Blackstone-backed AirTrunk is reportedly seeking about S$2 billion of debt as it prepares a potential Singapore REIT listing. The deal could do more than create another data-centre REIT: it may test how much public-market investors are willing to pay for AI infrastructure at a time when banks are becoming more selective about financing the sector.
The next big test for Singapore’s data-centre investment story may not come from an earnings announcement.
It could come from an IPO.
Blackstone-backed AirTrunk is reportedly seeking about S$2 billion of debt financing as it prepares a potential Singapore-listed REIT backed by its data-centre assets.
According to Bloomberg, the company has approached lenders for several debt tranches with maturities of roughly three to seven years, denominated in Singapore dollars and Japanese yen. The proposed debt could be used by the REIT to acquire AirTrunk assets and refinance existing debt, although terms have not been finalised.
That sounds like another financing transaction in a sector already awash with capital.
It is more important than that.
If the REIT proceeds, AirTrunk could give public-market investors something the private data-centre market has lacked in recent years:
a new observable price for institutional-quality AI infrastructure.
And that makes the proposed listing particularly relevant to investors in Singapore-listed data-cententre REITs such as Keppel DC REIT.
The question is not simply whether AirTrunk can list.
It is:
How much will investors pay for the AI infrastructure boom once they have to put a public-market price on it?
The IPO is still a proposal, not a done deal
The first distinction investors need to make is between what has been reported and what has actually been confirmed.
Reuters reported in July that AirTrunk was targeting a September or October launch for a Singapore REIT IPO that could raise about US$1.5 billion, or around S$1.9 billion at the exchange rate cited at the time.
The company had reportedly filed confidentially and begun meeting potential cornerstone investors.
But Reuters also noted that both the timing and size could change depending on market conditions. AirTrunk did not comment.
Earlier market reports suggested an asset pool of around S$5 billion and a target yield of roughly 6%.
Those figures should be treated as market reports, not confirmed IPO terms.
There is currently no basis for writing that AirTrunk will list at a 6% yield, or that the IPO will definitely raise S$1.5 billion.
That distinction will become critical once formal transaction documents appear.
Why the S$2 billion debt request is arguably more important
The IPO is interesting.
The financing environment behind it is even more interesting.
Bloomberg data cited in recent reports put Asian data-centre loans at almost US$29 billion since the start of 2025, with banks becoming more selective as their exposure to the sector approaches internal limits.
That creates a striking contradiction.
AI demand is accelerating.
But the amount of capital required to build the infrastructure supporting that demand is also becoming enormous.
Data centres require:
- land
- power
- cooling
- network infrastructure
- construction capital
- long-term financing.
And unlike software businesses, data-centre operators cannot scale capacity simply by spending more on servers.
They need physical infrastructure financed with very large amounts of capital.
AirTrunk itself illustrates the scale.
In August 2025, the company announced an A$16 billion refinancing excluding Japan, taking its total financing platform to more than A$18 billion including Japan. More than 60 banks and financiers participated.
AirTrunk also secured US$2.325 billion of green financing in July 2026 for its JHB2 hyperscale data-centre campus in Johor. The company described it as its largest single-asset financing to date.
The numbers make one thing clear:
AI infrastructure is becoming a capital-intensive asset class.
And that is where a REIT becomes strategically useful
Consider the basic structure.
Private investors own a large portfolio of data centres.
↓
The operator raises debt to develop and expand those assets.
↓
Assets mature and generate recurring rental income.
↓
Some assets can potentially be transferred into a listed REIT.
↓
The REIT raises public equity and debt.
↓
Capital is returned or recycled.
↓
The broader platform can potentially redeploy capital into new development.
This is the capital-recycling model.
It is one of the most important reasons a successful AirTrunk listing could matter beyond AirTrunk itself.
A REIT potentially gives a large private data-centre platform another source of capital.
Instead of relying entirely on:
bank debt + private equity
the platform can potentially add:
listed equity + REIT debt + public-market investors.
That broadens the pool of capital available to the sector.
AirTrunk already has the scale to make this interesting
This is not an untested data-centre operator attempting to access public markets.
Blackstone and Canada Pension Plan Investment Board completed their acquisition of AirTrunk in December 2024 in a transaction with an implied enterprise value of more than A$24 billion. At the time of the transaction, AirTrunk said it had more than 800MW of capacity committed to customers and land supporting more than 1GW of future growth.
Since then, the platform has continued expanding.
AirTrunk’s current pipeline includes developments across Singapore, Malaysia, Japan, Australia and India, while the company announced plans in June 2026 to invest US$30 billion in India and build more than 5GW of capacity there.
That makes a listed vehicle potentially more than an exit mechanism.
It could become part of the funding architecture supporting AirTrunk’s continued expansion.
But a REIT does not make leverage disappear
This is an important point for investors.
A listed REIT does not transform data centres into low-risk infrastructure.
It changes the ownership and financing structure.
Investors still have exposure to:
property values
lease cash flows
interest rates
debt
refinancing
equity-market valuation.
Suppose, purely for illustration, a future REIT had S$5 billion of assets and S$2 billion of debt.
That would represent approximately 40% debt-to-assets.
But that is not AirTrunk’s forecast gearing.
The actual ratio would depend on the final portfolio valuation, IPO proceeds, debt raised, transaction structure and existing liabilities.
The point is simply that a data-centre REIT can still be a leveraged investment.
And that makes the eventual yield critically important.
The number to watch isn’t the IPO size
Investors will naturally focus on:
How much money will AirTrunk raise?
But I would watch a different number first.
The IPO yield.
Why?
Because the yield tells us how the public market is pricing the underlying cash flows.
Imagine two outcomes.
Outcome A
AirTrunk achieves a relatively low initial yield.
That could suggest investors are comfortable paying a premium for:
- hyperscale assets
- strong tenants
- long leases
- AI-related demand
- development growth.
Outcome B
AirTrunk needs to offer a significantly higher yield.
That could indicate that investors are demanding more compensation for:
- leverage
- interest rates
- tenant concentration
- valuation risk
- development risk
- data-centre sector concentration.
In other words:
The IPO could tell us what the market thinks AI infrastructure is actually worth.
This is where Keppel DC REIT becomes interesting
For Singapore investors, the obvious listed comparison is Keppel DC REIT.
Keppel DC REIT is already a public-market vehicle giving investors exposure to data-centre infrastructure.
Its latest reported distribution for the six months ended June 2026 was 5.714 cents per unit, with payment made on September 18.
The REIT has also recently expanded its Japan exposure through the acquisition of two hyperscale colocation data centres in Greater Tokyo.
That means AirTrunk could arrive in the market at a particularly interesting moment.
Investors could potentially have two different ways to gain exposure to the same broad structural theme:
an established listed data-centre REIT
versus
a new vehicle backed by a major private data-centre platform.
But this is where investors need to resist a simplistic comparison.
AirTrunk’s IPO would not automatically validate Keppel DC REIT
A successful AirTrunk listing would not necessarily mean Keppel DC REIT is undervalued.
Nor would a weak AirTrunk IPO automatically mean Keppel DC REIT is overvalued.
The portfolios could differ materially in:
- geography
- tenant concentration
- lease length
- rental escalation
- occupancy
- asset age
- development pipeline
- leverage
- funding costs
- capital requirements.
The correct way to use the IPO would be as a benchmark, not a verdict.
That is an important distinction.
In fact, AirTrunk could create a new problem for existing data-centre REITs
Consider the scenario in which AirTrunk launches with exceptionally attractive assets and a competitive yield.
Investors could then ask:
Why pay a higher valuation for another data-centre REIT when I can buy a newly listed portfolio of high-quality hyperscale assets?
That could create competitive pressure on existing vehicles.
Alternatively, if AirTrunk needs to offer a high yield to attract investors, it could reset required returns across the sector.
Either way, the IPO becomes informative.
AirTrunk does not simply add another REIT.
It adds another reference point for the valuation of data-centre assets.
The bigger issue: AI demand versus cost of capital
This is where the AirTrunk story becomes much bigger than an IPO.
The AI infrastructure investment thesis effectively contains two separate premiums.
1. The operating premium
AI can potentially drive:
- demand for data-centre capacity
- higher utilisation
- rental growth
- new development
- longer customer commitments.
2. The financing premium
Investors and lenders decide how much those future cash flows are worth.
And those two variables can move in opposite directions.
For example:
AI demand ↑
while:
bond yields ↑
debt costs ↑
required REIT yield ↑
asset valuation multiples ↓
This creates one of the most important risks in the AI infrastructure trade.
A company can be completely right about AI demand and still deliver mediocre equity returns.
Why?
Because shareholders don’t own AI demand.
They own the cash flow left after the cost of capital.
That is why the S$2 billion debt request matters
The reported debt package is effectively a window into this problem.
AirTrunk needs enormous amounts of capital to expand.
Banks are willing to lend.
But banks also have balance-sheet constraints.
The reported financing discussions involve multiple SGD and JPY tranches with maturities of roughly three to seven years. The precise pricing and final structure have not been disclosed.
This suggests that the financing side of the AI infrastructure boom is becoming almost as important as the demand side.
And a REIT potentially solves part of the problem.
It can bring in another pool of capital:
Public equity investors.
The capital-recycling thesis
This may ultimately be the most important reason to watch the IPO.
Suppose private infrastructure investors build data centres using large amounts of debt and equity.
Once those assets become mature and income-producing, they don’t necessarily have to remain locked inside the private vehicle.
They can potentially be sold or injected into a REIT.
The REIT raises capital.
The private owner recycles capital.
That capital can then support:
new data centres
new land
new power capacity
new markets
and potentially more AI infrastructure.
This creates a virtuous capital cycle if public-market valuations remain attractive.
But it can also break down.
If REIT investors demand high yields, asset recycling becomes less attractive.
And that is precisely why the IPO valuation matters.
AirTrunk could therefore become a test of the entire capital-recycling model
There are two possible readings.
Bullish interpretation
A successful IPO demonstrates that:
institutional investors are willing to fund data-centre infrastructure through listed vehicles.
That would create another scalable source of capital for the sector.
Cautious interpretation
A difficult IPO demonstrates that:
private-market valuations for data-cententre assets are not automatically transferable to public markets.
That could force asset owners to accept higher yields and lower valuations.
The eventual outcome matters well beyond AirTrunk.
What investors should look for in the prospectus
Forget the headline IPO proceeds initially.
When the prospectus appears, these are the numbers that matter.
| Metric | Why it matters |
|---|---|
| IPO valuation | Establishes the public-market price of the assets |
| Initial yield | Shows required investor return |
| Gearing | Shows how much leverage sits inside the vehicle |
| WALE | Indicates cash-flow visibility |
| Lease expiry profile | Shows near-term repricing/refinancing exposure |
| Rental escalations | Indicates contractual growth |
| Rental reversion | Indicates potential market-rent upside/downside |
| Occupancy | Tests operating quality |
| Top-tenant concentration | Measures counterparty concentration |
| Development pipeline | Determines future capital requirements |
| Debt maturity profile | Determines refinancing risk |
| Cost of debt | Determines how much operating growth reaches equity holders |
And there is one question I would put above all of them:
How much of the portfolio’s apparent AI growth is already reflected in the IPO valuation?
“Hyperscaler” does not automatically mean low risk
This is another distinction worth making.
Data-centre investors often associate hyperscaler customers with strong credit quality and long-term demand.
But tenant quality is only one part of the equation.
Investors still need to consider:
- lease duration
- rental escalation
- renewal economics
- customer concentration
- power availability
- asset location
- competition
- development requirements.
A portfolio can have excellent customers and still be expensive.
That is why yield and valuation remain essential.
The comparison with Keppel DC REIT will be fascinating
When AirTrunk’s portfolio becomes public, investors will be able to construct a much more meaningful comparison.
| Question | AirTrunk REIT | Keppel DC REIT |
|---|---|---|
| Public-market valuation | TBD | Existing |
| Initial / current yield | TBD | Existing |
| Gearing | TBD | Existing |
| Portfolio | TBD | Established |
| Lease profile | TBD | Mixed |
| Tenant concentration | TBD | Existing disclosures |
| Rental growth | TBD | Existing portfolio data |
| Development pipeline | Potentially significant | Existing |
| Track record as listed vehicle | New | Established |
| Capital recycling | Potentially significant | Existing capability |
The table should remain incomplete until AirTrunk releases formal transaction documents.
That is preferable to manufacturing an apples-to-apples comparison before the underlying numbers exist.
The biggest risk to the AI infrastructure thesis
The obvious risk is that AI demand disappoints.
But I don’t think that is the most interesting risk right now.
The more subtle risk is:
AI demand remains strong, but the cost of capital rises faster than rental income.
Imagine:
Rental income grows 8%.
But:
- borrowing costs rise
- refinancing becomes more expensive
- REIT yields rise
- asset valuations decline
- equity issuance becomes more dilutive.
The underlying infrastructure can still be performing well.
Yet shareholders may not capture all of that growth.
This is why AI demand is not the same thing as AI investment returns.
What would make the thesis wrong?
There are several ways this story could fail.
1. AirTrunk delays or cancels the IPO
Market conditions could make a public listing unattractive.
2. The IPO requires a much higher yield
That would indicate public investors are less willing to pay the premium implied by private-market valuations.
3. Leverage is high
A heavily leveraged REIT would be more sensitive to rates and refinancing.
4. Tenant concentration is excessive
Strong individual tenants do not eliminate concentration risk.
5. Growth requires constant capital
If development absorbs large amounts of capital, headline rental growth may not translate cleanly into distributable income.
6. Public-market valuations diverge from private-market valuations
This is perhaps the most important one.
If private owners value data centres at aggressive multiples but public REIT investors demand significantly higher yields, the capital-recycling model becomes less attractive.
What AirTrunk could mean for Singapore’s REIT market
Singapore already has listed data-centre vehicles, including Keppel DC REIT and Digital Core REIT, while NTT DC REIT has also expanded the listed data-cententre universe.
An AirTrunk listing would add another major institutional-quality data-centre platform to that ecosystem.
But the significance would go beyond another ticker.
It could help establish whether Singapore’s public markets can become an important financing venue for large-scale digital infrastructure.
That is strategically important.
Singapore has already established itself as a major REIT market.
The next question is whether it can become a major public financing market for the AI infrastructure build-out.
The investment equation investors should remember
The AI infrastructure story is often presented as:
AI demand
↓
Data-centre demand
↓
Higher rents
↓
Higher earnings
↓
Higher valuations
But there is another line underneath it:
Capital expenditure
↓
Debt
↓
Interest costs
↓
Equity issuance
↓
Required yield
↓
Asset valuation
The second line can determine how much of the first line ultimately accrues to shareholders.
That is the part investors should not overlook.
So what should Keppel DC REIT investors watch?
AirTrunk’s IPO should be treated as a benchmark event, not an automatic buy or sell signal for Keppel DC REIT.
The important numbers will be:
- AirTrunk’s IPO yield
- Portfolio valuation
- Gearing
- Lease duration
- Tenant concentration
- Rental escalation
- Rental reversion
- Cost of debt
- Development pipeline
- Market reaction after listing
Then compare those numbers with Keppel DC REIT’s corresponding metrics.
If AirTrunk receives a premium valuation despite similar or greater leverage, that could indicate strong institutional appetite for hyperscale data-centre assets.
If it requires a large yield premium, that could indicate the public market is becoming more demanding about the price of AI infrastructure.
Either outcome would be useful information.
The deeper lesson for AI investors
The AI infrastructure boom has passed through several stages.
First came the question:
Will AI create enough demand for data centres?
Increasingly, that question is being answered by enormous investments from cloud and AI companies.
The next question is different:
Who will finance all the infrastructure required to satisfy that demand?
Banks have supplied enormous amounts of capital.
Private equity and infrastructure funds have supplied enormous amounts of equity.
Bond and green-loan markets have expanded.
And now public REITs may become another important source of capital.
AirTrunk sits directly at that intersection.
The real AirTrunk question
It is tempting to frame the story as:
“Blackstone is bringing another huge data-centre REIT to Singapore.”
That undersells what could be happening.
The more important question is:
Can Singapore’s public markets absorb the enormous capital requirements of the AI infrastructure boom at valuations attractive enough for private owners to keep recycling assets into listed vehicles?
If the answer is yes, the implications extend beyond one IPO.
It could strengthen the role of REITs as a financing mechanism for data centres.
If the answer is no, investors may discover that strong AI demand does not automatically translate into attractive public-market valuations.
And that distinction could become increasingly important as the sector moves from an infrastructure-construction story to a capital-allocation story.