Who Is Paying for APAC's Data Centre Boom?

The biggest APAC data centre transaction of 2024 did not involve a hyperscaler.

On 4 September 2024, Blackstone and the Canada Pension Plan Investment Board announced the acquisition of AirTrunk — a hyperscale data centre operator founded in Sydney in 2015 — from Macquarie Asset Management and Canada’s Public Sector Pension Investment Board. The implied enterprise value was A$24 billion, including capital expenditure on committed projects. The transaction was, and remains, the largest data centre acquisition in history, the largest M&A deal completed in Australia in 2024, and Blackstone’s largest single investment in the Asia-Pacific region.

Four years earlier, Macquarie’s consortium had acquired AirTrunk for A$3 billion. That is an eightfold uplift on the platform in the time it takes to build a large data centre from planning permission to operational status.

The buyers were not a hedge fund and an investment bank. They were a private equity firm investing across four distinct strategies and a Canadian pension fund investing on behalf of twenty-two million Canadian workers and retirees. The 12 per cent stake CPP Investments committed to at closing sits in the same portfolio as every future Canadian retirement cheque.

The AirTrunk deal is the piece of the story that reveals the rest of it. To understand where APAC’s data centre boom is going, it is useful to start with who paid for the anchor asset and why. What follows is the first Arc Brief editorial on the financial architecture of the buildout — where the disclosed money is coming from, what the underlying bet is worth, and how a reader with retirement or reserve exposure to the sector should read it.

The demand thesis for APAC data centres is stronger than the AI headline captures. Enterprise cloud penetration across the region is substantially below Western benchmarks — the Asia-Pacific cloud computing market was valued at approximately $203 billion in 2025 by Mordor Intelligence, projected to grow to $602 billion by 2031 at a compound annual growth rate near 20 per cent. India alone is forecast to grow at 25 per cent annually. This is not AI-hype demand. It is baseline enterprise IT infrastructure catching up on a decade of under-investment.

Data sovereignty laws are accelerating the pace. Vietnam’s Decree 53 requires local storage of customer data. India’s Digital Personal Data Protection Act, Australia’s Privacy Act amendments, Malaysia’s MyGovCloud framework, and Indonesia’s evolving data protection regime all push workloads onshore. According to IDC research, roughly 90 per cent of APAC enterprises now run multi-cloud workloads — but where those workloads physically sit is increasingly constrained by law. Sovereign cloud infrastructure in APAC is growing at approximately 25 per cent annually according to Grand View Research. The infrastructure required to comply with these laws is contracted, long-duration, and does not depend on whether AI adoption meets its most optimistic forecasts.

The AI layer sits on top of this baseline demand rather than replacing it. Microsoft disclosed in early 2026 that it holds approximately $80 billion in Azure orders it cannot fulfil because of power constraints. Amazon has committed A$20 billion to Australian data centre infrastructure through 2029. The five largest US hyperscalers are guiding to combined capital expenditure of between $600 billion and $725 billion in 2026, with Amazon alone at approximately $200 billion. The demand is neither speculative nor short-cycle. This is why capital allocators looking for long-duration, contracted, inflation-linked cash flows have moved into the sector at scale.

The disclosed picture is partial. What is visible is substantial.

GIC, Singapore’s sovereign wealth fund with approximately $936 billion under management, is an active investor in data centre infrastructure. Its disclosed positions include a strategic joint venture with GDS to develop hyperscale build-to-suit facilities in China, an 80 per cent stake in a joint venture with Equinix that owns three data centres in Tokyo and Osaka, a $525 million joint venture with Equinix for two hyperscale facilities in Seoul, investments in Vantage’s EMEA platform, a stake in the US operator EdgeCore, and a stake in the Czech infrastructure firm CETIN. Up to 50 per cent of GIC’s net investment returns flow back to the national budget under Singapore’s Net Investment Returns Contribution framework — meaning public goods including polyclinic subsidies and transport infrastructure are funded, in part, by returns generated across GIC’s overall portfolio.

AustralianSuper, Australia’s largest pension fund with approximately A$295 billion in assets across 3.4 million members, has invested US$2.2 billion in DataBank in the United States and US$2.5 billion in Vantage’s EMEA platform. Its head of mid-risk portfolios has publicly attributed the fund’s conviction in the sector to unprecedented demand for cloud and AI infrastructure. AustralianSuper’s assets are projected to reach A$1 trillion by 2035. The two disclosed data centre positions place the fund among the larger institutional investors in the sector globally.

Aware Super has committed A$460 million to Skyline JV to expand its APAC digital infrastructure presence. CPP Investments’ AirTrunk stake sits alongside data centre joint ventures across Hong Kong, Australia, Japan, Malaysia, and Singapore. Temasek, with approximately $521 billion under management, retains broader digital infrastructure positions across the sector.

The overall shape is straightforward. Sovereign wealth and pension capital across APAC and its Canadian counterparts have taken substantial disclosed positions in the region’s digital infrastructure. What is not disclosed at any comparable level of detail is the total exposure Australian super funds hold to data centres through indirect vehicles, or the specific tenant credit quality and contract structures behind each individual investment. Fund disclosures typically report broader infrastructure allocations without breaking out digital infrastructure by asset. The answer to “what is your fund’s data centre exposure worth” is often that no one publishes the answer in useful detail.

The Ireland precedent — €5.8 billion in stranded Dublin data centre projects as of late 2025, arising from a decade of favourable conditions meeting a grid moratorium — is not the only downside risk, but it is the most concrete. Facilities that were built or planned without proactive governance became liabilities when the constraint arrived. The specific dynamic that produced that outcome — speculative construction ahead of grid capacity, without contracted hyperscaler tenants — is the specific dynamic capital allocators should be pricing.

Three further risks sit in the current market. The first is a monetisation gap. Sequoia’s David Cahn has calculated an approximate $600 billion annual revenue gap between hyperscaler AI capital expenditure and the AI-generated revenue required to justify it. An MIT NANDA study in mid-2025 found that 95 per cent of enterprise generative AI pilots have produced no measurable P&L impact. Goldman Sachs’ chief economist has publicly stated that AI contributed close to zero per cent to United States economic growth in 2025. These are not fringe voices. They frame a legitimate concern that the capex layer is running ahead of the revenue layer.

The second is an accounting question about hardware depreciation. The investor Michael Burry has argued that major AI infrastructure buyers may be understating hardware depreciation by approximately $176 billion cumulatively across 2026 to 2028, on the basis that GPUs are being depreciated over five to six years while their economic life is closer to two to three. If that view is correct, hyperscaler cloud earnings are being reported at higher levels than the underlying economics justify.

The third is the capex-to-revenue ratio itself. Meta was operating at approximately 54 per cent of revenue in capex in 2026; Oracle at 57 per cent; Microsoft at 47 per cent; Alphabet at 46 per cent. Historically, only industrial utilities and telcos have sustained these ratios. This does not mean the underlying infrastructure has no value. It does mean the multiples currently being paid for exposure may not survive an eventual reset.

A fourth risk connects the first three to what follows. The tenant credit quality that makes a Tier 1 hyperscale asset valuable is itself a function of whether the AI monetisation gap eventually closes. Hyperscaler tenants today carry strong credit because they are generating cash flows from cloud services that are being reinvested at the capex-to-revenue ratios described above. If those ratios prove unsustainable, then hyperscaler credit quality in 2028 will not necessarily match hyperscaler credit quality in 2026. Tier 1 data centre exposure anchored by hyperscaler contracts is not independent of the AI capex debate. It is a specific bet that the current tenant credit profile survives whatever adjustment eventually occurs.

These risks are not evenly distributed across APAC data centre assets.

A data centre asset’s value is determined by five variables: the credit quality of its tenants, the concentration of those tenants and the geographic diversification of the assets they occupy, the duration of their contracts, the market tier in which it operates, and its grid and permitting position. All five are known at the point of investment. None of them require an AI demand forecast to evaluate.

The AirTrunk portfolio sits close to the strong end of that framework. Its facilities are in Sydney, Melbourne, Singapore, Hong Kong, Tokyo, Osaka, and Kuala Lumpur — Tier 1 markets where hyperscaler demand is genuinely constrained by supply. Its tenant base is composed of hyperscalers on long-duration contracts. Its capacity is largely committed. And its exposure is diversified across five countries and eleven sites, which materially reduces concentration risk against any single market, grid, or tenant. When Blackstone and CPP underwrote the A$24 billion valuation, they were not making an AI demand forecast. They were making a bet that these specific tenants, at these specific locations, on these specific contracts, would continue paying for capacity that the tenants themselves cannot replicate under current conditions.

That bet is not risk-free. The hyperscaler credit that supports the tenancy today is exposed to the AI monetisation debate named above. But the bet is structured to survive a wider range of outcomes than the Irish stranded profile. It does not depend on aggressive AI demand growth to remain viable — only on hyperscaler cloud revenue continuing broadly at current levels, distributed across enough markets and tenants that no single failure is catastrophic.

The Irish stranded profile is the mirror image. Facilities were built or committed without grid capacity secured, in a market that had become saturated relative to available power, with tenant contracts less binding or less fully executed. Ireland was also, structurally, a single-market bet. Diversification would not have prevented the moratorium. It would have prevented the moratorium from being a catastrophic event for any single investor’s whole position.

APAC will produce both types of exposure. Tier 1 hyperscale facilities in Sydney, Singapore, and Tokyo with long-duration hyperscaler contracts are structurally closer to the AirTrunk profile. Speculative Tier 2 builds in emerging APAC markets, without secured tenants or without a grid position, are structurally closer to the Irish profile. Between them sits a spectrum of exposures whose quality can be evaluated, if the underlying data is available.

The disclosure gap at the fund level is real. Most Australian and Singaporean funds do not publish tenant credit ratings, contract durations, or market-tier breakdowns for their data centre positions. The information exists inside the funds. It is not routinely made public.

Three questions a member could reasonably put to their fund would produce more clarity than any published fund report currently offers. First: what proportion of the fund’s data centre exposure is in Tier 1 APAC or comparable global markets versus Tier 2 or speculative markets, and how geographically diversified that exposure is across countries and grids. Second: what is the average contract duration, credit quality, and tenant concentration of the fund’s data centre positions — a portfolio anchored by a single hyperscaler is a different exposure to one contracted across three or four. Third: what proportion of the fund’s data centre exposure has secured grid connections and completed environmental permitting at the point the fund committed capital.

Funds that can answer those questions cleanly are demonstrating a level of underwriting rigour that reduces the risk of an Ireland outcome. Funds that cannot are worth asking again in six months.

Deals of the AirTrunk scale are underwritten by institutional investors with the resources and access to answer versions of those questions before capital is committed. Whether that same level of discipline is being applied to the rest of a fund’s data centre exposure is a question the fund itself can answer, if enough members ask.