Efficient does not just mean low power usage. It means built with consent, disclosed to the public, and governed by rules a community can actually see. Four APAC markets are already doing this. Most of the region is choosing not to.
When this publication uses the word efficient, it does not mean what the industry usually means by it. The industry’s definition is narrow and technical: power usage effectiveness, water usage effectiveness, renewable energy share. Those numbers matter. But a data centre can hit every one of them and still be the reason a court case has run for three years, or the reason a community found out what was being built next door only after the trucks arrived.
Arc Brief’s definition is broader. An efficient data centre is one that runs on abundant power and scarce water responsibly; that was sited with the affected community’s knowledge and input before construction began, not after; that discloses its operating standards to a public that can actually see them; and that is financed on terms an investor can evaluate rather than take on faith. A facility can be technically efficient and governance-inefficient at the same time. Most of the facilities this publication has covered over the past year are exactly that.
The good version — efficient in the fuller sense — already exists. It is being built in specific markets, under specific frameworks, right now. This is the piece that names them, and names what would have to change for the good version to become the regional standard rather than the regional exception.
Enjoying this intelligence dispatch? Receive Arc Brief in your inbox every Friday.
| Reform | Singapore | Japan | Korea | Australia |
|---|---|---|---|---|
| Disclosure | ||||
| Siting & design | ||||
| Open dialogue | ||||
| Financial transparency |
Singapore runs the most stringent binding technical standard in the region. Its second data centre call for applications, DC-CFA2, launched in December 2025 and closed for applications on 31 March 2026, requires a power usage effectiveness of 1.25 at full IT load — tighter than the EU’s 2030 target — at least 50 per cent of power from approved green energy pathways, and compliance with the national standard on IT equipment efficiency, according to law firm analyses from Norton Rose Fulbright and King & Wood Mallesons. Compliance with these requirements is a condition of being awarded DC-CFA2 capacity, enforced through the allocation process itself rather than published as public data. Singapore also allocated twenty hectares on Jurong Island specifically for a low-carbon data centre park, giving qualifying projects direct access to renewable energy infrastructure rather than asking them to source it independently. Operators must demonstrate actual clean energy supply, not offsets.
Japan’s Hokkaido and Kyushu programme works through incentive rather than mandate, and it is instructive precisely because it is a different model. As Intralink and Energy Tracker Asia have reported, Japan’s cabinet directed the Ministry of Economy, Trade and Industry and the communications ministry to convene a government body called the Watt-Bit Collaboration — bringing together academics, power industry representatives, and technology companies — to build a framework for redistributing compute capacity away from congested Tokyo toward regions with abundant renewable energy, cheaper land, and available grid headroom. Hokkaido and Kyushu are the priority destinations. The Ministry of the Environment opened a further subsidy programme in May 2026 covering up to half the cost of decarbonisation equipment at data centres, and a separate programme offers up to ¥210 billion under Japan’s GX 2040 vision for operators running entirely on decarbonised electricity. From 2029, new data centres built anywhere in Japan must meet minimum energy efficiency standards or pay a financial penalty. The mechanism is different from Singapore’s — carrots before the mandate arrives, rather than a mandate from the start — but the direction is the same.
Korea’s AI Basic Act, effective January 2026, is the first comprehensive binding AI law in the region, with extraterritorial reach and a transparency requirement that obligates operators of high-impact AI systems to notify users when AI is being used and to label AI-generated content, per analyses from Cooley and Law.asia. This is a narrower transparency mandate than the disclosure regimes described below — it governs what an AI system tells its users, not what a data centre discloses about its physical operations — but it establishes that binding, enforceable transparency obligations are politically achievable in the region, which several other APAC governments have treated as untested territory.
Australia has moved the furthest in the shortest time. As recently as March 2026, the federal government’s data centre standards were non-binding “expectations.” In July 2026, the Prime Minister announced that these will be converted into a mandatory National Framework, developed in consultation with state and territory governments — as reported by law firms Gilbert + Tobin and White & Case — with several binding requirements: large-scale data centre operators may be required to underwrite or supply their own power, pay the full cost of new grid connections rather than passing them to consumers, minimise water usage, and — critically — the framework is expected to optimise data centre siting through consultation with states, territories, and local communities. The direction matches the accountability gaps this publication documented in Issues 009 through 012 — the EPA licensing gap at Kemps Creek, the missing compliance reports at Lane Cove, the parliamentary inquiry that opened in May 2026 — though this publication has no basis to claim those specific findings caused the policy shift. The correction and the pattern are simply aligned. The framework is not yet law. The exposure draft and consultation process are still to come. But the direction — from voluntary expectation to mandatory rule, with community consultation named explicitly — is the correction the earlier issues in this series argued was missing.
Four governments, four different mechanisms. None is complete. Each is further ahead than the rest of the region.
Reading across these four markets, a shared reform agenda emerges. Not a single policy, but four distinct interventions that each address a different failure this publication has documented.
Disclosure. The EU’s Energy Efficiency Directive requires every data centre above 500 kilowatts to report power usage effectiveness, water usage effectiveness, and renewable energy share to a public database annually. Singapore enforces equivalent standards through its DC-CFA2 allocation process rather than public disclosure — compliance is a condition of winning capacity, not something the public can independently verify. No other APAC market requires either. The reform: operators disclose the operational metrics that determine a facility’s environmental footprint, in a format the public — not just the regulator — can see. This is the single most replicable reform on this list. It requires no new infrastructure, no new technology, and no negotiation with communities. It requires only that a government decide operators must publish numbers they already collect internally.
Community-friendly siting and design. Siting is not only which government approves a project. It is how the finished facility sits against the streets around it, and whether it pays its own way on the infrastructure it strains. Australia’s forthcoming National Framework moves on both counts: it proposes consultation with states, territories, and local communities before approval, and would require operators to underwrite new power supply and pay the full cost of new grid connections rather than pass that cost to consumers — the clearest APAC move yet toward operators funding the shared infrastructure their load creates, rather than leaving the bill for the utility and, eventually, the ratepayer. Japan’s Watt-Bit Collaboration achieves a version of the same goal from the other direction, directing capacity toward regions with spare land and grid headroom so the strain is avoided rather than mitigated afterward. Neither addresses the physical design of the facility itself — setbacks from homes, screened and enclosed generators, contained noise and waste heat — which remains a gap in every market examined here. The standard this publication would point to: siting and design conditions written into the permit itself, plus a binding requirement that operators fund the utility and public infrastructure upgrades their facility necessitates.
Open dialogue. The Watt-Bit Collaboration is the clearest example in the region of a standing, multi-stakeholder body — government, industry, and academic representatives meeting on an ongoing basis to manage capacity distribution as a shared problem rather than a series of unilateral corporate decisions. The Asia-Pacific Data Centre Association launched a comparable industry-only initiative, the Sustainable Digital Infrastructure Accord, in March 2026 — a voluntary, non-binding baseline of sustainability commitments across the sector, according to reporting from DC Byte. Voluntary industry coordination is a start. It is not the same as a standing body with government and community representation and the authority to shape binding outcomes. The reform: operator-community councils convened before construction and maintained through the operational life of a facility, with published minutes and genuine agenda-setting rights for the community side — not an advisory panel that meets once and is not reconvened.
Financial transparency. This is the reform Arc Brief’s own coverage is best positioned to argue for. Issue 013 described the sovereign wealth and pension capital financing this region’s buildout, and found that the specific tenant credit quality, contract duration, market tier, and grid position behind any given fund’s data centre exposure is rarely disclosed at asset level. A member of a pension fund can see that their fund holds data centre exposure. They cannot usually see whether that exposure looks like the AirTrunk profile — diversified, hyperscaler-tenanted, long-contracted — or the Irish profile that ended up stranded. The reform: a standardised disclosure format for infrastructure funds’ data centre positions, similar in spirit to the operational disclosure described above, so that a saver can evaluate the quality of their own exposure rather than trust that someone else has done so. The mechanism does not need to be invented from nothing. Australia’s phased Climate Disclosure Regime already requires large entities — a category most infrastructure funds fall into — to report climate-related financial risk, with the largest entities reporting since January 2025 and the next tier following from July 2026. Extending that same reporting architecture to require asset-level detail on infrastructure holdings, including data centres specifically, would be a scope expansion within a framework regulators are already enforcing, not a new regime built from scratch.
None of the four reforms above is technically difficult. None requires an invention. The reason they remain the exception rather than the rule is not technical. It is structural, and worth naming honestly.
Investment moves faster than governance everywhere, by design. A government that wants the jobs and the announcement moves at the investor’s speed, and one that insists on binding disclosure and consultation risks losing the project to a market that will not ask. Singapore can set the region’s strictest standard because its other advantages — connectivity, stability, rule of law — are strong enough that operators pay the compliance cost rather than leave. A smaller market rarely has that leverage.
Community engagement done well is expensive and slow, and no operator wants to bear that cost unilaterally while a competitor does not. This is the kind of problem a binding, sector-wide rule solves and a voluntary industry accord does not — why the difference between Japan’s government-convened Watt-Bit Collaboration and the industry-only Sustainable Digital Infrastructure Accord matters more than it first appears.
Disclosure is resisted because operators fear publishing operational metrics hands a competitive map to rivals. The EU precedent is instructive here: European operators have been reporting under public disclosure since September 2024, and this publication has found no evidence of the competitive harm operators elsewhere cite as a reason to resist similar rules.
And in nearly every market this publication has covered, the political cycle rewards the announcement over the governance. A ribbon-cutting is visible in one election term. A binding standard’s benefit — no grid freeze, no stranded asset, no backlash five years out — is not.
Fourteen issues ago, this publication asked what a data centre is. Since then it has covered how these buildings get financed, who bears the environmental cost, which communities are consulted and which are not, what happens when investment outruns governance, and what a reader’s own retirement fund is doing inside the sector without the reader’s knowledge.
The conclusion those fourteen issues point toward is not complicated. The infrastructure is being built regardless, because the life it enables has already been built. The good version of it — disclosed, consented to, governed, financially legible — exists today in enough places to prove it is achievable at the region’s current scale and speed. The bad version exists in more places, for reasons that are political and economic rather than technical.
None of the four reforms above exists yet in the specific places this publication has spent the most time. Vizag had neither disclosure of the environmental clearance process nor of the site’s operational metrics — the clearance was issued and the foundation stone laid within ten days, with no public data in between. Mekaguda has no open dialogue mechanism; the state’s documented response to community objection was to instruct an administrator to resolve it “by evening,” not to convene any standing council. Johor’s community-friendly siting and design arrived only after residents staged the region’s first public protest against a data centre — the state assemblyman’s admission that consultation should have come first was made five weeks after the protest, not five weeks before construction. Ireland’s stranded €5.8 billion is the clearest argument this publication has found for financial transparency at the fund level: that capital was, for years, visible only to the operators and their lenders, not to the public or the pension holders whose capital was adjacent to it. The four reforms are not abstractions. They are, specifically, the four things absent from each of those four cases.
Four markets are choosing, in different ways and to different degrees, to build the good version. The rest of the region has not yet made that choice, and the frameworks that would make it are not waiting to be invented. They are running in Singapore, in Hokkaido, in the drafting rooms of Australia’s forthcoming national standard. The question this publication will keep asking is which market moves next, and how long the rest of the region can defer the answer before the choice is made for it — by a grid that freezes, a community that finally reaches a court that listens, or a fund that discloses its exposure only after the loss is realised.
Three specific readers can act on this before this publication’s next issue. A regulator drafting a national framework can treat Singapore’s DC-CFA2 licensing process and Japan’s Watt-Bit Collaboration as working templates rather than hypotheticals — both are operating now and can be studied directly. An investor with data centre exposure through a pension or sovereign fund can ask their fund, specifically, whether that exposure is disclosed at the asset level Issue 013 described, or only at the portfolio level. A resident near a proposed or existing site can ask a single question of the developer or the approving authority: did the consultation happen before the permit was filed, or after? That distinction, more than any other in this piece, is what separates the good version from the one this publication has spent a year describing.
Next: Arc Brief’s next editorial returns to a single market for a closer look — what has been built, what is being contested, and where the leverage for the next twelve months actually sits.
Norton Rose Fulbright, “Singapore’s Green Data Centre Roadmap and DC-CFA2” (2026) · King & Wood Mallesons / Lexology, “Singapore Launches 200MW Data Centre Call for Application” (December 2025) · Withers, “Singapore’s Second Data Centre Call For Application” (January 2026) · Introl, “Singapore’s Green Data Center Gamble” (January 2026) · Eco-Business, “Japan launches programme to cut data centre emissions” (May 2026) · Intralink, “Thinking small: the international tech behind Japan’s urban data centers” (March 2026) · Energy Tracker Asia, “AI Data Centre Development in Japan and Clean Energy Transition” (2025) · Shulman Advisory, “Japan Data Center Update 19” (May 2026) · Law.asia, “Korea’s new AI Basic Act” (April 2026) · Cooley, “South Korea’s AI Basic Act: Overview and Key Takeaways” (January 2026) · Gilbert + Tobin, “Australian Government announces mandatory AI standards for large-scale data centres” (July 2026) · White & Case, “Australian AI update: PM’s AI and data centre speech” (July 2026) · DC Byte, “Community Pushback Is a Growing Risk for APAC Data Centre Development” (May 2026) · NSW Data Centre Guidelines (August 2026) · Lexology, “Data Centre - APAC Regulatory Guide,” on Australia’s Climate Disclosure Regime (2025–26) · Arc Brief Issues 009–013 for prior findings on Kemps Creek, Lane Cove, Mekaguda, Vizag, Johor, Ireland, and fund-level disclosure.