Who's Paying for the Compute? Czech Data Centres and the Quiet Race to Power Them

Czechia opened what is being described as its first dedicated AI data centre in 2026, a facility built by colocation operator MasterDC at its existing Kanice site outside Brno. The first phase delivers a modest 4 megawatts of capacity and around one hundred rack spaces, powered in part by an on-site and off-site photovoltaic installation — but the company has planned the site's total capacity at up to 25 megawatts, and has been explicit that the strategy is deliberately incremental rather than speculative: building AI-focused colocation, private AI cloud and GPU server rental capacity in step with confirmed demand rather than betting on a hyperscaler-style build-first approach. It is a small facility by global standards, but a symbolically significant one — the first sign that Czech AI compute demand is being met, at least partly, on domestic soil rather than routed entirely through hyperscaler regions in Frankfurt, Dublin or further afield.

The demand curve nobody can flatten

The scale of the challenge behind that first facility is genuinely global. The International Energy Agency has estimated that global data-centre electricity consumption grew 17 per cent in 2025, with AI-focused facilities specifically surging 50 per cent, and Morgan Stanley research projects hyperscalers alone could commit more than $1 trillion in capital to AI energy infrastructure across 2025 and 2026. For a small, energy-import-dependent economy like Czechia's, that global surge in demand for power-hungry compute lands directly on top of an electricity grid that was not designed with gigawatt-scale AI facilities in mind — which is precisely why the country's nuclear expansion plans, largely conceived before the AI boom accelerated, have taken on a second, more urgent rationale.

Nuclear as the answer nobody debates anymore

ČEZ, the majority state-owned utility, is in the middle of what is described as the largest energy investment in Czech history: an estimated $19bn project to build two new reactors at Dukovany, awarded to South Korea's KHNP after a lengthy and at times contentious tender process, with the state guaranteeing a 40-year income stream for the plant and financing the build through a loan CEZ will repay over three decades. Petr Závodský, the executive leading the Dukovany project, has been unusually direct about why the timeline matters now rather than in the abstract: robust electricity demand growth from data centres and electric vehicles is explicitly cited, alongside decarbonisation and energy security, as a reason the expansion needs to proceed on schedule. CEZ's existing Dukovany fleet has also been squeezing incremental capacity out of its four ageing reactors — reaching a combined 2,048 megawatts through a modernisation programme completed in early 2026 — while separately exploring extending the plants' operating lives to as much as 80 years, well beyond the original 60-year target.

The grid-versus-growth tension

The honest read on the Czech AI infrastructure buildout is that it is running on two very different timelines. Data-centre capacity, as MasterDC's Kanice project shows, can be built and switched on within months once permitting and site work clear. Nuclear generation capacity, by contrast, operates on a decade-plus horizon — the new Dukovany units are not expected online until the second half of the 2030s — meaning that whatever incremental AI compute demand Czech data centres add between now and then has to be absorbed by the existing grid, existing nuclear uprates, and whatever renewable and gas capacity utilities can bring online faster. Industrial electrification adds to the same load: the Czech battery and EV supply-chain buildout we covered in our automotive AI reporting is competing for the identical megawatts. That mismatch is not unique to Czechia; it mirrors a pattern the International Energy Agency has flagged globally, where planning and regulatory systems for grid connections are being stretched harder than the pace of AI compute demand itself.

What to watch

The practical question for Czech businesses evaluating where to run AI workloads over the next few years is less about model access — that's already largely solved via cloud providers — and more about latency, data-residency requirements under EU rules, and long-run power cost certainty. Domestic capacity like MasterDC's Kanice facility offers the first two; whether it can offer reliable, competitively priced power at scale will depend heavily on how quickly the country's nuclear and grid investments catch up with a compute demand curve that, on the current trajectory, shows no sign of flattening before the decade is out.