithmetic. Spreading the radio replacement to 2029 would have cost about €130 million over eight years, against €196 million over three for my plan, and would have avoided a €61 million write-off of equipment with years of life left. What like-for-like buys, though, is the same site with a different badge. In 2021 our average macro site drew about 4.3 kilowatts. It ran a 3G layer carrying under two per cent of our data traffic, a rectifier and battery set specified a decade earlier, and air conditioning that ran because the old cabinets needed it. A like-for-like swap would have kept every one of those features, and by 2029 we would have paid to visit every site again.

One visit, four changes

Our programme made four changes on each visit. The old radios came out and new radios with deeper sleep modes went in. The 3G layer was removed and its spectrum refarmed to 4G and 5G. Where the battery set was more than seven years old we replaced it, sized to the role of the site rather than to a single standard. Where the new equipment allowed it, we removed the cabinet air conditioning and moved to outdoor-rated units with free cooling.

The energy result was the argument I had made to the board, and it held. Swapped sites draw 27% less electricity for the same traffic, worth about 38 gigawatt hours a year at full run rate. The battery work mattered in a different way. During the July 2025 heatwave, power cuts in parts of the south lasted longer than our old batteries would have carried, and 212 of the swapped sites stayed on air through outages that would previously have taken them down. A board never sees that kind of benefit in a business case, because the outage it prevents does not happen.

The objection I respected most came from our sharing partner, Sestriva Telecom, whose traffic runs on 1,300 of the sites we host under our rural agreement. Its engineers worried that four changes per visit multiplied the risk to their customers as well as ours. We agreed that any shared site would have the swap and the 3G removal on one night and the power work on a second, and that Sestriva could veto a window in its busiest weeks. That rule cost us about seven weeks across the programme. It was worth it, because the partnership came through a period in which it could easily have turned adversarial.

Where combining changes went wrong

The approach has costs I would not hide from another board. We delayed our 5G standalone launch by a year, from 2023 to 2024, because the same engineering teams were needed for the swap. We froze new site builds in two regions for nine months, which deferred about 180 sites and slowed our coverage gains in places where we were already the weakest network. And the diagnostic problem Marco had predicted did occur. In two of our 23 clusters, in Calabria and in the hills behind Genoa, dropped-call rates stayed above baseline for six weeks after the swap instead of the ten days we had planned. With four changes made, the team spent the first fortnight ruling out causes that a like-for-like swap would never have introduced. The fault turned out to be a neighbour-list setting carried over from the old configuration, which had nothing to do with power or with 3G.

There was also a failure that the combined design made worse. We assumed the outgoing supplier would keep providing spares and support on reasonable terms until the last site was done. In the final nine months, lead times for spare parts on the remaining 900 sites stretched to three weeks, and we kept faults open longer than we should have. I should have negotiated end-of-programme support in 2022, when we still had bargaining power, instead of relying on goodwill. Every supply contract we sign now prices exit support at signature.

None of this argues for loading unlimited change onto a forced programme. We kept two things out deliberately: we did not change transmission links on the same visit, and we did not combine the swap with new antenna designs, because both would have multiplied the diagnostic problem without a matching benefit. The test is not how much can be done on one night, but which changes would otherwise require a separate visit within a few years.

If another operator faces a similar programme, I would offer three tests rather than a template. First, describe the site as you would want it in ten years and count how many of those changes need a visit you are about to make anyway. Second, agree with any sharing partner, before the first site, which changes can share a night and which cannot, and give the partner a real veto. Third, price the outgoing supplier's support for the final year of the programme at the start, while it still wants something from you.

The removal was not a choice Corvalta made. What we did on each site while the crew was there was our choice, and I would make it the same way again, with a better contract for the last nine hundred sites.