Context and synthesis

International Experience: How Other Systems Are Managing Large-Load Growth#

The United States is not the first system to have a large-load surge collide with a slower grid, and several jurisdictions that hit the wall earlier offer a preview of where U.S. policy may head. Three European cases are especially instructive because each chose a different instrument — a connection moratorium tied to self-supply, a congestion-driven flexible-connection regime, and an efficiency-gated capacity release — and each has now lived with the consequences long enough to judge. Three further cases follow them: Australia, which is writing ride-through obligations before experiencing the events that prompted them; China, the one large system directing siting rather than pricing it; and the wider Asia-Pacific, where access to a constrained grid has become a rationed good.

Ireland — the moratorium that became a self-supply mandate#

Ireland is the clearest cautionary tale, because data centers reached a scale relative to the system that the U.S. as a whole has not. They consumed roughly 22% of Ireland's metered electricity in 2024, with EirGrid's median scenario projecting 31% by 2034 — Dublin is Europe's second-largest data-center cluster at about 1,150 MW. Facing supplyshortfall risk in the greater Dublin area, the regulator (CRU) imposed a de facto moratorium on new grid connections in November 2021. In December 2025 it replaced the moratorium with a Large Energy User Connection Policy that does not simply reopen the door: new data centers above 10 MVA must provide on-site flexible generation or storage sized to 100% of their grid connection, site in “unconstrained” locations away from congested nodes, and match 80% of annual demand with new Irish renewables within six years. In parallel, EirGrid moved on April 1, 2026 to impose fault-ride-through obligations on large demand facilities, with a two-year derogation for existing sites — the same ride-through discipline Section 4 describes, arriving in Ireland on essentially the same schedule as ERCOT's NOGRR282. Notably, Irish government policy explicitly rejects fully islanded, fossil-powered data centers as out of line with climate goals, closing the very off-grid pathway Section 8 examines.

The Netherlands — congestion, and the right to a flexible connection#

The Dutch grid suffers the worst congestion in Western Europe, and the response has redefined what a grid connection means. TenneT has declared effectively zero headroom for large new connections across much of NoordHolland into the mid-2030s, with connection queues exceeding 5 GW and a national grid-upgrade program measured in the hundreds of billions of euros that will not fully relieve pressure before the 2030s. Rather than queue everyone, the Netherlands has shifted from a “right to connection” toward a right to a flexible connection. Four instruments carry that shift. It prices 24/7 firm capacity as a premium product, requires large loads to participate in congestion management, and offers off-peak and non-firm connection contracts — TenneT allocated 9 GW of off-peak capacity against more than 70 GW of applications. It also contracts batteries as “congestion mitigators.” This is the clearest real-world implementation of the flexible/non-firm service that Section 6 identifies as a leading U.S. lever — born of necessity rather than choice.

Singapore — the efficiency-gated capacity release#

Singapore, land- and power-constrained, imposed an outright moratorium on new data centers in 2019, then reopened deliberately. A 2022 pilot allocation came first, followed by the May 2024 Green Data Centre Roadmap. The roadmap unlocks at least 300 MW of new capacity but gates it on efficiency and green-power conditions: a PUE target below 1.3 and at least 50% green power for priority access, with faster approvals and earmarked grid allocations for the best performers. Singapore's model treats grid capacity as a scarce public resource rationed by efficiency, a different philosophy from the U.S. market-clearing approach but one increasingly echoed in the efficiency and waste-heat mandates appearing in European law.

Australia — writing the ride-through rule before the event#

Australia matters to this report for a reason none of the European cases supply: it is addressing the Section 4 problem prospectively, and citing the American record as its justification. In March 2026 the Australian Energy Market Commission published a draft rule creating technical access standards for large inverter-based loads, raising the threshold that triggers them from 5 MW to 30 MW and requiring covered facilities to remain connected through voltage and frequency disturbances and to restore demand in a controlled manner afterwards. Consultation closed on 7 May 2026 with a final rule expected mid-year. The Commission’s stated reasoning cites the July 2024 Virginia event by name — sixty data centres removing about 1,500 MW during a single fault — alongside incidents in Ireland and Texas.

The queue evidence is equally familiar. AEMO disclosed a transmission connection pipeline for the first time in its March 2026 quarterly report: eleven projects above 5 MW totalling 5.4 GW of maximum demand, about 60% in New South Wales and 40% in Victoria, against a national fleet that consumed roughly 3.9 TWh in FY25, near 2% of grid-supplied electricity. New South Wales alone lists 44 projects in development totalling 11.4 GW, while industry expects about 1.2 GW to actually energise in Sydney by 2030 — close to a tenfold ratio between requested and expected load, and the term used locally for the difference is the same one this report uses. AEMO now models data centres as a separate demand category and describes them as active grid participants rather than as load, which is the modelling consequence of Section 4 stated in planning terms.

China — directing siting instead of pricing it#

China is the only large system attempting to solve the problem by allocation rather than by price, which makes it the useful contrast case even though its instruments do not transfer. Installed data-centre capacity stood near 32 GW at the end of 2025 and is expected to reach about 40 GW by the end of 2026, on a path to more than 60 GW by 2030. Consumption should roughly double to around 289 TWh by 2030, taking the sector from about 1.2% of national electricity demand to 2.3%. That remains a smaller share than in the United States, on a system nearly three times the size, growing at 19% compound after 38% over the preceding five years.

The organising instrument is the East Data West Computing programme, which designates eight national computing hubs in the western provinces and ten clusters in the east, directing latency-tolerant workloads toward the west where wind and solar resources are concentrated and reserving the east for latency-sensitive services. Data-centre development is a named priority of the 15th Five-Year Plan covering 2026 to 2030. The conditions attached are binding rather than aspirational: new large facilities must achieve a power usage effectiveness below 1.25 and projects inside the national hubs below 1.2, against a national average target under 1.5, and the 2025 green data centre action plan requires new projects in the hubs to source at least 80% of their energy from renewables.

Two observations follow for a reader of this report. Directed siting dissolves the queue-integrity problem of Section 1 by removing the queue: a developer does not request interconnection at fifteen points to preserve optionality when the location is assigned. What it removes with it is the information a queue carries, since an administratively chosen site reveals nothing about where demand would have gone. And the mechanism most Chinese operators use to meet the renewable requirement is green electricity certificate procurement, which raises precisely the accounting question set out in Section 8 — an annual certificate does not establish that the facility ran on renewable power in any given hour.

Japan, Korea, Singapore, Malaysia and India — access as a rationed good#

The wider Asia-Pacific pattern is consistent enough to state briefly. Japan, India and South Korea have begun steering data-centre development toward regions with stronger grid availability and lower-carbon supply, while Singapore, Malaysia and South Korea have turned to regulatory frameworks that ration access to constrained systems outright. Reporting from the region also indicates grid-support obligations moving toward the Australian model, including ride-through duties on facilities above 30 MW. Where geography constrains the network and no unconstrained alternative region exists, these systems converge on administrative allocation rather than queue position — the same destination as the European cases above, reached from a different direction.

Table I1 — What other systems did, and what transfers. Five jurisdictions reached the same constraint earlier and answered it differently. The final column is the point: the instruments travel, the institutional conditions that made them work often do not.

Sources: CRU (Ireland); ACM and TenneT (Netherlands); IMDA and EMA (Singapore); AEMC and AEMO (Australia); Chinese government programme documents.

Every one of these systems ended up somewhere the U.S. debate now heads: making firm grid access conditional — on self-supply (Ireland), on flexibility (Netherlands), or on efficiency (Singapore) — rather than an entitlement. The U.S. advantages are its scale, its spare land, and its deep capacity markets; its disadvantage is the fragmented authority of Section 7, which makes a coherent national conditionalaccess policy far harder to write than in a single-regulator system. The lesson is less any specific rule than the direction of travel: as large-load growth outruns the grid, “connect and we will build for you” gives way everywhere to “connect on conditions,” and the systems that wrote those conditions early suffered less disruption than those that improvised late.

The comparison in numbers#

The instruments described above are responses to very different magnitudes, and the table below sets the six markets on comparable terms. Two columns deserve caution. Installed capacity and consumption are reported on different bases by different authorities — metered consumption in Ireland, contracted capacity in Australia, installed gigawatts in China — so the rows should be read down rather than across. And the final column is forecast rather than record: it is included because policy in each jurisdiction answers the projection rather than the present, but every figure in it carries the fivefold uncertainty that Section 1 describes.

Table I2 — Data-centre development by market. Latest reported consumption, share of national electricity, fleet and pipeline, and forecast. Forecast figures are projections by the cited bodies, not commitments.

Sources: IEA, Energy and AI; LBNL, 2024 US data-centre report and 2025 Update; Rystad; Ireland Central Statistics Office; AEMO; AEMC.

Read down the share column and the governing pattern appears. The instrument each jurisdiction reaches for tracks the share of its own system that data centres occupy, not the absolute size of the fleet. Ireland at 23% imposed a moratorium and then a self-supply condition. Australia at about 2% is writing technical performance standards. The United States at roughly 4% is arguing about tariffs and cost allocation. China at 1.2% is directing siting for industrial-policy reasons rather than reliability ones. The concentration figures explain why: national shares understate the problem everywhere, since 26% in Virginia and 79% in Dublin are the numbers the operators of those systems actually face. A jurisdiction acts when the local share becomes unmanageable, whatever the national average shows — which is the same conclusion Table D1 reaches for the United States, arrived at from six directions.