Credit: Getty Images
15 Oct 2026, 10:44 pm
The global race to build artificial intelligence (AI) infrastructure is creating substantial demand for capital. For projects involving Chinese capital, equipment or technology, the financing and risk-sharing models developed for roads, ports and power plants are unlikely to be directly transferable.
‘Chinese elements’ can take many different forms, including capital, lending, equipment, technology, contractors and customers. The same is true of participation from other countries. What matters for bankability is not whether a project has a Chinese, American, European or other national element. It is what that element does, how critical it is to the project’s continued operation and how it interacts with the legal and regulatory regimes applicable to the project.
AI infrastructure sits at the intersection of infrastructure, technology and equipment finance. Unlike traditional projects, where assets typically share similar risk profiles and economic lives, AI facilities combine physical infrastructure with rapidly evolving technology and complex contractual arrangements. As a result, bankability depends less on the value of individual assets and more on whether the entire ecosystem can continue functioning together over time.
A modern AI project may include land, buildings, grid connections, dedicated power generation, cooling systems, fibre networks, computing equipment, software licences, customer contracts and specialist operating arrangements. Although these assets collectively support revenue generation, they may be owned, financed and operated by different entities.
This fragmentation can make commercial sense. Long-term infrastructure investors may be willing to finance land, buildings and energy assets through long-dated debt, while computing equipment may be funded through shorter-term facilities that better reflect its depreciation profile.
However, dividing ownership and financing arrangements can create structural weaknesses. A lender with security over the property-owning entity may have no rights over the customer contracts that generate revenue. An equipment financier may control expensive servers but lack any practical ability to access or operate the data centre. Critical software, intellectual property or technical expertise may be held by affiliates outside the financing structure altogether.
These challenges become more concentrated in cross-border projects. The local project company may own the facility and hold the necessary permits, while procurement, technology services, intellectual property and sponsor support sit in other jurisdictions. In such circumstances, even comprehensive local security may not provide lenders with an independently operable business if issues arise.
Consequently, financiers need to evaluate projects not simply on what the borrower owns, but whether the business can continue operating and generating revenue if relationships break down or sponsor support disappears.
The physical and technological components of an AI facility often have very different economic lives.
Buildings, substations, cooling systems and grid connections may remain productive for decades. Computing equipment, by contrast, may become obsolete within a much shorter period. This creates an inherent mismatch between debt tenor and collateral value.
As a result, AI projects are likely to require layered capital structures. Long-term project debt may support the core infrastructure, while separate facilities finance computing equipment. Equity and subordinated capital may be needed to fund recurring upgrades and technology refresh cycles.
That complexity creates competing incentives among creditors. Project lenders typically seek to preserve the value of an operating business. Equipment financiers may have an incentive to quickly realise rapidly depreciating hardware rather than preserve it within the operating business.
For this reason, arrangements between different lenders need to do more than simply determine who gets paid first. They should also set out how the project would be supported, restructured or sold if it runs into financial difficulty.
Cross-border deals add another layer of risk. Restrictions relating to foreign exchange, guarantees, remittances and enforcement can all affect whether contractual protections are practically effective when they are most needed.
Traditional project finance often distinguishes clearly between construction risk and operating risk. AI infrastructure rarely fits that model.
A building may be physically complete before it receives power. Grid connections may be energised before computing equipment arrives. The facility itself may be operational while advanced processors remain delayed by export controls, customs issues or sanctions-related restrictions.
As a result, construction progress does not necessarily correspond with a reduction in credit risk.
A data centre may be physically complete yet remain commercially incomplete because it lacks sufficient power, computing capacity or customer acceptance.
Completion tests therefore need to extend beyond physical construction milestones. They may need to incorporate energisation, equipment installation, performance validation, customer acceptance and commercial operation.
Similarly, the conversion of construction debt into long-term financing should depend on the project's demonstrated ability to produce the cash flows assumed in the financing model, rather than the mere completion of a building.
The viability of any AI infrastructure project ultimately depends on two critical inputs: electricity and computing technology.
In relation to power, the availability of electricity alone is insufficient. Lenders require assurance that the borrower has an enforceable right to access adequate capacity for a period consistent with the financing tenor. That assessment may involve analysing grid connection arrangements, curtailment rights, tariff risks, minimum purchase obligations and transferability following enforcement.
Where generation assets or grid infrastructure are separately owned or financed, creditors will also need confidence that these arrangements can survive distress elsewhere in the project structure.
Technology creates similar challenges.
The borrower must not only acquire the necessary equipment during construction but also maintain, replace and upgrade it throughout the life of the financing. Advanced computing increasingly sits within overlapping export-control, sanctions, investment-screening and technology-security regimes.
Equipment available today may become difficult to replace in future. Alternative technologies may be legally accessible but insufficiently compatible with customer requirements or performance expectations.
Bankability therefore cannot be assessed solely at financial close. Financiers must consider whether critical technologies can continue to be sourced, serviced and substituted throughout the financing period.
Financing documents must strike a balance. They need to permit commercially necessary upgrades while preventing changes that materially erode performance, collateral value or revenue generation. Overly restrictive provisions risk leaving facilities dependent on obsolete technology, ultimately reducing the value of the lender's security.
Like any project financing, AI infrastructure ultimately depends on reliable cash flow.
Customer contracts are critical because they determine when revenue starts flowing. Lenders need to understand when customers are required to start paying, whether they are committed to buying a minimum amount of capacity, and the circumstances in which they may be able to reduce their commitments or walk away from the contract.
Support arrangements also deserve close attention. Guarantees, completion support and other financial backstops only provide real protection if they apply when problems actually arise. Even a financially strong sponsor may offer limited comfort if the financing structure leaves lenders with little practical recourse when things go wrong.
Risk analysis is becoming more complicated as AI projects increasingly rely on a mix of funding sources, including project loans, bonds, leases, private credit and other financing structures.
While these arrangements can make it easier to raise capital, they can also make it harder to identify where the real risks lie. What appears to be a diversified financing structure may ultimately depend on the same small group of customers, technology providers or sponsors. Understanding those underlying dependencies can be more important than simply counting the number of contracts or entities involved.
One unusual feature of AI infrastructure is that commercial success can increase rather than reduce certain risks. As a facility grows in scale, computing capacity or strategic importance, it may attract greater regulatory and national-security scrutiny.
For investments with Chinese elements, that scrutiny may extend well beyond the host jurisdiction. The source of capital, equipment and technology, together with relationships among project participants, may bring projects within the scope of competing national security regimes.
For lenders, this matters because regulatory sensitivity does not stop at investment approval. It can affect refinancing, changes of control, enforcement and the identity of potential replacement owners. A restriction that reduces the buyer pool can ultimately become a recovery-value issue.
The value of a data centre rarely lies in any individual asset. Instead, it resides in a functioning combination of land, power, equipment, software, customer contracts and operational expertise. Legal control over the assets does not necessarily give a secured creditor effective control over the operating business.
A creditor may be able to enforce against land, buildings and equipment yet still be unable to operate the business if power rights, software licences, technical knowledge, customer relationships or regulatory approvals cannot follow the assets.
These challenges are particularly relevant in projects where intellectual property, technical support, spare parts or customer relationships may sit outside the local project structure.
As a result, enforcement planning should begin at the structuring stage rather than when a default occurs. Critical technology and operating rights may need to be held or licensed at project level, direct agreements should provide realistic step-in rights, and liquidity arrangements may be required to preserve power, cooling, cybersecurity and specialist personnel during a restructuring.
AI infrastructure requires lenders to think beyond conventional asset-level security. At its core, AI infrastructure finance involves funding an interconnected bundle of assets, rights and dependencies. Land, power, servers, software and customer relationships may all sit in different entities, depreciate at different speeds and be governed by different legal regimes. Their standalone value matters less than their ability to function together as a single operating business.
For investors and lenders considering AI opportunities, this requires a shift away from traditional project-finance assumptions. Bankability can no longer be assessed solely through asset-level security or construction risk analysis. It must be evaluated across the full lifecycle of the project, taking account of technology replacement, regulatory change, operational dependencies and enforcement realities.
The central question is no longer simply whether capital, equipment and technology can enter a project. It is whether those inputs can continue to support the business over time, whether they can be lawfully replaced when needed, and whether the enterprise can ultimately survive if the original sponsor, customer or technology model falls away. That challenge is likely to define the next generation of cross-border AI infrastructure finance.
Written by
