On 27 June 2026, one of the UK's most powerful AI supercomputers went dark. Dawn, the University of Cambridge's flagship research machine, was knocked offline during the June heatwave — reportedly after cooling infrastructure at the West Cambridge data centre failed to cope with the heat. Access was only fully restored on 6 July, more than a week later.
The timing could hardly have been worse. Just a day earlier, on 26 June, the Met Office had confirmed a new UK June temperature record of 37.7°C at Lingwood, Norfolk, with amber and red extreme heat warnings in force across much of England. And Dawn was not alone: the same week, Queen Alexandra Hospital in Portsmouth declared a critical incident after chillers failed at its on-site data centre, disrupting digital services and forcing some planned care to be postponed.
The cost of a hot week
Dawn is not an ordinary computing cluster. Researchers use it for cancer vaccine development, molecular screening for Parkinson's disease, and climate modelling — including British Antarctic Survey work on sea ice. When the machine stopped, that research stopped with it. Researchers affected by the outage have reported no data loss and say paused jobs will not need to be repeated, but a week of lost compute time on nationally significant research is a real cost, even before counting the engineering effort of recovery.
Nor is this a new failure mode. During the record heat of July 2022, when the UK first passed 40°C, two data centres serving Guy's and St Thomas' NHS Foundation Trust in London failed on the same 40°C day, leaving clinicians unable to access patient records. Full restoration took six weeks and cost the trust around £1.4 million.
The pattern is consistent: UK data infrastructure was largely engineered for a climate the country no longer has. Cooling systems specified against historical temperature norms are now routinely asked to operate beyond them. The Met Office issued its heat warnings days in advance; the heat itself arrived exactly on schedule. What failed was not the forecast — it was the link between the forecast and operational action.
Heat is the most forecastable threat a data centre faces
Unlike a fibre cut or a power surge, a heatwave never arrives unannounced. The June event was flagged by forecasters days ahead, and the third heatwave of the summer — under way now, with heat health alerts in England running to 11 July — was signalled days before the first uncomfortable afternoon.
That lead time is the entire opportunity. A data centre operator who knows on Monday that Thursday will hit 35°C can act while conditions are still normal: pre-cool the facility overnight, bring standby chillers online and test them under low load, defer non-critical batch workloads out of the peak-heat window, postpone maintenance that would take redundant cooling capacity offline, and staff up for the days when margins are thinnest. None of these measures is exotic. All of them depend on one thing — trusting the forecast early enough to act on it.
Building the forecast into operations
This is precisely the kind of decision-making One Call API 4.0 was designed to feed. A single integration gives an operations team current conditions at the facility's exact coordinates, a high-resolution 15-minute forecast timeline for the next 48 hours, daily forecasts for longer-range planning, and official government weather warnings — with alert references embedded directly in the weather response, so an amber heat warning can trigger an operational playbook automatically rather than waiting for someone to check the news.
For a data centre, the practical shape of this is simple: thresholds. When the forecast shows ambient temperatures approaching the facility's design limits, the monitoring stack raises the flag days out — enough time to pre-cool, reschedule, and verify redundancy. The 15-minute timeline then matters most on the day itself, when cooling load tracks outdoor temperature closely and operators need to know whether the peak comes at 14:00 or 16:30, and how long it will hold.
The forecast is the cheapest redundancy there is
Upgrading cooling plant to a new climate reality is expensive and slow. Integrating a weather feed into operational monitoring is neither. As UK summers continue to break records — three heatwaves before mid-July this year — the organisations that treat weather data as core operational infrastructure, rather than background information, will be the ones whose research, patient records, and services stay online when the temperature climbs.
Dawn is back up. The next heatwave is already here. The question for every infrastructure operator is whether the warning will be built into their systems before it is needed — because the forecast, unlike the cooling, has never failed to show up.