Megawatts & Megabytes: How Trade Finance Fuels the AI Boom

Over the last few months, a defining trend has emerged in data center financing, involving the use of global trade solutions - notably Letters of Credit (LCs) - to secure access to vital development resources. Specifically, "power LCs" have become critical tools for developers to secure energy access, a pressing bottleneck for data center development, particularly for AI-focused hyperscalers, which can be up to ten times more energy-intensive than traditional data-centers.

Until recently, this trend was primarily concentrated in the US. Are these solutions now arriving in Europe? Drawing on recent landmark transactions, Aydin Akkaya, Head of Global Trade for Telecom, Media, Tech, Arnaud Stevens, Head of Global Trade Americas, and Rebecca Lo, Trade & Working Capital Solutions Originator, explain the mechanics of these global trade solutions across all the data center value chain.

Arnaud Stevens: Power access has become the defining constraint for hyperscale growth, whereas until last year, the market's main focus was securing chips. What explains this explosion in energy demand isn't just the growing number of data centers, but the power intensity required for AI, which is up to ten times higher than that of traditional data centers.

To build large, sophisticated data centers, developers must then secure reliable, high-density power, knowing that any delay can cost billions in lost market share or stranded capital. Consequently, power demand is exploding while new capacity lags and grid interconnection queues drag on for years.

To secure energy access, we are structuring large, syndicated power LC facilities in favor of grid operators and utilities, allowing developers to back their power purchase agreements (PPAs). Ultimately, these power LCs act as a selective filter in the development process, proving how advanced and serious a project really is. Grid operators and utility companies want to ensure there is significant credit support in place before allocating massive power volumes. They need protection in case a developer is unable to offtake the committed volumes or fail to connect the data center to the grid. In the US, this is currently the number one issue.

Rebecca Lo: For this same reason, we also structure performance LCs, which power generation companies or grid operators - for instance - can draw upon, should the data center developer fail to perform. While drawdowns are rare - usually because the beneficiary and the applicant work closely to meet key developmental milestones, or because struggling developers sell their projects to other operators - these LCs provide a crucial layer of reassurance that suppliers highly value.

Arnaud Stevens: Essentially, we adapted an existing technology: the large, syndicated LC or borrowing-base facilities traditionally used for commodity merchants. Previously, LCs were issued under developers' large revolving credit facilities (RCFs). However, while RCFs have sublimits for LCs, the growing scale of these projects means developers want to preserve their RCF "dry powder" for funded, cash-drawing needs.

This is why we designed specifically targeted, syndicated power LC facilities. A prime example is the landmark facility we structured for Switch - a true trailblazer in establishing a massive, dedicated power LC facility entirely separate from traditional corporate RCFs.

It is also worth noting that, like RCFs, LCs are off-balance-sheet instruments that are rarely drawn. Consequently, they consume far less cash liquidity, allowing banks to issue the massive volumes required for these giant projects.

Rebecca Lo: The average tenor for these power LCs is usually one year. They are structured as one-year evergreen LCs, meaning they automatically renew annually. Typically, they remain in place throughout the construction phase, which generally spans two to three years.

Aydin Akkaya: Yes, even if the US remains the biggest market for data centers, cloud, AI for the time being. Currently, in Europe, these needs are usually handled on a bilateral basis, as sponsors or clients typically have enough capacity within their existing corporate facilities. We haven't yet seen the same syndicated structures as in the US. However, given the massive scale of upcoming projects in Europe - such as gigafactories, particularly in France - we anticipate a growing corporate demand for these exact types of syndicated guarantee facilities.

Aydin Akkaya: Yes, and this is a crucial question. Power is actually one of what I call the “4 Ps” - the four vital issues for data center developers, alongside permitting, procurement, and people.

Permitting, notably, is highly challenging in Europe, especially in France, where it takes three to six years to get a data center built and operational. During this timeframe, developers must engage with the local ecosystem to get their projects approved, navigating sustainability regulations and local community concerns. The overall costs associated with this permitting phase can be massive if the project is delayed by the permitting phase.

To mitigate these delays, developers are increasingly negotiating fast-track clauses with local authorities or governments, for which they must post financial guarantees. This is precisely where we step in. If the developer fails or walks away, the beneficiary can draw down on the guarantee.

Aydin Akkaya: Procurement, indeed, has become a growing bottleneck. There is unprecedented demand for every single component necessary to deploy cloud and computing capacities - from electrical infrastructure, cables, and wiring, to batteries, power generators, substations, memory and chips. Consequently, suppliers are demanding substantial advance payments - representing 10% to 30% of the total contract value - sometimes two to three years ahead of delivery or even production. We help developers navigate this by structuring tailored facilities to finance these advance payments directly on their balance sheets.

Here again, we have adapted structured trade products widely used in commodity finance. A prime example is the landmark Mistral AI transaction, which we funded through an offtake-backed GPU finance facility. To transition from renting third-party computing capacities to owning their own hardware, Mistral set up a dedicated SPV to purchase high-performance GPUs and install them in a specialized, high-density AI factory near Paris operated by Eclairion.

To fund this, we structured a five-year facility modeled exactly like a traditional Pre-Export Finance (PXF) structure. Instead of underlying physical commodities like oil, cocoa, or copper, the asset here is computing capacity pre-sold to dedicated offtakers, whose contractual payments directly service the debt. This structure pioneered in the US with players like Lambda Labs and Coreweave, and it is now actively spreading across Europe and Asia.

Arnaud Stevens: Additionally, bills of exchange are instrumental in securing highly sought-after data center equipment by allowing developers to extend their payment terms. Inventory and equipment financing solutions also offer vital payment guarantees to tier-one suppliers of gas turbines or specialized electrical equipment. When you consider that developers must wait up to two or three years to receive custom equipment from manufacturers like Siemens or Schneider Electric, these trade structures provide the long-term financial security needed to lock in vital production slots early.

Aydin Akkaya: Indirectly, YES! We do not have a direct solution to solve the shortage of skilled labor of the overall sector but we have talented people within Global Trade that understand the sectorial dynamics and are ready to deploy expertise throughout the value chain of AI.