The data centre we can't buy yet
Why the Gulf's digital build-out is a financing problem wearing a technology costume.
Source: OECD
"It's in the market, but you cannot buy it."
That line, at the MENA Water-Energy Nexus Forum at DIFC, was about a solid oxide fuel cell built to power data centres. It is gas-fired, but water-positive — it produces around 270 litres of water per megawatt-hour and recycles most of it back into its own process — and its carbon dioxide comes out as a concentrated, capture-ready stream rather than diluted exhaust. The single producer is sold out, with nothing left to sell. The bottleneck wasn't the science. It was bankability.
It is one of several technologies now competing to ease the sector's power and water load. None of them is the whole answer. What they share is a more interesting common feature: most of them already work. The constraint on deploying them sits in the capital structure, not the lab.
This week, AirTrunk's new paper, Financing the Future: Building Asia Pacific's Sustainable Digital Backbone, makes a related point from the other end of the same sector — the kit that powers a data centre, and the capital that builds one. It is worth reading alongside what the Gulf is currently trying to do.
What's on the shelf
The toolkit for a lower-impact data centre is no longer speculative. A non-exhaustive list of what is in the market today:
- Liquid and direct-to-chip cooling — moves heat far more efficiently than air, and is the enabling condition for dense AI workloads, climbing from roughly 5–10 kW per rack toward 50–130 kW and beyond. Air-cooled halls cannot host them.
- Closed-loop and waterless cooling — fluid coolers and zero-water evaporative designs that fill once and recirculate, cutting external water draw close to zero.
- Waste-heat recovery — absorption chillers that run cooling off the facility's own waste heat, freeing grid capacity where expansion is constrained.
- On-site generation with capture-ready output — high-efficiency fuel cells, like the DIFC example, that produce their own water and concentrate their carbon dioxide for potential capture.
- Storage and load-shifting — batteries that let a 24/7 facility shift demand off the peak, supporting grid stability and qualifying for incentive-based financing.
- Recycled and non-potable water — treated sewage effluent and recycled-water schemes that take cooling off the potable supply entirely. Particularly relevant on a desalinated base.
The list keeps growing, and the point is not which technology wins. It is that the binding constraint sits downstream of the technology.
Stranding risk and adaptation
A data centre is a long-lived asset financed over many years. Whether it holds its value or quietly impairs depends on how well it absorbs changes in its operating environment — in resources, regulation, and the workloads it must host. Three pressures behave differently.
Energy and the carbon dependency. A facility drawing grid power inherits the grid's decarbonisation passively: as the regional mix cleans up — more solar, sustained nuclear baseload from Barakah — its energy-related emissions fall without the operator doing anything. But the dependency is only neutralised where the power is grid, renewable, or contracted through a PPA. On-site fossil generation is different: a facility leaning on its own gas turbines, or a gas-fed fuel cell, carries a carbon dependency tied to a fuel whose cost and acceptability are exposed to tightening policy. Capture-ready output is a hedge — but only if the capture is actually built and financed.
Water — the live concern, and the GCC already regulates it. A facility designed around evaporative cooling, in an economy where water is desalinated, is exposed on three fronts: rising cost, curtailment risk, and tightening discharge rules. The Gulf already has the regulatory architecture: Abu Dhabi's Trade Effluent Control Regulations (2022), Dubai Municipality's trade-waste permitting, and the mandatory Estidama Pearl and Al Sa'fat green-building systems — with policy moving toward non-potable-first cooling. A data centre built around potable evaporative cooling is adapting into, not ahead of, the regulatory direction. Adaptation is capital expenditure, partly constrained by the building as originally designed. That is where real impairment sits.
Technical obsolescence and efficiency — a moving benchmark. This is the sharper stranding risk in the AI era. A data hall built for air-cooled densities cannot host modern AI workloads without a cooling retrofit, and the benchmark keeps moving. Power Usage Effectiveness has fallen steadily; leading operators run portfolio PUEs around 1.3, the hyperscale frontier below 1.2, against legacy facilities often above 1.5. A legacy asset doesn't get worse in absolute terms; it falls behind a benchmark that improves every year.
In the Gulf the cost trajectory is structural: ambient heat means cooling is 30–40% of total energy; water is desalinated, carrying an energy penalty and a scarcity-exposed price; and measurement and compliance now cost real money under Federal Decree-Law 11 of 2024 and its national MRV system. A high-PUE, high-WUE asset sits on a steeper cost curve — and the gap widens as resource prices climb.
Which UAE assets sit in the exposed cohort? The honest read is at the level of vintage. The UAE's existing fleet of roughly 350–400 MW is dominated by Tier III colocation built between 2015 and 2021 for conventional densities and air cooling — that cohort faces the retrofit-or-reprice question. The new gigawatt-scale AI builds (Stargate UAE, Khazna's expansions) are designed for liquid cooling from the outset. The exposure is generational, not operator-specific.
How AirTrunk did it — an evolution, not a single deal
AirTrunk's model is instructive precisely because it was incremental. In September 2021 it converted its core corporate debt — A$2.1 billion — into the first sustainability-linked loan for a data centre operator in Asia Pacific, and the first anywhere to use operating PUE as a KPI: how the asset performs in service, not what the design promised. That first deal carried three KPIs — operating PUE, carbon, and diversity — with targets independently assured.
Each refinancing then raised the bar, adding Water Usage Effectiveness and Carbon Usage Effectiveness. By 2024–25, AirTrunk had linked 100% of its debt platforms — around A$18 billion — to sustainability targets. The mechanical point: the baseline came from the operator's own portfolio operating data across multiple live sites; targets were calibrated to outperform that history; third-party verification made the numbers bankable. The KPI lived inside the loan — margin steps down when targets are beaten, up when missed — so the lender prices the resource-performance trajectory directly. The financing instrument and the stranding risk are two views of the same thing.
Is the Gulf ready, or does this need piloting first?
Closer to pilot-then-refinance than plug-and-play — but the raw material is already here. A credible SLL needs three things: a measured baseline, ambitious targets, and independent verification. There are roughly 110 operating data centres across the GCC, with large portfolio operators — Khazna runs 30-plus facilities, alongside Equinix, Gulf Data Hub and Moro/DEWA — whose installed fleet is a genuine baseline. Decree-Law 11's MRV system is starting to generate the verified data layer assurance depends on.
What's less mature: regional WUE baselining is thinner than energy; off-site renewable procurement is constrained by current UAE rules; and the bench of structuring agents and second-party-opinion providers on regional data-centre deals is still developing. So the credible path is to pilot — structure first deals on conservative, well-verified targets benchmarked against the existing fleet, prove performance over a cycle, and refinance at a lower margin. The existing GCC fleet is not a reason to wait. It is the baseline that makes a pilot bankable.
The real problem
This is no longer a technology problem. The kit is largely on the shelf — cooling that barely touches water, waste-heat reuse, storage, fuel cells that make their own water. What is missing is the finance and governance to deploy it at scale, and — in the case of that sold-out fuel cell — to capture what comes out.
For the Gulf, the reframing matters. A data centre here is not just real estate. It is the infrastructure on which the region is staking its post-oil diversification, and it is being built in gigawatts. Treated that way, sustainability stops being a reporting exercise and becomes a question of resilience and sovereignty: whether the digital backbone the region is betting its future on is built to adapt, or built to strand.
The capital is willing. The technology is mostly ready. The work now is structuring the finance so the two meet — and proving, on the region's own operating fleet, that performance-linked capital does what it claims.
Figures on AirTrunk's financing and the SOFC are drawn from the companies' own published materials. UAE regulatory references are to the Abu Dhabi DoE Trade Effluent Control Regulations, Dubai Municipality trade-waste permitting, the Estidama Pearl and Al Sa'fat green-building systems, and Federal Decree-Law 11 of 2024.