S&P Global Data Center & Energy Innovation Summit: Lending and investing

The 2025 S&P Global Data Center & Energy Innovation Summit brought together industry experts on lending and investing to share their views on the rapidly changing landscape of the data center market and their expectations for the future. The conference provided a platform for experts to share know-how and strategies, and to discuss trends in various facets of the market. While there are some isolated pockets of concern, the market as a whole looks poised for continued expansion.

The data center market has seen unprecedented growth in the last two years, largely driven by AI workloads. Capital expenditure by hyperscale companies has skyrocketed, and experts expect over four times the expenditure in 2025, compared with 2024. The industry is also seeing many changes. The lack of power is making investors and lenders wary about renewables, even though companies maintain decarbonization goals. There is also a shift in the allocation of capital as ever-progressing GPUs become the linchpin of competitive performance. Risk is also evolving as data centers move into secondary and tertiary markets, where latency could become a concern in the long run. Overall, the experts agree that the explosive growth is not indicative of a bubble in data center builds, as the industry is relatively flexible and can halt builds if demand collapses. However, there are portions of the market that should remain alert to changing winds.

How have markets changed in the past two years, and what is the outlook for the future?

Equity market

For the past two years, AI has been a critical driver of share performance. Hyperscale tech stocks such as Alphabet Inc., Amazon.com Inc., Microsoft Corp. and Meta Platforms Inc. have made up the bulk of that performance over the last two years. The reason for this is tied to capex and the quantity of capital dedicated to building data centers that can accommodate AI modeling and large numbers of high-density GPU chips. Considering the numbers alone, looking back to 2019, those mega-cap tech stocks spent about $80 billion in capex in aggregate that year. This year, they are on track to spend over $365 billion. This has translated directly to the success of companies like NVIDIA Corp.

Green bonds and renewables

The demand from institutional investors globally is still strong; however, there has been a sharp decline in the green bond market in the last year. Through August, the global green bond market has declined by about 30%. This could be due to shifts in US attitudes toward renewables. While the future is uncertain, it is clear that the largest institutional investors are maintaining decarbonization goals. In addition, given the amount of capital expected to be deployed in the data center space over the next five years, there is a need to diversify funding sources and broaden the investor base. This is where green bonds can play a role.

On the other hand, this significant expansion of energy infrastructure will likely result in some of the hyperscalers deviating from their climate targets to some extent. It seems that no single technology will be able to provide the necessary power, and companies are moving toward what they can attain in the shortest time frame.

Lending

There has been a massive increase in lending. Forecasts show a threefold increase in project finance construction from 2024 to 2025. With a hyperscale tenant, it is typical to see a roughly 80% loan-to-cost ratio on the construction of the data center, although this may overstate how much is being funded by debt, as the 80% estimate covers just the cost of the data center shell. The power supply and the equipment inside the data center, including servers, are not included in this estimate. The cost of data center construction is about $15 million per megawatt. This could be, however, $50 million to $100 million per megawatt when considering the full buildout, with chips and the equipment that goes into the shell. So, the actual spending being debt-financed in the data center is quite low in relation to the full project.

Risk is a function of data center location and purpose

Location, coupled with the facility's function, is of paramount importance when assessing the risk involved with a data center buildout. The first consideration is the type of facility. AI training facilities that use GPUs have a depreciation consideration. Banks cannot provide long-term assumptions; they must take more of a five-year view in terms of value. Neoclouds tend to use a six-year depreciation on GPUs. Companies will argue that they still have legs, but banks must take a relatively punitive view on depreciated value.

As an example, there is a lot less risk of obsolescence for a cloud provider in Northern Virginia than an AI training facility in western Texas or rural North Carolina. In addition, tertiary markets, particularly more rural markets, can be seen as riskier than sites closer to population centers due to concerns about latency. While these sites can be used for AI training, it is not clear that these sites will be able to transfer information fast enough to support inferencing once that becomes the primary need. Banks generally find it unwise to invest against cash flows that are assumed after the initial maturity of the loan.

In the first six months of 2025, the insurance industry saw upward of $100 billion in insured losses, mostly from severe convective storms and wildfires. For example, Texas, which has had difficulty with extreme weather events in the past, tends to be seen as more risky than milder climates. Stability of the grid at the regional level is also a consideration. Areas where the grid is known to struggle during environmental extremes without data centers are a riskier proposition than areas with a more robust infrastructure. For example, an area that is prone to hailstorms but also heavily reliant on solar power has risk built in prior to the presence of a data center.

Does the AI market suggest a bubble?

From an investing perspective, the word "bubble" evokes the thought of a massive lack of profitability in the system. That doesn't seem to be the case when considering the leaders of the space, such as NVIDIA and the hyperscalers. There will be winners and losers, but the starting point must be considered, and we are starting from very steady footing regarding big tech. In terms of lending, issues may arise when the current construction loan needs to be refinanced. Much of what is being seen in the space is three-year interest-only debt; this is not an issue for the hyperscalers. On the other hand, for smaller tenants, it is unclear where the refinancing market will be if something goes wrong in the next five years.