The Data Refineries

Nikita Melashchenko · (14 July 2026) · opinion

Two news items from opposite ends of the data economy landed in my library within a week of each other. In Ireland, data centres now consume 23 per cent of the country’s metered electricity, up from 5 per cent a decade ago, more than urban and rural households combined. The regulator spent years rationing new grid connections around Dublin, and operators seeking more than 10 MW must now bring their own generation or batteries to the party. In India, the Finance Minister used the 2026-27 budget to promise foreign companies a tax holiday until 2047 for providing cloud services to global customers from data centres located in India, with Google, Microsoft and Amazon already pouring billions into the country.

In this post I described data localisation as the creation of artificial monopolies over a resource whose value grows with circulation, fiefdoms guarding their spice reserves. What these two stories tell me is that the game now has (perhaps, always had) a physical board. The binding constraint on where data lives is no longer only law. It is electricity, land and grid capacity, and states have noticed.

Note also that India’s measure inverts the classic localisation playbook. A traditional mandate fences domestic data in. The tax holiday instead pulls foreign data through, subsidising India as an export platform for data processing rather than sealing it off as a protected reserve. That is localisation by carrot, and my payoff matrix handled it poorly, because the carrot state is not restricting circulation at all. It is bidding for the harvesters and refineries rather than hoarding the data. If the fiefdoms of Arrakis were about who owns the resource, the new competition is about where the machinery sits, on the bet that data gravity, jobs and jurisdiction follow the machinery.

Ireland is the caution about what winning that auction costs. Eighty-plus data centres for five million people delivered tax revenue, investment and a seat in the global cloud, and also a grid at the edge of its capacity, public protest and a de facto moratorium in the capital. New York has now drawn the same conclusion in advance. Governor Hochul’s executive order pauses approvals for hyperscale projects for a year, the first statewide moratorium in the United States, while the state works out what the projects do to its grid, its water and its ratepayers, and the accompanying push would make data centres pay for grid upgrades and strip their tax subsidies. The subsidy India is offering is measured in decades of forgone tax; the price Ireland is paying is measured in gigawatt hours and political patience. Neither shows up in the trade agreements. Fiscal incentives conditioned on using local infrastructure sit in a grey zone between services commitments and subsidy disciplines, and the digital trade chapters that argue about data flows have almost nothing to say about the industrial policy of compute.

There is a small-state angle here that New Zealand has already lived through once. The pitch arrived in 2021 dressed as export diversification, when Amazon announced a hyperscale build in West Auckland with a promised $10.8 billion contribution to the economy over fifteen years and a thousand jobs. This May the company’s newly filed accounts showed a $45 million write-down on the abandoned site, with the investment redirected into servers and leased space in other operators’ data centres. Renewable-heavy, politically stable, temperate jurisdictions are exactly where the hyperscalers go looking. However, the machinery moves in both directions, and a promised data refinery is not a built one. The Ireland’s numbers are the due diligence a small state should run before saying yes. A data centre strategy is an energy policy and a land-use policy before it is a digital trade policy, and a state that discovers this after the FTAs are signed has discovered it too late.

The data must still flow, but I suspect the next decade of data governance will be fought over data centres and their associated infrastructure as much as regulatory data perimeters. As trade scholars and lawyers I guess we will need to learn to read grid maps.