EU EED, Singapore DC-CFA2 and the US state read-across: efficiency as procurement constraint

TL;DR · EU EED + cross-jurisdictional read-across

• Efficiency is migrating from ESG preference to procurement constraint. The EU Energy Efficiency Directive (EED) data-centre trajectory (mandatory reporting → rating scheme → potential minimum performance standards) is the leading edge. Singapore's DC-CFA2 already turns PUE ≤ 1.25 into an allocation gate. Specific US state jurisdictions are following. The regulatory arbitrage window is narrowing first in Singapore and the EU while the US remains fragmented; a Singapore-style allocation gate has not yet spread to EU or US markets, but the direction is clear.
• The EU is progressing from reporting to a full procurement framework, not merely contemplating one. Recast Directive 2023/1791 (in force October 2023) requires reporting for data centres above 500 kW IT installed. Delegated Regulation (EU) 2024/1364 formalised the first-phase Union rating scheme. In March 2026 the Commission published a draft regulation for feedback on the second-stage rating scheme, explicitly framing it as intended to inform procurement of more sustainable digital assets and services. The Q3 2026 Data Centre Energy Efficiency Package is expected to include the second-stage rating Delegated Act plus launch of minimum-performance-standard work covering both new and existing data centres, with adjacent parameters (grid integration, flexibility, clean energy sourcing, waste-heat recovery, siting). The EU is not yet Singapore — there is no single PUE gate — but the reporting → rating → procurement pressure → minimum-standard sequence is already unfolding.
• Singapore DC-CFA2 is the sharpest current example of efficiency as procurement gate. The Infocomm Media Development Authority (IMDA) DC-CFA2 allocation framework (following the earlier 2019-2022 moratorium) requires PUE ≤ 1.25 at 100% IT load, plus green-energy commitments and IT-efficiency requirements. Winning an allocation slot depends on the efficiency spec. Vendor selection reprices around efficiency-first specifications.
• US state jurisdictions are developing a heterogeneous mosaic. Virginia, Oregon, Arizona, California and adjacent utility-territory frameworks are developing a mosaic of utility, permitting, water and grid-capacity constraints that increasingly make efficiency part of site economics. There is no coherent national procurement gate equivalent to Singapore DC-CFA2; the direction is toward more scrutiny, but the mechanisms and timing differ substantially by state.
• Vendor implications concentrate on UPS topology, PDU efficiency and thermal integration. Higher-efficiency UPS topologies (double-conversion online with ECO-mode capability, 800V DC power-rack / BBU architectures per V2/V3 — see P4), higher-efficiency PDU networks and tighter cooling-electrical co-design capture regulatory-preference share.
• Schneider Electric and ABB have the European position; Vertiv and Eaton compete through EU subsidiaries; Delta is growing EU share. European vendors with mature high-efficiency product families capture the EED-driven vendor selection. Non-European vendors compete through EU subsidiaries and local manufacturing capacity.
• F-gas is cooling, not P4. The revised EU F-gas Regulation (Regulation 2024/573, in force March 2024) governs SF6 in MV switchgear and refrigerants in cooling systems. Relevant to P3 site substation switchgear (see P3) and to hall cooling systems. Not directly a P4 UPS / PDU obligation.
• The killer thesis: regulatory arbitrage on efficiency has a closing window. Operators who lock in efficient designs now capture the regulatory-preference share in Singapore, EU and specific US state jurisdictions. Operators who wait face incremental compliance and permitting cost when the rating scheme and minimum performance standards bite.

Framework anchor

Regulatory overlay · procurement constraint. Part of the hall-storage investment stack: this piece covers efficiency as procurement constraint; ride-through storage covers battery economics; monetizable storage covers the hall-BESS revenue stack. See the hall-storage investment stack hub for the three-panel view.

This piece is a standalone deep-dive on the regulatory efficiency framework touching P4 (Hall Power). Cross-refers to P4 · Hall Power for the UPS / PDU / BESS architecture context, P3 · Campus Power for the F-gas Regulation on MV switchgear, and P2 · Grid Access for the ISO/RTO and utility interconnection framework that also feeds efficiency constraints.

01What the EU EED actually requires

Efficiency is migrating from an ESG-preference layer to a procurement-gate layer. The regulatory shift matters when efficiency stops being something the operator reports and becomes something the buyer, allocator or grid uses to decide whether the project gets built. The regulatory arbitrage window is narrowing first in Singapore and the EU while the US remains fragmented. Operators who lock in efficient designs now capture the preference share; operators who wait pay compliance and permitting cost when the standards bite.

And the constraint is becoming multi-dimensional. The EU 2026 Data Centre Energy Efficiency Package addresses grid integration, flexibility, clean energy sourcing, waste-heat recovery and siting alongside straight energy-efficiency metrics. The 2028+ procurement attribute moves past a single PUE threshold; it is a bundle across energy + water + carbon + grid + flexibility + heat recovery + siting. That is much harder to arbitrage than a single-metric threshold.

The EU Energy Efficiency Directive was recast as Directive (EU) 2023/1791 in force October 2023, updating and strengthening the 2012 Directive. Article 12 of the recast Directive introduces obligations specific to data centres above a defined threshold. The mechanics matter for anyone specifying hall power infrastructure destined for EU deployment through the 2024-2028 phase-in.

The reporting layer (in force)

Data-centre operators in the EU with an installed IT power of at least 500 kW must publish annually a set of energy performance and sustainability indicators. The specific indicators are defined by delegated acts adopted by the European Commission and cover power usage effectiveness (PUE), water usage effectiveness (WUE), energy reuse factor (ERF), renewable-energy factor (REF) and other metrics. Reporting is via the EU database framework; publication is public.

The reporting obligation entered force through 2024 with the first reporting cycle covering 2024 operations reported in 2025. The framework aligns with the earlier EU Code of Conduct on Data Centres (voluntary) but is now mandatory for the covered population.

The rating scheme (in development)

The EU rating scheme runs beyond speculative. The first phase was formalised through Delegated Regulation (EU) 2024/1364. In March 2026 the Commission published a draft regulation for feedback on the second-stage rating scheme, explicitly framing the scheme as intended to inform procurement of more sustainable digital assets and services. The Q3 2026 Data Centre Energy Efficiency Package is expected to include the second-stage rating Delegated Act, launch a public consultation for minimum-performance standards for both new and existing data centres, and address broader parameters including grid integration, flexibility, clean energy sourcing, waste-heat recovery and siting. The rating scheme therefore creates real procurement-preference pressure now, not only in a future scenario.

The minimum performance standards (potential future)

The Commission has flagged that minimum performance standards for new-build data centres may follow the rating scheme. If adopted, minimum standards would function like the EU Ecodesign Directive for other product categories: a threshold below which new-build data centres cannot be placed on the market. The specific threshold levels and applicability timing remain under development; the direction of travel is clear.

The obligation trajectory in one picture

Fig 1

EU EED data-centre obligation trajectory 2023-2030

Directional timeline; specific delegated-act adoption dates are subject to Commission scheduling. The direction and sequencing are established.

02Singapore DC-CFA2: the sharpest current example

Singapore is the leading current example of efficiency as an allocation gate (reporting-layer treatment stays weaker).

The Infocomm Media Development Authority (IMDA) operated a data-centre moratorium from mid-2019 through 2022 in response to constrained power availability and land-use concerns. The moratorium was lifted with the introduction of the Data Centre Call for Application 1 (DC-CFA1) in 2022, which awarded allocation slots to a small number of operators against specific criteria including PUE, green-energy commitment and IT-efficiency capability.

DC-CFA2, launched in 2024, tightened the criteria. Applicants must demonstrate PUE ≤ 1.25 at 100% IT load, green-energy sourcing commitments (aligned with Singapore's national decarbonisation goals), and IT-efficiency requirements covering the accelerator and rack configuration. Singapore moves past regulating equipment efficiency; it is rationing scarce data-centre capacity through an efficiency criterion. That is the real significance and the reason DC-CFA2 rests at the sharp end of the global efficiency-as-procurement-gate spectrum. The mechanism turns efficiency from an ESG preference into a procurement constraint at the site-allocation stage.

Why DC-CFA2 matters as a template. The Singapore framework demonstrates how allocation policy can turn efficiency into a procurement constraint. The operators winning DC-CFA2 slots have specified hall power infrastructure to meet the PUE ≤ 1.25 spec: higher-efficiency UPS topologies, higher-efficiency PDU networks, tighter cooling-electrical co-design, and (increasingly) 800V DC architecture (see P4) where MGX-compatible racks can deliver higher rack density with reduced conversion losses. The Singapore template is being watched by regulators in EU member states, US state PUCs and other Asian jurisdictions considering allocation-based frameworks.

03US state jurisdictions: the mosaic read-across

The US does not have a federal data-centre efficiency framework. State-level constraints are emerging as a heterogeneous mosaic of utility, permitting, water and grid-capacity mechanisms. The direction of travel is toward more scrutiny of data-centre efficiency, but there is no coherent national procurement gate equivalent to Singapore DC-CFA2. The case studies below are illustrative reads of the mosaic, not asserted statewide procurement rules; each requires site-level and docket-level diligence for underwriting to a specific project.

Virginia

The State Corporation Commission (SCC) has approved specific tariff structures for hyperscaler-linked loads; specific dockets and Dominion Energy tariff filings for AI-buildout loads have referenced efficiency-related considerations. Direct efficiency-conditioning of ratebase treatment remains developing (settled framework stays absent); specific outcomes run docket-by-docket, not statewide. Dominion Energy's tariff filings for AI-buildout loads have referenced efficiency thresholds. Localities within Virginia (including Prince William County and Loudoun County) have introduced site-permitting conditions covering energy efficiency, water use and cooling technology. The specific thresholds are locality-dependent.

Oregon

The Oregon PUC and specific utility service territories (particularly Portland General Electric and PacifiCorp) have introduced conditions on hyperscaler load service. Efficiency reporting is part of the framework in specific tariff structures.

Arizona

Arizona Corporation Commission oversight of specific utility service territories, combined with local water-use restrictions in Phoenix-area jurisdictions, creates efficiency pressure that runs beyond primarily electrical (water use is the binding constraint on some sites) but still drives vendor selection.

California

California Energy Commission and CPUC oversight of data-centre efficiency has been evolving. Title 24 building code applies to data centres; the state's climate framework and CPUC decisions on utility-served hyperscaler load together drive efficiency requirements. California operates a different set of constraints from Texas or Virginia; efficiency plays a larger role.

Texas

Texas PUC oversight of ERCOT-served hyperscaler load is comparatively less efficiency-focused than California, Virginia or Oregon. Growth runs capacity-constrained; efficiency-constrained framing understates the gating factor; the Texas mosaic emphasises grid interconnection over efficiency mandates. But state-level and locality-level frameworks continue evolving.

Fig 2

Jurisdictional efficiency framework comparison 2026

Comparative directional read of how efficiency requirements enter each jurisdiction's data-centre framework. Vertical axis is directional; treat as a structural comparison; a precise ranking stays absent.

04Vendor implications

Higher-efficiency UPS topologies, higher-efficiency PDU networks, tighter cooling-electrical co-design and 800V DC-ready product lines capture the regulatory-preference share in EU, Singapore and specific US state jurisdictions. Vendors with mature high-efficiency portfolios lead; vendors developing catch up.

UPS topology: double-conversion online with ECO-mode

The double-conversion online UPS topology typically operates at high 90s efficiency at hyperscaler load (typical published ranges vary by vendor, load and operating mode). ECO-mode capability (also called line-interactive or eco-conversion depending on vendor) allows the UPS to run at higher efficiency during utility conditions that permit it, switching back to full double-conversion during disturbances. Vendors with mature ECO-mode capability in their hyperscaler-class products capture regulatory-preference share.

The V2 hybrid 800V DC architecture (see P4) introduces DC UPS / BBU on the 800V bus, which changes the efficiency calculation. Fewer conversion stages between grid and rack means lower cumulative loss. Vendors positioned for both AC ECO-mode and 800V DC topologies capture the full transition-window preference.

PDU networks: higher-efficiency variants

PDU energy losses are typically small relative to UPS losses but compound across the campus. Higher-efficiency PDU vendors (Vertiv Geist, Starline, Anord Mardix, Legrand) with mature high-efficiency product families capture preference on efficiency-first specifications. 800V DC PDU / busway variants (for V3 row-level 800V, see P4) are a distinct product category.

Cooling-electrical co-design

PUE is dominated by cooling energy load. Vendors offering integrated cooling-electrical product families (Vertiv thermal management, Schneider Electric EcoStruxure, and adjacent) capture value from tighter co-design. Direct liquid cooling and immersion cooling adoption changes the electrical specification and the PUE math simultaneously; efficiency-driven procurement favours vendors capable of both sides.

Regional vendor positioning

Tier 1

European vendors with mature high-efficiency portfolios

Schneider Electric. French-headquartered global vendor with APC data-centre brand history. Galaxy UPS product family with ECO-mode capability. EcoStruxure integrated cooling-electrical platform. Strong European hyperscaler installed base. Positioned to capture EU EED-driven vendor selection.

ABB Electrification. Swiss-Swedish heritage; strong European presence. High-efficiency UPS product families including DC UPS variants. MV / electrical infrastructure adjacency (see P3).

Tier 2

Non-European vendors competing through EU subsidiaries

Vertiv. US-headquartered data-centre pure-play. EU subsidiary operations serve European hyperscalers. Product families (Trinergy / Liebert / EnergyAware) include high-efficiency variants; EU-manufactured products meet EU-compliance requirements.

Eaton. US-headquartered diversified electrical + power. Strong European subsidiary footprint. Product families include high-efficiency UPS. Cross-sell into broader P3 to P4 electrical stack.

Tier 3

APAC vendors growing EU share

Delta Electronics. Taiwan-headquartered global vendor. Growing European share through the 2024-2026 window. High-efficiency UPS + 800V DC power-rack capability (concrete V2 evidence per Nvidia GTC 2026 — see P4).

Mitsubishi Electric, Toshiba, Hitachi. Japanese vendors with mature high-efficiency UPS product families for domestic and APAC markets; European presence growing but smaller than the Tier 1 European incumbents.

05What the EED does not directly require

Worth being precise about scope. The EU EED does not directly mandate:

• Specific UPS topologies. The Directive requires reporting of PUE, WUE and adjacent metrics; it does not specify how those metrics are achieved. Vendor selection is indirect (through operator response to reporting and rating pressure), not direct (through Directive text specifying a topology).
• Specific chemistry choices for UPS batteries. Lithium vs VRLA is not directly addressed by EED; the EU Battery Regulation (Regulation 2023/1542) handles battery-specific obligations (due diligence, carbon footprint, recyclability).
• Refrigerants and F-gas. The revised F-gas Regulation (Regulation 2024/573, in force March 2024) governs SF6 in MV switchgear and high-GWP refrigerants in cooling systems. Relevant to P3 (site substation switchgear) and to hall cooling systems (P4-adjacent but separate). Not directly a P4 UPS / PDU obligation.
• Renewable-energy sourcing. The EED requires reporting of the renewable-energy factor; specific procurement mandates come from adjacent EU frameworks (Renewable Energy Directive, national implementations, EU ETS carbon pricing).

06 The regulatory maturity ladder: REPORT > RATE > PROCURE > STANDARDISE > EXCLUDE

Data-centre efficiency regulation follows a repeatable five-stage progression across jurisdictions. Each stage moves the economic consequence of inefficiency closer to the investment decision.

1. REPORT. "You must tell us." Reporting cost only; no direct procurement consequence.
2. RATE. "We compare you." Rating-scheme comparison; reputational and customer-selection pressure.
3. PROCURE. "Efficiency ratings become relevant to purchasing / allocation decisions." Direct procurement-preference pressure; buyers cite ratings.
4. STANDARDISE. "You cannot operate below the threshold." Minimum performance standards excluding below-threshold designs from market placement.
5. EXCLUDE. "Inefficient designs eventually lose access to scarce capacity, approvals or economically viable sites." Efficiency becomes site viability itself — the endpoint where the operator without an efficient design loses the deal entirely.

Singapore currently operates at stage 5 for capacity allocation (DC-CFA2 as an efficiency-conditioned scarce-resource allocation). The EU is at stages 1 through 3, moving to 4 through the Q3 2026 Package and preparatory minimum-standard work. The US mosaic occupies heterogeneous combinations of stages 1, 2 and 5 depending on jurisdiction and mechanism.

The prior five-stage framework, mapped

Efficiency regulation follows a repeatable sequence across jurisdictions. The stages run with soft boundaries and jurisdictions can sit at multiple stages simultaneously across different mechanisms; the sequence is a useful diagnostic for underwriting where a given jurisdiction stands and where it is heading.

The investor question at each stage: at what point does efficiency move from disclosure to economic exclusion? Stages 1-2 create reputational and preference pressure without direct exclusion. Stage 3 excludes at allocation. Stage 4 excludes at market placement. Stage 5 excludes at specific-site level. Vendors positioned on the leading efficiency frontier capture the transition at each stage; vendors that don't face relative share loss as jurisdictions climb the stages.

07 Singapore DC-CFA1 and DC-CFA2 detail

DC-CFA1 (2022)

The initial call for application after the moratorium period. IMDA awarded allocation slots to a small named group of operators against criteria including PUE, green-energy commitment and IT-efficiency capability. Public disclosures at the time named specific operator awards (a limited slate); the aggregate allocation size was constrained relative to industry demand, establishing the baseline that Singapore data-centre capacity runs rationed, with demand-served treatment ending.

DC-CFA2 (2024)

Tightened criteria: PUE ≤ 1.25 at 100% IT load; green-energy sourcing commitments; IT-efficiency requirements covering accelerator and rack configuration. The green-energy commitment aligns with Singapore's national decarbonisation goals (renewable power sourcing, PPA structures, adjacent grid-services). IT-efficiency criteria include accelerator-per-rack density and cooling-technology preferences.

DC-CFA3 and beyond (developing)

Whether Singapore's next allocation round tightens further, holds at DC-CFA2 threshold, or introduces new criteria (embodied carbon, water-use effectiveness, alignment with 800V DC architecture progression) is a real underwriting variable. Watch IMDA disclosures on next-round criteria; each tightening compresses the eligible-operator set.

08 US state jurisdictional case studies

Each of the below is a case study of the mosaic, not an asserted statewide procurement rule. Underwriting a specific project requires site-level and docket-level diligence.

Virginia case study: SCC + Loudoun / Prince William permitting

Virginia hosts the largest US data-centre cluster (Ashburn / Loudoun County + adjacent Northern Virginia territories). Constraints have been developing along three tracks: State Corporation Commission (SCC) tariff and ratebase decisions for Dominion Energy's hyperscaler-linked load service; Loudoun County zoning and permitting conditions on new-build sites (including specific setback, water-use and cooling-technology conditions); Prince William County adjacent zoning and permitting. Grid-connection capacity is a binding constraint that drives site-selection separately from efficiency; the two intersect where efficiency-driven cooling-architecture choices affect water-use and permitting pathways.

California case study: CEC + CPUC + Title 24

California's data-centre efficiency framework combines Title 24 building energy code (which includes data-centre specific provisions and has been updated across successive triennial code cycles), California Energy Commission (CEC) load-management and efficiency oversight, and CPUC utility-rate structures for data-centre load service. Water-use and climate-related state frameworks add additional layers. California's data-centre growth has been slower than Virginia / Oregon / Arizona partly because efficiency requirements combined with power-cost and land-cost dynamics produce different site economics.

Oregon case study: PGE + PacifiCorp service territories

Oregon's data-centre cluster (Hillsboro / adjacent Portland General Electric territory; Prineville / adjacent PacifiCorp territory) operates under tariff structures that include efficiency-reporting provisions in specific hyperscaler load agreements. PUC oversight is at the tariff-approval level; an EED-style national framework stays absent.

Arizona case study: water-first constraints

Arizona (particularly Phoenix-area jurisdictions) hosts significant hyperscaler data-centre deployment; the binding constraint often comes from water availability; direct electrical efficiency plays a smaller role. Water-use restrictions in specific municipalities drive cooling-technology choices (favouring air-cooled or hybrid approaches over water-intensive cooling); vendor-selection consequences are indirect but real.

Texas case study: capacity-constrained, less efficiency-focused

Texas ERCOT-served hyperscaler load growth is comparatively less efficiency-focused than California / Virginia / Oregon. Growth runs capacity-constrained; efficiency-constrained framing understates the gating factor; the Texas mosaic emphasises grid interconnection and generation-side firm capacity over efficiency mandates. Site-level and locality-level frameworks continue evolving but currently do not aggregate to an efficiency-first procurement gate.

09 UPS efficiency standards depth

IEC 62040 (UPS standards family)

The IEC 62040 series governs UPS product design, testing and performance. Includes 62040-3 (performance and test requirements), 62040-4 (environmental aspects) and adjacent parts. UPS efficiency measurement and reporting under this framework produces the vendor-published efficiency numbers that feed into procurement comparisons.

ENERGY STAR Enterprise Server and adjacent

The US ENERGY STAR programme covers enterprise servers and adjacent data-centre equipment categories. Data-centre facility level does not currently have a single ENERGY STAR certification, but the equipment-level programmes drive hyperscaler procurement preferences for IT equipment. The equipment-side efficiency requirement flows into the facility-side PUE math.

ISO / IEC 30134 (data-centre KPIs)

The ISO / IEC 30134 series defines data-centre performance KPIs including PUE (30134-2), REF (30134-3), WUE (30134-9) and adjacent. These are the metric definitions that jurisdictional reporting frameworks (including EU EED) reference; consistency of definition across jurisdictions matters for cross-comparability of reported metrics.

Ecodesign Directive parallel

The EU Ecodesign Directive (Directive 2009/125/EC) applies to energy-related products and would provide the framework model if EU EED minimum performance standards for data centres are eventually adopted. Ecodesign has been used across categories from consumer electronics to industrial equipment; extension to data-centre facilities would follow a similar structural pattern.

Architecture-endpoint reference. SemiAnalysis' multi-part "Inside the 800VDC Revolution" covers the architectural endpoint that efficiency regulation acts on: centralised rectification, MV UPS positioning and the 800V DC transition inside the hall. Regulatory minimum-performance standards (Ecodesign parallel, IEC 62040 series, ENERGY STAR, ISO/IEC 30134) act on the same equipment layer SemiAnalysis maps; this piece covers how the regulatory ladder (REPORT > RATE > PROCURE > STANDARDISE > EXCLUDE) alters procurement gates upstream of that equipment.

10 PUE math walk-through

PUE = Total facility energy / IT-equipment energy. A hall drawing 10 MW total facility energy to deliver 8 MW to IT equipment has PUE 1.25. The 2 MW gap covers cooling (typically the largest component), UPS and PDU losses, lighting, security systems, and adjacent auxiliary loads.

Cooling typically dominates the non-IT energy load: a 10 MW facility might see 1.2-1.5 MW for cooling (depending on climate, cooling architecture and rack density), 0.15-0.25 MW for UPS + PDU losses, and 0.1-0.2 MW for adjacent auxiliary loads. Driving PUE toward 1.25 or below requires optimising primarily on cooling, secondarily on UPS / PDU efficiency, and tertiarily on adjacent loads. The order of magnitude explains why cooling-architecture choices dominate the PUE improvement curve, and why 800V DC (which reduces the UPS / PDU loss layer) captures a smaller but real share of the PUE improvement.

The 800V DC PUE arithmetic

Fewer conversion stages between grid and rack reduces cumulative conversion loss. The V2 hybrid 800V DC architecture (see P4) reduces the AC-to-DC-to-DC path for AI-workload racks; V4 native 800V DC further reduces facility-level conversion. Directional PUE improvement from architectural conversion-stage reduction is smaller in absolute terms than cooling-architecture improvement (immersion / direct liquid cooling can shift PUE by 0.05-0.10 depending on baseline; conversion-stage reduction typically shifts PUE by 0.01-0.03) but combined they support jurisdictional efficiency requirements at the leading edge.

11 Renewable-energy sourcing overlay

Efficiency (PUE, WUE, ERF) and renewable-energy sourcing (REF) are related but distinct regulatory dimensions. The EED requires reporting of both; Singapore DC-CFA2 has both PUE and green-energy commitment criteria. Vendor selection at P4 (UPS, PDU, BESS) runs primarily driven by efficiency requirements; renewable-energy sourcing runs primarily driven by upstream P1 contracting (see P1 · Electrons) — the specific PPA structure, on-site generation, virtual PPA / EAC procurement, and adjacent frameworks.

The intersection: a hyperscaler operator meeting jurisdictional efficiency criteria at P4 without matching renewable-energy sourcing at P1 may still fail an allocation-gate framework where green-energy commitment is a separate criterion. Singapore DC-CFA2's green-energy commitment runs alongside PUE; failing either criterion loses the allocation. Watch for jurisdictions adopting combined efficiency + renewable-sourcing gates (potential future direction in EU EED minimum performance standards and adjacent frameworks).

12 Vendor scorecard

Analyst scores — directional underwriting judgement, not measured market shares. Scores are qualitative judgements on a 1-10 scale on each dimension, equally weighted for the composite. Inputs are our reading of vendor product-portfolio disclosures, EU / Singapore project wins and industry reporting. Treat as a structured comparison forcing explicit judgement, not as market-share metrics.

Schneider Electric leads on the composite reflecting French headquarters + APC / EcoStruxure European installed-base depth + cooling-electrical co-design capability. Vertiv is close behind on data-centre pure-play focus and high-efficiency UPS portfolio depth. Delta rates higher on 800V DC readiness (from GTC 2026 evidence — see P4) but lower on EU installed-base depth. ABB and Eaton sit in the middle. Non-European vendors compete through EU subsidiaries; European incumbents lead on regulatory-preference channel.

13 Decision tree for a regulatory-driven vendor screen

Question 1. Which jurisdiction is the target project in? Singapore, EU (which member state), US (which state), UK, Australia, other.

Question 2. Which stage of the five-stage regulatory progression applies? Reporting-only (1), rating-active (2), allocation-gated (3), minimum-standard-gated (4), or utility-permitting-gated (5)?

Question 3. Does the target project fall within the covered population (e.g., above 500 kW IT installed for EU EED; within the DC-CFA2 allocation round for Singapore; within specific US utility service territory)?

Question 4. Which specific efficiency criteria bind (PUE threshold, WUE threshold, REF requirement, IT-efficiency requirement)?

Question 5. What is the vendor's product-family compliance readiness (high-efficiency UPS topology, PDU efficiency, cooling-electrical co-design, 800V DC option) for the specific project spec?

Question 6. What is the timeline sensitivity? A project contracting today for delivery in 2028 may cross regulatory thresholds mid-build if the jurisdiction climbs stages during construction.

14 Investment contests

Contest 1. EU EED rating-scheme design and vendor response

How the Commission's delegated acts formalise the rating scheme determines which specific metrics hold the most vendor-selection weight. Vendors with mature product families across the leading metrics (PUE, WUE, ERF, REF) capture disproportionate preference share once the rating scheme is operational.

Contest 2. Singapore DC-CFA3 and beyond

Whether Singapore's next allocation round (DC-CFA3 or successor) further tightens the PUE and green-energy thresholds. Tighter thresholds compress the eligible-operator set and favour vendors positioned on the leading-edge efficiency spec. Watch IMDA disclosures on next-round criteria.

Contest 3. US state jurisdictional convergence or divergence

Whether US state efficiency frameworks converge on common metrics (making a coherent mosaic) or diverge (creating jurisdiction-specific compliance overhead). Convergence favours vendors with jurisdiction-agnostic high-efficiency products; divergence favours vendors with strong per-state relationships.

Contest 4. Federal action in the US

Whether US federal action on data-centre efficiency (through DOE, EPA or Executive Order) creates a national framework analogous to EED. Federal action would compress state-level divergence and create a single national efficiency baseline; absence of federal action preserves state-level mosaic.

Contest 5. Adjacent APAC and MENA framework adoption

Whether Japan, South Korea, India, UAE and Saudi Arabia adopt Singapore-style allocation-based frameworks or EU-style rating schemes. Adoption in any of these jurisdictions expands the regulatory-preference vendor-selection window beyond EU + Singapore + US states.

15 Falsification triggers

Falsification trigger 1

EU EED rating scheme delayed beyond 2028. If delegated acts on the rating scheme slip materially, the transition from reporting to rating-driven procurement preference compresses and the regulatory-arbitrage window stays open longer.

Falsification trigger 2

Singapore reverses or loosens DC-CFA2. Any change to the PUE ≤ 1.25 threshold or the green-energy commitment framework at Singapore would reduce the strongest current template for allocation-based efficiency. Watch IMDA policy signals.

Falsification trigger 3

US state efficiency frameworks stall at reporting only. If US state PUCs and localities remain at reporting-only frameworks without moving to permitting conditions or tariff constraints, the US read-across weakens.

Falsification trigger 4

800V DC transition captures efficiency gains through architecture, not vendor product-family upgrades. If the V2 / V3 / V4 800V DC transition (see P4) captures the efficiency improvements primarily through architectural changes; vendor product-family upgrades carry less of the lift, the efficiency-first vendor-preference thesis compresses relative to the 800V-transition thesis.

Falsification trigger 5

Renewable-energy sourcing becomes the binding constraint; PUE recedes. If regulator emphasis shifts from energy-efficiency metrics to renewable-energy procurement, vendor selection runs less driven by UPS / PDU topology and more by P1 upstream PPA structure (see P1).

16 What would change my mind

What would falsify the efficiency-as-procurement-constraint thesis? Six distilled tests.

1. EU EED rating scheme slips materially or gets watered down at delegated-act stage. The regulatory arbitrage window stays open longer than expected.
2. Singapore DC-CFA2 reverses or loosens the PUE threshold. The strongest current allocation-gate template weakens.
3. US state efficiency frameworks stall at reporting-only. The US read-across from EU / Singapore weakens; regulatory arbitrage on efficiency persists in the US.
4. 800V DC architectural transition delivers efficiency gains that make vendor product-family differentiation less relevant. The V2 / V3 / V4 architecture capture displaces the vendor-preference thesis.
5. Renewable-energy sourcing becomes the binding constraint rather than PUE / WUE. Vendor-selection weight shifts from P4 UPS / PDU efficiency to P1 upstream PPA structure.
6. Efficiency metrics diverge across jurisdictions faster than they converge. Compliance overhead compresses vendor cross-jurisdictional efficiency advantage.

17 The 24-month watch calendar

2026-H2EU EED first reporting cycle results. Reported PUE / WUE / ERF / REF numbers for EU data centres above 500 kW IT installed power.2026-2027EU Commission delegated-act adoption on rating scheme. Formalisation of the classification framework.2027Singapore DC-CFA3 announcement (if follow-on round announced). Any tightening of PUE threshold or green-energy commitment criteria.2027Virginia SCC and locality permitting outcomes. Named efficiency-conditioned approvals or denials on hyperscaler-linked builds.2027-2028US federal action or absence-of-action signals. DOE / EPA / Executive Order signals on national data-centre efficiency framework.2028EU EED minimum performance standards proposal. Whether the Commission formalises minimum standards following the rating scheme.2028+Adjacent APAC and MENA framework adoption signals. Japan, South Korea, India, UAE, Saudi Arabia signals on data-centre efficiency framework adoption.

18 FAQ

What is the EU EED?

Directive (EU) 2023/1791 recast in October 2023, the EU Energy Efficiency Directive updated and strengthened the earlier 2012 Directive. Article 12 introduces obligations specific to data centres above 500 kW IT installed power: mandatory reporting of energy performance and sustainability indicators (PUE, WUE, ERF, REF among others), with the Commission working toward a common rating scheme and potentially minimum performance standards.

What are the specific PUE / WUE thresholds?

The EED reporting itself does not set thresholds; it requires reporting of the metrics. Threshold-setting comes through the rating scheme (in development) and potentially through minimum performance standards (potential future). Singapore DC-CFA2 already sets a threshold: PUE ≤ 1.25 at 100% IT load.

Is Singapore DC-CFA2 an allocation gate or just a reporting requirement?

An allocation gate. IMDA awards specific data-centre allocation slots against DC-CFA2 criteria including PUE ≤ 1.25, green-energy sourcing and IT-efficiency requirements. Winning a slot depends on meeting the criteria; not winning a slot means no allocation.

Which US states have the tightest data-centre efficiency frameworks?

Virginia (SCC + locality permitting), Oregon (PUC + utility service territory conditions) and California (CEC + CPUC + Title 24) have the tighter current frameworks. Arizona has water-use pressure that drives vendor selection indirectly. Texas is comparatively less efficiency-focused; growth is capacity-constrained.

Does the EED apply to hyperscaler-owned facilities in the EU or only to colocation?

The 500 kW IT installed power threshold catches both hyperscaler-owned and colocation facilities above that scale. In practice, essentially every hyperscaler-owned EU data centre falls within scope; large colocation facilities also fall within scope.

Which vendors are best positioned for EU EED-driven vendor selection?

European incumbents with mature high-efficiency portfolios (Schneider Electric with Galaxy + EcoStruxure, ABB Electrification) lead. Non-European vendors compete through EU subsidiaries (Vertiv, Eaton). Delta Electronics is growing EU share with high-efficiency UPS + 800V DC power-rack capability. Cooling-integrated product families capture cross-domain value.

Is F-gas Regulation part of the same story?

Adjacent but distinct. The revised EU F-gas Regulation (Regulation 2024/573, in force March 2024) governs SF6 in MV switchgear and high-GWP refrigerants in cooling systems. Relevant to P3 (site substation switchgear) and to hall cooling systems (P4-adjacent). Not directly an EED / UPS / PDU obligation.

How does the efficiency framework interact with the 800V DC transition?

800V DC architectures deliver efficiency gains through architectural changes (fewer conversion stages, higher-voltage distribution, integrated conversion at pod / row level). See P4 · Hall Power. Vendors positioned for both AC ECO-mode topologies and 800V DC architectures capture the full transition-window regulatory preference.

19 Closing

Efficiency is migrating from an ESG-preference layer to a procurement-gate layer across the EU, Singapore and specific US state jurisdictions. The EU EED trajectory (reporting → rating → potential minimum standards) is the leading edge. Singapore's DC-CFA2 turns efficiency into an allocation gate. US states are following through a mosaic of PUC, ratebase and permitting mechanisms.

For P4 vendor selection, the implication is concrete: higher-efficiency UPS topologies (double-conversion online with ECO-mode; 800V DC UPS / BBU), higher-efficiency PDU networks, and tighter cooling-electrical co-design capture the regulatory-preference share. European vendors with mature high-efficiency portfolios (Schneider Electric, ABB) lead the EU market; non-European vendors compete through subsidiaries. The APAC vendor cohort (Delta, Mitsubishi Electric, Toshiba, Hitachi) is growing EU share.

The regulatory arbitrage window on hyperscaler data-centre efficiency is closing. Operators who lock in efficient designs now capture the preference share; operators who wait pay compliance and permitting cost when the standards bite. That is the killer investment implication.

See also: the fourth companion, Load-side flexibility · How firm AI load becomes a flexible grid asset, covers the software-defined competitor to physical storage with structural edge on the connection-acceleration value stream. Introduces the reclassification premium framework — the economic value created when firm AI capacity acquires a defined flexibility right.