BlackRock and IFM close in on the $25B Stack data-center deal

The world's biggest asset manager wants to own the wires. Not the AI models, not the chips — the buildings the chips live in.

A consortium backed by BlackRock and IFM Investors is in exclusive talks to buy Stack Infrastructure's Asia-Pacific data-center portfolio for up to $25 billion from Blue Owl Capital. Bloomberg News reported the talks on 24 September 2026; Reuters, careful as ever, could not independently verify them. Talks, in deal-speak, are what fail right before they succeed — and exclusive talks are what fail right before the press release is written.

What is for sale: 11 to 14 data centers across Tokyo, Osaka, Sydney, Melbourne and Johor Bahru, with 1.2 to 1.4 gigawatts of IT capacity. Other coverage counts Singapore and Seoul among the markets too. The precise map matters less than what it is: the physical ground floor of the hyperscalers' Asia expansion — the rooms where AWS, Microsoft Azure and Google Cloud actually live.

Start with the map, because the map is the argument. Tokyo and Osaka anchor the Japanese end — the world's third-largest economy, land-scarce, power-expensive, and allergic to building anything slowly. Sydney and Melbourne anchor the Australian end — a continent with cheap land, abundant sun, and a government that has spent the decade courting the cloud. Johor Bahru, across the strait from Singapore, is the newest and cleverest piece: land and power at a fraction of Singapore's cost, close enough to serve the city-state's demand without sitting inside its moratorium-era constraints. If Singapore and Seoul are in the package, as some coverage suggests, the portfolio spans the region's four most important compute markets. Whoever buys it buys the floor plan of Asia's AI decade.

The buyers are the interesting part. BlackRock's AI Infrastructure Partnership — AIP, a roughly $100 billion commitment pool launched in late 2024 specifically to own AI compute infrastructure — has Nvidia, xAI, Microsoft and MGX among its investors. Technology's giants, in other words, are paying the landlord alongside the landlord. Next to it sits IFM Investors, the Australian infrastructure house with about $202 billion under management. BlackRock itself is roughly a $15 trillion operation. This is not a speculative bid. It is the furniture of the financial system arriving.

The structure of the money deserves a closer look. AIP is not BlackRock's balance sheet — it is a dedicated infrastructure vehicle, which matters because it signals how the world's largest asset manager now classifies a data center: not as real estate, not as technology, but as infrastructure, the same category as toll roads and power grids. The presence of Nvidia, xAI, Microsoft and MGX as co-investors adds a second signal. When the tenants' suppliers and the tenants' owners sit on the same side of the table as the landlord, the old adversarial model of landlord-versus-tenant has quietly been replaced by something else: a vertical alliance of everyone who profits when compute gets more expensive.

One outlet reports the deal is already closed, with BlackRock taking 60% via its Global Infrastructure Partners arm and IFM 40%. Others say talks. The split may matter less than the direction: the capital is moving toward power, and it is not being subtle about it. GIP's appearance is itself a tell — BlackRock bought the infrastructure specialist precisely to do deals like this, and a 60/40 split with IFM reads like risk-sharing on a number this large: even $15 trillion operations prefer company when the ticket is $25 billion.

And the seller's arithmetic is the deal's quiet headline. Blue Owl Capital assembled this portfolio in 2021–2023 at a blended cost of about $14 billion — back when data centers were considered, in the industry's own phrase, boring boxes. Selling at $25 billion crystallises a return of roughly 78% in under four years. The boring boxes turned out to be anything but.

The timing of that assembly is the whole story in miniature. In 2021–2023, the industry was still recovering from a decade in which data centers were treated as commodity real estate — low-margin, capital-hungry, and about as glamorous as a warehouse. Blue Owl bought when the narrative was dull and the yields were steady. Then the large language models arrived, and every forecast for compute demand was revised upward by an order of magnitude. The portfolio did not change. The world changed around it. That is what a 78% return in four years usually is: not genius, but position — being in the right boring asset when it stops being boring.

There is a price clue in the history. Bloomberg reported in May that Stack weighed selling its Asia operations at over $30 billion. The $25 billion now discussed is the negotiated middle — the seller came down, the buyers came up, and both sides left convinced. When a $5 billion haircut still leaves a 78% return, nobody is weeping. But the gap between $30 billion and $25 billion is worth a second thought: it is the market's measure of how much scarcity is worth versus how much uncertainty costs. Exclusive talks, unverified reports, five jurisdictions of regulatory review — each of those discounts the price by a little, and $5 billion is what "a little" looks like at this scale.

The deal's economics explain the confidence. Roughly 92% of the portfolio sits under 12-year take-or-pay leases with hyperscalers — take-or-pay meaning the tenant pays whether it uses the capacity or not, which is to say, the revenue is closer to a government bond than a rental contract. Rents escalate about 3.8% a year, and power costs pass through to the tenant. Forward EBITDA near $1.1 billion implies an entry multiple of roughly 23 times — expensive by old infrastructure standards, entirely normal where scarcity rules.

Unpack that, because the jargon is doing real work. A 12-year take-or-pay lease is the landlord's dream contract: the tenant commits to pay for capacity for over a decade, and the "take-or-pay" clause means the bill arrives even if the servers sit idle. AWS, Microsoft Azure and Google Cloud are the tenants — the three companies whose demand for compute is the least cyclical thing in the technology economy. A 3.8% annual escalator means the rent rises faster than most central banks' inflation targets, every year, for twelve years. And power pass-throughs mean the landlord's single largest operating cost — electricity — is the tenant's problem, not the landlord's. Put together, the buyer is purchasing a stream of payments that behaves like a sovereign bond with a growth kicker. That is what the 23x multiple is pricing: not the buildings, but the contracts inside them.

Scarcity is the whole story, and it has two parents. The first is demand: AI's appetite for compute has turned Asian power grids and data halls into the scarcest asset class on earth. You cannot train a frontier model in a rented corner of someone else's server; you need dedicated, cooled, grid-connected floor space, and there is not enough of it. The second is regulation: sovereign data-residency mandates — governments requiring their citizens' data to stay on national soil — add a regulatory premium to every megawatt. A data center in Johor Bahru is not just a building; it is a building that certain data is legally allowed to be inside.

The power angle is the one the headline misses. A 1.2-to-1.4-gigawatt portfolio is not just compute capacity — it is a claim on electricity grids in five markets, secured, contracted, and already connected. In Tokyo and Sydney, where grid connections for new large loads are measured in years of waiting, an existing energized data center is worth more than the sum of its servers. The buyers are not just buying the right to house chips; they are buying the right to consume power at scale in places where power at scale is increasingly rationed. In the AI century, a grid connection is a license, and licenses trade at a premium.

The deeper pattern is worth naming. This is the same financialisation wave that turned wind farms, toll roads and fiber into "infrastructure" — now arriving at the one infrastructure the 2020s actually run on. The pension fund that once bought a motorway now buys a server hall. The yield looks similar. The tenant, this time, is the cloud.

That wave has a logic and a limit. The logic: infrastructure investors learned over two decades that the safest returns come from monopolies on necessities — everyone needs roads, power, water, and now compute. The limit: each successive asset class in the wave has been more complex and more contested than the last. A toll road does not consume a city's electricity. A server hall does. The pension funds are about to discover that owning the AI century's plumbing means owning its politics too — its grid queues, its water permits, its community opposition. The yield is infrastructure-like. The headaches will not be.

What could still kill it is worth stating plainly, because the coverage keeps hedging. Reuters could not verify the talks. One outlet says closed; the others say talking. Five jurisdictions means five regulatory reviews, and data centers have become strategic assets — the kind governments now screen. The final terms are unknown: the multiple, the debt package, the exact facility list, the protections for the period between signing and closing. Exclusive talks are exclusive until they are not. Every $25 billion deal in history has a moment where it wobbles, and the wobble usually arrives right about now.

But the direction of travel does not wobble. Whether this portfolio closes at $25 billion or $24 or $26, the signal is the same: the largest pools of capital on earth have decided that compute infrastructure is a core holding, not a trade. The next Asia-Pacific portfolio to come to market will be priced against this one. The 23x multiple is not just a number in a deal memo — it is the new reference price for the scarcest asset class of the AI era. And somewhere in a pension fund office, a manager who once bought motorways is explaining to a board why the server hall in Johor Bahru belongs in the same portfolio. The board, having seen the 78%, will not need much convincing.

Western coverage — Reuters/Bloomberg via SRN News — reads this as a capital-allocation story: BlackRock's AIP deploying its $100 billion war chest, the GIP machinery in action, the 60/40 split as evidence of how seriously the largest pool of capital in the world takes compute. The verification caveat — talks may fail — gets the headline treatment, because in the Western telling the deal's newsworthiness and its uncertainty are the same story.

In this telling, the 23x multiple is the analytical hinge: expensive, but explainable, because 12-year take-or-pay leases at 92% occupancy are closer to sovereign credit than to commercial property. The story is about finance discovering that AI's bottleneck is not talent or capital — it is a power socket. The Western lens is comfortable with that discovery; it has been pricing bottlenecks for a century. What it watches now is whether the multiple holds when the next portfolio comes to market — because a reference price is only real when it repeats.

The Western coverage also lingers, with professional interest, on the co-investor list. Nvidia, xAI, Microsoft and MGX putting money into the landlord's vehicle is read as vertical integration by other means: the compute buyers ensuring the compute exists. In the old economy, an automaker did not co-own the steel mill. In this one, the model-builders co-own the buildings. The West reads that as rational. It is also, quietly, a concentration of power that would have troubled the West's own antitrust instincts in any other decade.

Eastern coverage — CaproAsia out of Hong Kong, Markets Edge — reads this as a geography story. The map is the message: Tokyo, Osaka, Sydney, Melbourne, Johor Bahru. The $202 billion Australian manager, the $15 trillion American one, and the Asian ground underneath them. In this telling, the deal is Asia's grid being re-priced by someone else's demand curve.

The read from Hong Kong is pointed on the 78%: Blue Owl bought the "boring boxes" in 2021–2023 and sold the scarcest asset class on earth in 2026. The lesson drawn is not about data centers at all — it is about how fast a commodity becomes a chokepoint when the world's models all want the same floor space. The Eastern lens has watched this movie before with other commodities, and it knows the sequel: the chokepoint gets financialized, the financialization gets securitized, and the region that hosts the asset ends up renting back its own scarcity.

The Eastern coverage is also the most attentive to the tenant list. Twelve-year take-or-pay contracts with three American hyperscalers, on Asian soil, cooled with Asian electricity, now owned by American and Australian capital. In this telling, the $25 billion is a down payment on a specific future: one in which Asia provides the land, the power and the water, and the rents flow to New York and Sydney. The sovereign data-residency mandates — the regulation that makes each megawatt more valuable — are read here as the region's belated attempt to keep some of the value at home.

Global South coverage — the Economic Times' infrastructure desk — reads the lease economics and sees who is not in the room. Twelve-year take-or-pay contracts with the three American hyperscalers: in this telling, the deal is a map of digital dependency, priced at 23x EBITDA.

The read from Delhi is structural: the world's data increasingly lives in buildings owned by New York and Sydney, rented by Seattle and Mountain View, and cooled with Asian electricity. The sovereign data-residency mandates are the response — every government now wants its data on its own soil — and this $25 billion is, among other things, the market pricing that anxiety. The South notes the irony without savoring it: the mandates meant to keep data local are making the buildings that hold it more valuable to foreign owners.

The Southern lens asks the question the other two skip: what does the 23x multiple mean for everyone who is not BlackRock? It means the cost of compute — the raw material of the AI economy — is being set by the return expectations of the world's largest asset manager. Every startup in Jakarta or Lagos that rents cloud capacity will pay, in its monthly bill, a fraction of that multiple. The deal is not just a transfer of buildings. It is a transfer of pricing power over the century's key input, from the people who use it to the people who own it. The South has seen this film with oil, with shipping, with grain. It knows how it ends.