Q&A: LBBW - Jan Weismüller on the realities of data centre financing, risk conversion & Europe’s fragmented market

 AI has become a buzzword, or rather a phrase, that dominates most conversations of late, and it is only amplified by the astronomical amounts of capital that is being pledged towards this sector. While the majority of the public discourse centers around the use case of AI and its integration in everyday life, the data centres that are needed to facilitate this integration are often an afterthought.

With or without AI, data centres play a pivotal role in modern society, and their importance is only growing with the rapid push to digitize. Financing this immensely capital hungry sector is a challenge unto itself, but it’s further complicated by the nippy technological innovations and geopolitical tensions which have contrived the need for “digital sovereignty”.

We sat down with LBBW’s Global Head of Infrastructure, Energy and Transport Finance, Jan Weismüller to understand the considerations and complexities associated with arranging capital for this sector.

The German bank recently played a pivotal part in the financing for CloudHQ’s 315MW data centre cluster in Virginia in addition to a considerable pipeline of deals, within the sector, in other parts of the US and Europe.

Weismüller goes over the differences between the US and European markets and how commercial banks navigate the various risks to arrange large chunks of capital – sometimes multiple billions of pounds in just one deal – for these assets. He also weeds through the hype vs the realities of the current market and Europe’s potential for building up its data centre infrastructure to be on par with the US.

How do financing models differ in US and Europe?

Weismüller: The US is quite a mature market when it comes to data centre financing, it has a deep capital market capacity. There, stabilised assets – following construction – are pushed into the capital markets mainly through ABS structures. It’s a very large, highly liquid and an attractive market for stable, long-term infrastructure financing.

There is also a size difference between the US and Europe. Hyperscale market in the US is more than twice the size of the European counterpart, moreover, the assets tend to be more centralised and consolidated. This translates to larger ticket sizes and larger transactions.

Tenor on commercial loans are slightly shorter in the US. Mostly mini-perm structures that get refinanced into the capital markets soon after construction. Lease terms on the other hand tend to be longer in the US. The US also has structural advantage in this regard as in most cases you tend to get higher quality counterparts.

The market in Europe, in contrast, is more fragmented.

Will European markets evolve to replicate the American ABS model for long term data centre financing, or will new ones emerge to cater to local needs?

Weismüller: Replicating US patterns in Europe is difficult because the financial markets for data centres are very different here, the landscape is much more fragmented. We are slowing moving towards the US model for high quality sponsors, but we can’t simply replicate the US model because we also have a stronger ESG and regulatory oversight here.

Because of this, we will need to develop a hybrid model and there is also trend towards energy linked structures which include corporate PPAs and the incorporation of onsite generation along with battery storage.

Power is the key factor in sourcing locations for data centres in Europe.

Here, we are also seeing the involvement of promotional banks, meaning legal entities carrying out financial and development activities by mandate of a member state at central, regional or local level. There is a conversion of risk appetite at the top of the market which will have to reflect the overall European regulatory framework.

How would lenders go about assessing the risk profile for projects that include a power generation element?

Weismüller: The general assumption is that you do require long term power contracts to be in place. It’s a key element of the risk profile. The primary considerations are tenant risk, construction/delivery risk, location and last but not least power availability. These are the elements with which we assess the viability of a data centre rather than pure asset value.

Power is typically precontracted and we need to have a very clear visibility of this. But, in the past we have seen deals in the market where the power element was a key element of the project delivery, where the developer also builds their own power supply. In these cases, the power generation assets may or may not be financed separately.

The proposal to build gas plants to power data centres is common in the US but is it feasible in Europe?

Weismüller: Commercial lenders have been lending to these projects in the US and it’s also in discussion in Europe. Here, it’s definitely a challenge given the strict ESG metrics and because of this, not every lender will include them in their portfolios but it’s a reality the market has to face with as there is no other viable alternative at the moment both with respect to power availability and delivery reliability.

How do lenders view technology risk?

Weismüller: We don’t see that technology risk is often taken as a part of the risk bundle.

Banks need to have a view on the involvement to cooling systems where there is a lot of innovation of late to cater to the AI processing conditions but at the end of the day the customer/tenant signs off on the lease contract. The developer and the customer are ultimately responsible for the technology. If there is an end use contract in place, the lenders don’t have the risk of the tech not being suitable for the customers.

The risk on the technology side is completely borne by the hyperscaler, the computing power aspect is not a part of the financing package. The way we look at it, the data centre is a fridge that provides housing, cooling and security. What they put in the fridge is of less concern to the lenders.

Is there an AI bubble and if so, how will that affect build out of hyperscale capacity?

Weismüller: First, we need to differentiate between the different AI use cases. Most of what we see now will come back to the hyperscalers because they remain the primary work channel for AI workloads and GPU capacity.

When people talk about AI its mostly in reference to training centres for models which is a totally different use case compared to hyperscalers which will come in for AI inference. Inferencing happens close to the load centres, close to the customers and would ideally have low latency.

Training on the other hand can withstand high latency and be located in remote places which have ample grid access and where the power is cheaper.

Anyone using AI at home or for a business will need low latency and because of this, it will be decentralised.

This is where the FLAPD markets in Europe come into focus. These are the tier 1 markets, then you have tier 2 and tier 3 markets and finally the very remote hyperscale locations like northern Norway where you have cheaper power for training models.

LBBW mainly focuses on higher tier markets in Europe because we see a longer use case in terms of marketability of these data centres.

What are the considerations for financing AI inference facilities?

Weismüller: It’s all blended really. Every hyperscaler will have the option for AI integration. Think about Copilot or Siri, they are not housed in special facilities that only cater to AI processing. This doesn’t mean all the customers will have use for AI applications. Any application you use will need to have a home. Most hyperscale facilities will have the infrastructure to house AI applications so it’s hard to separate these for the purpose of financing.

Do you see the data centre demand shifting between the US and Europe?

Weismüller: There is a really high demand in Europe for hyperscale, you see this across all the metrics in the market, in terms of lease rates, the availability of power and land. Some European markets are closed to data centre development because there are constraints on build outs due to the pressure on the grid. There is a clear demand overhang as of now.

The market is more active in the US, much more buildout mainly because power is more widely available. Recently, some large-scale AI projects were cancelled in Europe so it’s clear that there is no demand shift towards Europe.

With that said, competitive pressures in industry and research, but also sovereignty concerns about data security and integrity will put pressure toward further European data centre expansion.

If Europe wants to be part of the data centre push, there need to be more favourable development conditions. Issues surround permitting and power availability need to be addressed.

We see a lot of potential for this sector especially within our core markets. At LBBW we have demonstrated that we are willing to lead these transactions.

This is a very capital hungry sector, and we need to diversify the source capital to accommodate the rate of consumption which is another main focus for us.