Making
Singapore's
Digital
Infrastructure
Future‑Ready

The draft Digital Infrastructure Bill establishes a dedicated licensing regime for data-centre operators. Its structure places environmental sustainability within the continuing obligations of operating regulated data-centre infrastructure rather than treating sustainability solely as a voluntary objective.

Section 33 is particularly significant. It requires a licensee, subject to requirements to be prescribed, to ensure:

• compliance with minimum facility-level energy-efficiency requirements;
• compliance by installed information technology equipment with equipment-level energy-efficiency requirements; and
• compliance with minimum facility-level water-efficiency requirements.

The legislation therefore recognises at least two distinct energy-efficiency domains:

This distinction is important.

A data centre is not merely a building that consumes electricity. Nor is it merely a collection of computers. It is a system in which energy is sourced, converted, distributed and ultimately consumed to perform computation.

The proposed regime does not appear to treat energy efficiency as a one-time design requirement.

Section 34 requires periodic reports concerning compliance with the relevant Part of the legislation, regulations, licence conditions and applicable codes of practice. The Authority may require such a report to be audited and accompanied by an auditor's report. Section 35 permits the Authority to issue or approve codes of practice relating to measures for ensuring energy and water efficiency.

The resulting regulatory structure can be represented as:

with Audit and Remediation branching from Reporting.

This is important because it creates an opportunity for evidence-based innovation. If compliance is measured and audited, the framework need not rely solely on prescribed technologies. It can potentially recognise actual performance.

The question is therefore whether the regulatory framework should provide only:

or also:

This white paper recommends both.

Codes of practice are necessary. They provide operators with certainty and allow regulators to establish minimum standards.

However, detailed technical requirements can create an unintended problem if compliance becomes inseparable from the use of particular architectures or methods.

A regulatory framework developed around current technology may inadvertently establish:

The result can be paradoxical.

A new technology may produce a better environmental outcome but face difficulty because it does not operate according to the method assumed by the existing compliance framework.

This risk is especially relevant in rapidly developing fields such as:

• artificial intelligence;
• computing architectures;
• advanced cooling;
• power electronics;
• energy storage;
• renewable-energy integration; and
• dynamic workload management.

The legislation should therefore distinguish between:

Both have legitimate roles. The progressive principle should be:

This white paper recommends that the Bill, regulations or codes of practice expressly permit an Alternative Performance-Based Compliance Pathway.

Under such a pathway, an operator or technology provider could demonstrate that an alternative system achieves:

• the required level of energy efficiency or better;
• equivalent or better security;
• equivalent or better operational resilience;
• equivalent or better reliability; and
• independently verifiable performance.

The architecture would be:

This would make the regulatory framework technology-neutral in practice, not merely in principle.

Data-centre energy consumption should be understood as a system. A simplified chain is:

Efficiency can be lost or improved at every stage. The regulatory challenge is that no single metric necessarily captures the entire chain.

For example, PUE is commonly expressed as:

This is valuable for assessing facility overhead. A PUE approaching 1 indicates that a greater proportion of facility energy is reaching IT equipment rather than being consumed by supporting infrastructure.

But consider two hypothetical systems:

If both facilities have the same PUE, PUE alone does not distinguish the computational efficiency of the two workloads. This is not a criticism of PUE. It is a recognition of its purpose. A metric should not be criticised for failing to measure something it was not designed to measure.

The policy implication is that facility metrics should be complemented, not discarded.

The conventional measurement chain may stop at:

But the economic and social purpose of a data centre lies beyond the equipment. The equipment exists to produce:

• information;
• inference;
• storage;
• communication;
• simulation;
• transactions;
• scientific results;
• digital services; or
• other computational outcomes.

The complete chain is therefore:

A progressive efficiency framework should ultimately be capable of asking:

This paper refers to that concept as Computational Outcome Efficiency. It can be expressed conceptually as:

This is not proposed as a single universal numerical metric. Different workloads have different purposes. A medical model, database query, scientific simulation and language model cannot necessarily be compared through one common output unit.

The concept should therefore initially be applied through like-for-like or mission-equivalent comparison. The appropriate question is:

The energy demand associated with data centres is increasing rapidly. The International Energy Agency projects major growth in global data-centre electricity demand, with AI being an important driver. Its 2025 analysis projected global data-centre electricity consumption reaching around 945 TWh by 2030, while its 2026 update reported continued rapid growth and noted that aggregate demand can rise even while energy consumed per AI task falls.

This reveals a fundamental distinction between:

An individual AI operation may become more efficient while the number, complexity and autonomy of AI operations increase faster.

Therefore, sustainability cannot rely solely on improving the efficiency of each unit of hardware. It must also consider:

• workload utilisation;
• computational redundancy;
• unnecessary processing;
• task scheduling;
• model and architecture efficiency;
• reuse of prior computation where appropriate;
• whether a task remains necessary; and
• whether a smaller computational pathway can achieve the required outcome.

This is consistent with broader international analysis identifying not only infrastructure efficiency but also server efficiency, utilisation and shutdown of underused equipment as material data-centre energy-efficiency opportunities.

Most energy-efficiency strategies ask:

A deeper question is:

This produces two complementary approaches.

The two approaches should not compete. They multiply.

This gives rise to a central principle:

This proposition should not be interpreted as an argument against computing or AI. It is an argument for purposeful computing.

Data centres operate in changing conditions. Variables include:

• workload demand;
• server utilisation;
• energy availability;
• grid conditions;
• renewable generation;
• storage state;
• ambient temperature;
• cooling demand; and
• urgency of computation.

A system that is optimal at one moment may not be optimal at another.

Standardised test conditions remain indispensable because they permit comparison. However, operational compliance should also be capable of recognising measured real-world performance. The regulatory framework should therefore support both:

This principle is already compatible with a regulatory regime built around continuing reporting and audits.

Efficiency improvements can propagate through the data-centre system. For example:

Similarly:

A component-level metric may fail to recognise the full system effect. This white paper therefore proposes that, where causation can be measured and verified, the compliance framework should permit system-level efficiency accounting.

The governing principle should be:

The system boundary must be defined transparently to prevent double counting.

The purpose of this section is not to propose that Government adopt any particular proprietary technology. It is to demonstrate the range of innovations that a progressive regulatory framework should be capable of evaluating.

The innovation landscape can be divided into three broad domains:

Innovation may improve:

• how energy is sourced;
• how energy is converted, stored and delivered; or
• how much computation is required.

A regulatory framework focused only on one layer may fail to recognise improvements elsewhere.

The first domain concerns the relationship between variable renewable resources and usable electrical output. New architectures may seek to improve:

• utilisation of available renewable energy;
• operation under variable environmental conditions;
• direct energy admission;
• reduced conversion stages;
• dynamic matching of available energy and load; and
• integration with storage.

The author's Solar Driver™ work is one example of an experimental architecture directed at real-time alignment between available solar conditions and delivered output. The underlying proposition is that energy performance should be evaluated across time and actual operating conditions rather than solely through a peak operating point. The author's own materials describe Irradiance-Output Correlation as a proposed method for examining the relationship between available irradiance and output over time.

At present, these concepts should be treated as innovation candidates requiring independent, application-specific validation before any claim of data-centre compliance or performance improvement is made.

The policy relevance is broader than the particular technology:

Data centres require both efficiency and continuity. The engineering objective is therefore not simply:

but:

Alternative power architectures may seek to reduce:

• conversion stages;
• switching losses;
• resistive losses;
• unnecessary intermediate processing;
• storage and discharge losses; and
• duplicated infrastructure.

The author's broader energy work includes proposed architectures involving energy admission, storage, switching and power-flow management. These technologies have not been established by the draft Bill as compliance mechanisms and would require appropriate engineering validation for data-centre use.

Their relevance to policy is again methodological:

An alternative architecture that safely and reliably achieves the required electrical outcome with lower verified total loss should be capable of entering an approved test and compliance pathway.

The third domain may become increasingly important as AI-related demand grows. Conventional computational systems frequently receive a request and process information to determine an output.

An alternative approach may begin by defining the required mission or outcome and reducing the computational search space before expensive processing occurs. Conceptually:

If such an architecture can produce an equivalent or superior defined outcome with less compute, then:

The author's work in Mission-First computing and Relational Code™ explores this broader proposition. A related patent document describes adaptive computational pathways and the reduction of computational redundancy as intended objectives of a relational computing framework.

These are proposed architectures rather than established regulatory metrics. The policy point is nevertheless significant:

Where the same required digital outcome can be achieved with less computation, the resulting energy reduction is a legitimate form of efficiency.

The three innovation domains can be brought together in a common architecture:

Possible operating states may include:

This paper refers to the general concept as Resource-Intelligent Computing. It is not a proposal that every data centre must use one particular control system. It is a regulatory principle:

The distinction in section 33 between facility-level and equipment-level energy efficiency should be preserved. This is a strong foundation. The two categories recognise that an efficient building cannot compensate indefinitely for inefficient IT equipment, and efficient IT equipment does not eliminate facility overhead.

The legislation or its subsidiary framework should be drafted broadly enough to permit future recognition of verified reductions in the computational resources required to achieve a defined or equivalent digital outcome. This need not immediately become a mandatory universal metric — it could develop through voluntary reporting, sandboxes, pilots, sector benchmarks, procurement incentives or alternative compliance pathways.

The Authority should be empowered expressly to recognise an alternative method where an operator demonstrates equal or better required outcome, equal or better resilience, and lower verified resource consumption. The burden of proof rests on the party seeking recognition; evidence should be measurable, reproducible, auditable, based on an agreed system boundary, independently verified where appropriate, and compared against a valid baseline.

Standardised tests are essential for comparability. Actual operating conditions are essential for reality. The regulatory framework should permit both — allowing regulators to identify technologies that perform well in real-world tropical, variable-load or AI-intensive environments while retaining standardised benchmarks.

Where an innovation produces verified effects across multiple layers, the framework should permit those effects to be recognised. Safeguards should include defined measurement boundaries, prevention of double counting, baseline consistency, treatment of rebound effects, treatment of reliability changes and independent verification where material.

Singapore should consider a dedicated sandbox for technologies that may improve data-centre sustainability but do not fit neatly within established architectures — including alternative power-conversion, advanced storage, renewable integration, dynamic energy admission, workload shifting, low-compute AI, mission-based computing, advanced cooling, waste-heat utilisation and integrated energy-compute optimisation.

Minimum thresholds are necessary, but they can create a compliance ceiling if the regulatory incentive ends once the threshold is reached. Singapore should recognise year-on-year verified improvement, performance beyond minimum standards, demonstrated reductions in resource intensity, successful deployment of validated new technologies and transparent sharing of non-proprietary performance data.

A progressive framework should be open to innovation but rigorous about evidence. A possible hierarchy: Level 1 — Laboratory evidence; Level 2 — Pilot evidence; Level 3 — Independent field validation; Level 4 — Compliance recognition; Level 5 — Scaled deployment evidence. Innovation should not be rejected because it is new, and it should not be accepted merely because it is new. Evidence is the bridge.

Without attempting to replace Parliamentary Counsel's drafting, the following policy language illustrates the intended direction:

A further provision or code could provide:

The precise legal drafting would require governmental review. The policy objective is clear:

Singapore cannot compete indefinitely through unlimited land, unlimited water or unlimited domestic energy resources. Its comparative advantage lies elsewhere:

The draft Bill creates an opportunity to combine all three. The policy choice is not simply between:

A third path exists:

Singapore's Green Data Centre Roadmap already seeks sustainable growth and includes work on both facility and IT-equipment efficiency. The draft Bill can provide the legislative architecture for the next stage. Singapore could become a global testbed for: