The Lease Is the Bond. So What Is the Gulf Selling?
European Data Centres, Read from the Gulf | Part One of Seven
Written by Lanre Okunnuga, August 2026
Europe has pre-let 83% of its data centre pipeline before construction finishes. Saudi Arabia had 467MW operational in the first quarter of 2026 against a national target commonly cited in the 1.3 to 1.5GW range by 2030.
I have spent this summer writing a series on European data centres on my own platform, working through the Netherlands, Ireland, the UK, Germany, Spain, Finland and Johor one market at a time. This piece turns the same lens on the Gulf, because the comparison surfaces something that neither region shows on its own.
Here is the route. The easy read of those two opening numbers is that Europe builds against contracts and the Gulf builds against hope. That read is wrong in a specific and interesting way, and the first two sections deal with it. Underneath it sits the harder question of what is actually being sold in each region and for how long, which is where the piece spends most of its time. From there the consequences separate depending on whether capital is held as debt or as equity, because the same contract does different work on either side. The piece closes on demand mobility, on the state money funding all of this, and on what the Gulf's sovereign anchoring does and does not cover.
The reason this matters to anyone holding these assets is straightforward. What is sold determines what the asset is worth, who is able to buy it later, and at what cost capital can be raised against it. Those are ownership questions before they are credit questions.
The easy read
European capacity is spoken for years ahead of delivery. JLL's Q4 2025 figures showed 83% of the European pipeline already pre-let. By the second quarter of 2026 colocation vacancy across FLAP-D had tightened to 6.4%, down from a peak of 16.9% in 2021, with Frankfurt at 3.1%. Live capacity across the five markets has reached 3.8GW. Pre-leasing, in JLL's own framing, has become a necessity rather than a choice, and partial pre-lets are now standard.
The reason is not that European operators are more disciplined. The scarce input in Europe is grid connection and planning consent. Powered land has risen 82% since 2021, from €1.24 million to €2.26 million per megawatt. Lead times in primary markets reach ten years. When capacity is that hard to create, tenants commit early because waiting costs more than committing.
Gulf capacity is built ahead of the identified customer, and for the mirror-image reason. Renewable generation at or near two cents a kilowatt-hour, delivered industrial tariffs among the lowest anywhere, and available land together remove the constraint Europe cannot solve. There is no queue to wait in, so nothing forces a tenant to commit three years early.
Why that read stops short
Constraints producing different behaviour is not the same as one region taking more risk than the other. Both regions are building capacity that does not yet have a permanent home. Europe's happens to be spoken for on paper because scarcity forced the paperwork early.
The distinction that carries weight is not whether a contract exists. It is what the contract is for, and how long it runs.
The deeper read: space on a fifteen-year term, or compute by the hour
The financing market has a phrase for the European model. Clifford Chance titled a briefing this July “The Data Center Lease Is the Bond,” and the phrase is precise rather than clever. A rated take-out on a stabilised European or American data centre is priced off the contracted lease cash flow, which means the rating of the bond is in substance a rating of the lease.
The structure is standardised. A fifteen-year initial term with tenant extension options, sized so the debt amortises or reaches an agreed balloon inside that term. An absolute triple-net lease, under which the tenant carries taxes, insurance and maintenance, produces a fixed bond-like rent. Fixed annual escalators. A single investment-grade tenant.
One practitioner put it more directly: a 100MW facility leased to Microsoft on a twenty-year triple-net lease is a Microsoft corporate bond with a building attached.
Applied Digital's Polaris Forge 3 transaction in May 2026 shows the shape at scale. Three hundred megawatts of critical IT load, a fifteen-year take-or-pay lease with an investment-grade hyperscaler, $7.5 billion of contracted base-term revenue and up to $18.2 billion including options. The building is incidental. The instrument is a fifteen-year obligation from a counterparty that will still exist in fifteen years.
HUMAIN is positioned explicitly as an exporter of AI compute rather than a domestic consumer of it. Its stated demand rests on a captive base of ministries, PIF portfolio companies and state-owned enterprises directed there through procurement preference, alongside two announced partnerships and a marketplace layer where third parties sell agents running on Saudi infrastructure.
The two partnerships are worth describing precisely, because the distinction this piece turns on applies to them. AWS entered a strategic partnership in May 2025, expanded in November 2025 to plans to provide, deploy and manage up to 150,000 AI accelerators in a Riyadh AI Zone, and is described as HUMAIN's preferred AI partner. xAI and HUMAIN signed a framework agreement announced on 19 November 2025 for the joint development of GPU data centres in Saudi Arabia, anchored by a flagship facility of 500MW or more.
That 500MW figure describes the planned capacity of the facility. It is not a disclosed commitment by xAI to take 500MW of capacity, and the two are different things. Neither arrangement has been disclosed with a contracted term, a take-or-pay provision, a minimum quantity or committed pricing. Both are real and publicly announced. Neither is the kind of instrument that carries a European financing.
There is a further feature of the xAI arrangement that a European lease structure would not present. HUMAIN invested $3 billion in xAI's Series E financing, becoming a significant minority shareholder. HUMAIN disclosed the investment itself in February 2026, and Reuters and the Financial Times reported it the same day. The exact percentage was never made public. Following SpaceX's acquisition of xAI in early February 2026, those holdings converted into SpaceX shares.
So the anchor partner was one the counterparty had put three billion dollars into. Nothing improper follows from that, and related-party arrangements are ordinary in sovereign-led industrial programmes. It does mean the commitment is not the same instrument as a fifteen-year obligation from an unrelated investment-grade tenant, and valuation reflects that difference whether or not anyone intends it to.
That is a different product from the European one, and the difference is duration. A colocation lease sells fifteen years of space at a fixed rent to one creditworthy tenant. Selling compute sells processing at a price, to whoever wants it, for as long as they want it. The revenue can be larger and higher margin. It behaves differently over time, because it can stop.
Saudi analysts have been direct about the consequence. A review of the HUMAIN financing programme observed that without anchor customers in the form of government workloads, hyperscaler commitments, AI lab contracts or long-term cloud agreements, lenders see speculative compute. The word is theirs.
Where the two sides of the capital structure diverge
Financing means debt and equity, and the same lease does different work depending on which is held.
For debt, the lease is the instrument. Lenders underwrite the tenant's credit rating rather than the asset, size the facility to the contracted term, and take out construction risk through a rated bond, a private placement or a securitisation priced off that cash flow. Where the contracted term is short or the counterparty unrated, the same asset supports less debt at a higher cost.
For equity, the lease is what makes the asset saleable. A fifteen-year investment-grade lease produces a cap rate, a set of comparables and an identifiable pool of buyers: infrastructure funds, listed REITs, insurance capital. An asset whose revenue is compute sold on short contracts has no equivalent valuation anchor and a thinner buyer list, because the buyer is underwriting a business rather than acquiring a contracted income stream. That is an exit consideration rather than a credit one, and it sits with whoever holds the equity.
In the Gulf the equity is frequently sovereign or sovereign-adjacent, and family office capital often arrives beside it rather than in place of it. Co-investing alongside a sovereign anchor and holding a minority interest in an asset the sovereign controls are different positions, and they resolve differently at exit.
This mismatch is not particular to the Gulf
The same duration problem has appeared in American and European colocation. Neocloud providers lease long-dated capacity from operators, fill it with GPU clusters, and resell that compute on short-duration contracts. A landlord signs a fifteen-year lease to a tenant whose own revenue is contracted for eighteen months.
The result is an asset financed against a counterparty that may not service the lease for its full duration. A neocloud tenant relying on one customer for most of its revenue means a landlord who looks diversified can hold indirect exposure to a single hyperscaler. Facilities built to one tenant's power density and cooling specification do not re-lease quickly when that tenant fails.
The maturity mismatch is what happens anywhere compute is sold rather than space leased. The Gulf runs the model at sovereign scale and without the intermediating layer.
Demand that can move
European demand is captive. Data residency requirements and latency-bound workloads mean the customer has to be physically present, which is why 83% of the pipeline is pre-let before a slab is poured. The workload cannot go elsewhere.
Gulf demand is portable. A training run does not care which continent it happens on. That portability is the commercial advantage, because it lets Saudi Arabia and the UAE compete on delivered power cost against markets that cannot approach it. It is also the exposure. Demand won on price can be lost on price, and a buyer who chooses Riyadh this year can choose Texas or Johor next year.
Cisco's survey work found most Saudi firms expect to use AI agents while fewer than a third hold meaningful GPU capacity. Domestic demand is real and underserved. It is not large enough to absorb the 1.3 to 1.5GW national target commonly cited for 2030, let alone the 6.6GW by 2034 that HUMAIN's chief executive has described as the company's own pipeline ambition. The build works if compute is sold abroad, which means the demand underneath it has somewhere else to go.
On state money, before anyone reaches for the word speculative
Nobody is financing this cycle on demonstrated demand. The United States appropriated $52.7 billion under the CHIPS Act with $24 billion in tax credits behind it. China's Big Fund III raised $47.5 billion, taking three rounds past $95 billion. HUMAIN's chief executive has put the company's own programme at roughly $77 billion.
Europe's number deserves a closer look. The European Chips Act carries a headline of €43 billion. The Commission's own direct allocation is €4.2 billion, of which €2.7 billion was moved across from Horizon Europe and €1.4 billion from the Digital Europe Programme. The remainder is expected to be mobilised from member states and private capital.
That is a headline reached by moving money between existing pockets. It is the same distinction I have been careful about all year, between announced project value and capital actually deployed, and it cuts in every direction. Europe is not underwriting its buildout on demand either. It is doing so against a slower constraint and a smaller cheque.
What the sovereign anchor covers
PIF is not going to default. That is worth saying plainly, because it removes the reflexive objection and leaves the more interesting question standing.
Sovereign backing addresses credit. It does not address utilisation. A fully performing loan secured against a half-empty campus still resolves poorly for whoever holds the equity, and in the Gulf the sovereign usually holds the equity. What the anchor genuinely provides is patience, and patience is not a small thing. Gulf capital can wait through a demand cycle in a way that capital borrowed against a political term cannot.
So the megawatts get built. The variable is which instrument sits underneath them. A fifteen-year triple-net obligation from an investment-grade tenant behaves one way through a cycle. A share of revenue from compute sold into a market where the buyer has alternatives behaves another. Both can work. They are not the same asset, and the market does not price them the same way.
Next: What happens when the grid says no, and what a real impairment looks like on a live project.
This is analysis of how these markets are structured and financed, written by someone who spends his time following how capital moves across borders and into infrastructure. It is not investment advice and not a recommendation on any asset, programme or jurisdiction.