The Grid Doesn't Care About the Investment Announcement

In late 2021, the Irish government commissioned emergency gas generators to be installed near the port of Dublin, at an estimated cost of €1 billion. The generators were not part of any planned infrastructure programme. They were an emergency response to a grid that had absorbed more than a decade of data centre growth without a governance framework capable of managing the load. By that point, data centres accounted for 22% of Ireland’s metered electricity consumption — a share that had risen from 5% in 2014. Irish households were each paying approximately €100 per year extra on their electricity bills to cover the cost of grid pressure the planning system had not anticipated.

The generators at Dublin Port are not the story. The story is when the warning signs first appeared and what happened next.

In June 2020 — more than a year before the moratorium — the newly appointed Environment Minister received a detailed written briefing from his civil servants recommending a review of the security of Ireland’s electricity and gas supplies. The briefing was acted on. A review was commissioned. An early warning report was shared with a Cabinet sub-committee that included the Taoiseach. From 2010 to 2019, the Irish grid had issued 13 system alerts — warnings that supply might not meet demand. Between December 2020 and October 2021 alone, there were 7. EirGrid wrote to the regulator in June 2021 drawing urgent attention to the need for 200 megawatts of emergency generation by winter 2021. In November 2021, the moratorium arrived.

The warning signs were legible eighteen months before the freeze. The freeze arrived anyway.

Ireland was not the first. Amsterdam went earlier.

By 2019, Amsterdam’s data centre cluster had grown to the point where the city’s grid operator declared a bottleneck in energy demand. The municipality imposed a moratorium on new data centre construction. Amsterdam had been Europe’s most mature colocation market and one of its most attractive locations — a cable landing station, a neutral interconnection hub, a city with strong digital infrastructure and a government that had competed aggressively for hyperscaler investment. The moratorium did not care. The grid did.

Frankfurt followed in 2022, when the Main-Taunus district — home to one of Europe’s densest data centre clusters — imposed new development restrictions in response to grid and planning pressure. Dublin was already in its own freeze by then.

The sequence is Amsterdam 2019, Dublin 2021, Frankfurt 2022. Three of Europe’s four major data centre markets, in a rolling series of reactive governance responses to the same set of policy choices: attract aggressively, govern lightly, discover that the grid has limits. AWS’s submission to Ireland’s Commission for Regulation of Utilities, obtained through a freedom of information request, was explicit about what the warning signs had looked like in retrospect: “Over the past decade, we have had opportunities to do reinforcement work, prepare the grid for growth and investment. Those opportunities were not taken.”

The rules that Dublin now applies are stricter than anything it would have imposed proactively in 2015. New data centres above 10 MVA must provide their own dispatchable generation or storage capable of meeting their full electricity demand. They must be capable of feeding power back to the national grid during peak stress. 80% of annual energy demand must come from renewable generation within six years of opening. As of late 2025, an estimated €5.8 billion in Irish data centre projects hold planning permission but have no grid connection to plug into — they have cleared land, bought equipment, committed capital, and stopped. A further €7.8 billion sits in a pipeline whose viability depends on rules that were still being finalised when the investment was committed. The chairman of Digital Infrastructure Ireland described the situation to Irish radio as “self-sabotage.” Ireland’s IDA had already conceded that new data centres were “unlikely to happen in Dublin and the East Coast, at this point.”

This is what reactive governance costs. Not a fine, not a planning rejection, not an environmental assessment. A freeze — arriving with no transition, no accommodation for projects mid-build, and rules on exit that are harder than any rules the government would have chosen at leisure.

The arc has five phases. The warning signs that appear at each phase are documented. APAC markets are not at the start of an unknown trajectory. They are at known positions on a sequence that Amsterdam and Dublin have already completed.

Australia is the furthest along. The NSW parliamentary inquiry into data centres opened in May 2026. A Dublin city council local authority refused a Google data centre development citing insufficient grid capacity — the equivalent event in Australia was Penrith City Council’s formal objection to the AirTrunk Kemps Creek campus, on grounds that the site was unsuitable for a development of that scale. The NSW EPA found the Environmental Impact Statement for Kemps Creek insufficient during its exhibition period. The warning signs that Ireland’s Cabinet sub-committee received in April 2021 — eighteen months before its moratorium — are arriving in Sydney now.

Japan is in the same phase from a different direction. In Tokyo, community resistance is mounting faster than regulatory response. The Tokyo Metropolitan Government published non-binding guidelines in March 2026. The Japan Data Centre Council’s executive director told Kyodo that “right now, anxiety is prevailing.” Grid connection wait times in inner Tokyo now stretch up to ten years. The data centre market’s own leadership is describing a dynamic that Dublin’s operators would recognise from 2019.

India and Malaysia are one phase behind — in the growth phase, where the investment is accelerating and the warning signs are appearing but not yet registering as crisis. Malaysia’s regulator rejected 30% of data centre applications in early 2024 citing water concerns — a reactive instrument, not a prospective governance framework. India extended two coal plants by at least five years in late 2023 specifically because data centre demand was rising too fast to wait for alternatives. These are the system alerts. From 2010 to 2019, Ireland had 13. In ten months of 2021, it had 7.

Vietnam, Indonesia, the Philippines, and Thailand are earlier still — in the attraction phase, where the investment announcements are being made and the governance frameworks have not yet been written for what is being built.

Singapore is the only APAC market that has traversed the full arc on its own terms rather than being forced through it.

In 2019 — the same year Amsterdam imposed its moratorium — Singapore imposed its own pause on new data centre approvals. Singapore’s grid was not in crisis. Its pause was a deliberate choice to reassess the governance framework before the stress arrived rather than after. That timing is the entire difference between what Singapore has now and what Dublin has now.

The DC-CFA2 programme that Singapore launched in December 2025 requires a PUE of 1.25 at full load — more stringent than the EU’s 2030 target. It mandates at least 50% supply from green energy sources. It requires compliance with Singapore Standard SS 715:2025 on IT equipment efficiency. These are demanding conditions. They are demanding conditions that Singapore chose rather than conditions that a grid crisis imposed. The operators who want access to Singapore’s market have accepted them because they were negotiated proactively, with enough lead time for capital planning, rather than announced into a freeze where the alternative was a stranded project.

Dublin’s new conditions — 80% renewable, full self-supply above 10 MVA — are objectively stricter than Singapore’s. They arrived with €5.8 billion already stranded. Singapore’s arrived with the market intact and $12 billion in committed AWS investment still flowing.

The difference is not the standard. It is the sequence.

Singapore is the clearest case. It is not the only signal.

Korea did not wait for the EU AI Act to force binding AI governance. The AI Basic Act, effective January 2026, is risk-based, extraterritorial, and covers the same high-impact categories — healthcare AI, hiring algorithms, public services — that Brussels designated as high-risk. Korea chose its governance moment on AI before the external forcing mechanism arrived. The gap is on the physical infrastructure side — PUE thresholds, mandatory energy reporting, water efficiency standards — where Korea remains general industrial regulation. Partial early action is still earlier than none.

Japan is attempting something different and more modest: using incentives to redirect demand within the arc rather than regulate it. The Environment Ministry subsidises relocation of AI training workloads to Hokkaido at 50% of capital cost. Hokkaido has surplus renewable energy, natural cooling from its climate, and no grid stress. Tokyo is at Phase 3 — the warning signs are visible, the guidelines are non-binding. Hokkaido is at Phase 1 with conditions that could absorb growth without crisis. The subsidy is not a standard and it is not a regulation. It is an attempt to move the problem before the problem moves the policy. Whether it works depends on how many operators take it up before Tokyo’s grid forces the question.

Australia’s parliamentary inquiry is the mechanism that could matter most in the near term — but it cuts both ways. Unlike Tokyo’s guidelines, a parliamentary inquiry produces recommendations that can become law. The question is timing: whether the recommendations arrive before the constraint that forces them, or after. Ireland’s equivalent moment — the Cabinet sub-committee briefing in April 2021 — produced a review, not a regulation. The moratorium arrived seven months later. Australia’s inquiry opened in May 2026. The arc does not pause while it deliberates.

China is the only other APAC market at the reset phase on physical infrastructure standards — binding PUE below 1.3 from 2025 in major data centre regions, with water consumption limits in force. The political context is different and the model is not transferable. But the operational benchmark is real: China’s major data centre markets are running to a mandatory standard that most of APAC is still treating as aspirational.

McKinsey estimates that up to a third of global data centre demand is currently unmet. The investment flowing into APAC is not speculative — it is filling a documented gap, at speed, into markets that are at Phase 2 and Phase 3 of a five-phase arc that three European markets have already completed.

The window in which APAC governments can act at Phase 3 rather than wait for Phase 4 is the next twelve to eighteen months. Australia is the closest to the edge — the parliamentary inquiry, the planning objections, the EPA findings are the system alerts. India and Malaysia have eighteen to twenty-four months before grid stress and community backlash compound into a constraint that looks like a freeze.

The rules that come out of a Phase 5 reset are always stricter than the rules a government would have chosen in Phase 2. That is not a moral argument. It is a description of what reactive governance produces when infrastructure has already been built to voluntary benchmarks and the only remaining instrument is a freeze.

Dublin had the warning signs in June 2020. The €5.8 billion answer arrived in November 2021. Amsterdam had the warning signs earlier. The sequence is documented. The arc is not a prediction. It is a record.