The Fund is expected to capitalise more than US$8 billion of hyperscale development assets across the US in partnership with Stream Data Centers, a leading provider of digital infrastructure solutions.
Principal Real Estate, which manages over US$100 billion in real estate assets globally, is the dedicated real estate investment team of Principal Asset Management. It has been investing in data centres since 2007, with data centre assets under management and a construction pipeline of US$11 billion.
In a recent interview, Sebastian Dooley, Senior Fund Manager at Principal Asset Management, discusses the challenges and opportunities associated with building data centres, particularly in arid areas like Phoenix, Arizona, where Principal Asset Management’s fund is involved in a data centre project.
The availability of water and power is a key challenge. Initially, in data centre development, the first big challenge was that data centres used too much electricity. They are giant fridges, Dooley says.
See also: The AI trade in Europe is about data centres and power
Since data centres consume excessive electricity, DC developers have explored alternative forms of cooling. Water cooling became a popular method to reduce electricity consumption.
“We’ve done data centre development in dry arid areas where water is an issue, and we are using a closed-loop water system. You condense and capture water, resulting in less waste. There is an additional cost, but we monitor water with water usage efficiency (WUE). It’s similar to power usage efficiency (PUE), but for water. We try to keep WUE as close to zero as possible. For our newbuild in Phoenix, we are getting the WUE down to similar amounts of wastage as a residential house in the US,” Dooley says.
The onus is on the capital provider, who ensures it’s building assets for the future. “The onus is on us to show that the data centre is sustainable,” he adds.
See also: AI data centres give private credit its mojo back
In California, the home of Silicon Valley and the big tech companies, the power infrastructure is stretched. “We are finding that sourcing power, and power infrastructure such as the transmission lines, is a big issue,” Dooley says.
Phoenix is neighbouring California, which is more sparsely populated, has lower land costs, and is ideal for solar power generation. Think Johor and Singapore, but with no international border.
Power consumption to rocket
A Bloomberg News analysis of wholesale electricity prices for tens of thousands of locations across the US, published on Sept 30, reveals the effects of the AI boom on the power market.
Electricity prices were tracked and aggregated monthly by Grid Status, an energy data analytics platform. Bloomberg analysed this data in relation to data centre locations, from DC Byte, and found that electricity now costs as much as 267% more for a single month than it did five years ago in areas located near significant data centre activity.
Approximately two-thirds of the power consumed in the US is generated from a state or regional grid, where the system operator manages the trading of energy, according to Bloomberg. These wholesale commodity costs are passed on to households and businesses through their utility bills, which also include other charges to maintain and expand the network, Bloomberg reports. That can affect even customers who aren’t in proximity to a data centre, since their energy relies on the same grid.
In the analysis, wholesale electricity prices are measured in real-time on a daily basis by Locational Marginal Pricing (LMP) points on the power grid, referred to as nodes. LMPs are primarily made up of the cost to produce the energy and congestion. Analysing data from 25,000 nodes used by seven regional transmission authorities, Bloomberg News estimated the change in wholesale prices in the lower 48 states since 2020.
For more stories about where money flows, click here for Capital Section
“Their power needs are only set to accelerate. In recent weeks alone, Nvidia Corp said it will invest as much as US$100 billion in OpenAI to support new data centres, while Microsoft Corp. struck a multi-year deal worth almost US$20 billion to obtain cloud computing power from Amsterdam-based Nebius Group NV, using a New Jersey data centre. OpenAI and Oracle Corp. forged a partnership to build 4.5 gigawatts (GW) worth of data centre capacity, enough energy to power millions of American homes,” Bloomberg reports.
Globally, data centres are expected to consume more than 4% of electricity by 2035, according to BloombergNEF. Put another way: If the facilities were a country, data centres would rank fourth in electricity use, behind only China, the US and India, Bloomberg says.
In the US, power demand from data centres is set to double by 2035, to almost 9% of all demand, according to BNEF. It could be the biggest surge in US energy demand since air conditioning caught on in the 1960s, market watchers reckon. That comes as the grid is already struggling to update ageing infrastructure and adapt to climate change, they add.
As part of their sustainability push, the big tech companies have announced joint ventures with nuclear power providers. In October 2024, Amazon announced an agreement with Energy Northwest, a consortium of state public utilities, to enable the development of four advanced small modular reactors (SMRs). The reactors will be constructed, owned, and operated by Energy Northwest and are expected to generate approximately 320 MW of capacity for the first phase of the project, with the option to increase to a total of 960 MW — enough to power the equivalent of more than 770,000 US homes. These projects will help meet the forecasted energy needs of the Pacific Northwest, beginning in the early 2030s, an Amazon press release says.
In Virginia, Amazon signed an agreement with utility company Dominion Energy to explore the development of an SMR project near Dominion’s existing North Anna nuclear power station. This will bring at least 300 MW of power to the Virginia region, where Dominion projects that power demands will increase by 85% over the next 15 years.
In September 2024, Constellation Energy announced the signing of a 20-year power purchase agreement with Microsoft, setting the stage for the restart of Three Mile Island Unit 1, to be known as the Crane Clean Energy Center.
Financing data centres
On a rental basis, data centres’ yield-on-cost is somewhere between 8% and 12%. “On average, it will take you 10 years to recover your cost, depending on what you’re building,” Dooley says.
The cost of debt is likely to be 2.25 percentage points above the base rate. “This price is for a very compelling development. Rates are very competitive compared to other development asset classes, especially if you’ve signed your lease with a triple-A-rated big tech firm,” Dooley says. It typically takes 18–24 months to construct the shell, along with the power infrastructure.
The development cost is usually based on the price per kilowatt-hour (kWh). “On average, it costs EUR1.5 million ($2.2 million) per megawatt-hour (MWh) to buy land. The build cost is roughly EUR12.5 million per MWh, and EUR1.5 million per MWh for construction, or EUR14 million per MWh. Hence, a 1GW data centre would cost EUR1.4 billion.
The new data centres usually have a capacity of 100 million MW to 120 million MW, as hyperscalers require such a high capacity. “These are very sizeable projects from a capital outlay perspective. If we’re looking at a European basis for cloud, normal-sized assets are 100 to 120 MW that are being announced. The scale of capital required for the industry has grown massively,” Dooley says.
The cost includes the shell, power supply, additional generators, and power infrastructure, but does not include the chips. The data centre operator supplies the chips.
With so much infrastructure, depreciation and keeping the data centre relevant are likely to be key focuses for capital expenditure. However, depreciation of a data centre isn’t on a straight-line basis, as different parts of the infrastructure have different lifespans and life cycles. The batteries last seven to 10 years. Cooling systems can last anywhere from 12 to 20 years. The backup generator systems can last for 25 to 30 years. During the life of the data centre, different infrastructure will have to be replaced regularly.
“Because data centres need to be able to offer very reliable services and facilities constantly, the infrastructure replacements are done proactively, and the maintenance is done regularly. You’re constantly investing in these assets. If you’re looking at a market like Singapore, where it’s difficult to build new data centres, you’ve got to continually reinvest in the data centres when cooling systems and the floor loading standards change,” Dooley says.
“Fundamentally, your long-term value for these assets is from having a reliable power supply, connectivity, and the permission to run as a data centre. These data centres are in locations where it’s becoming increasingly difficult to create new data centres and demand is continuing to grow,” he adds.
Moratoriums
Like Singapore, many developed economies have moratoriums in place. Currently, Ireland has a moratorium on building more data centres. Amsterdam and the Netherlands have also undergone several moratoriums.
Generative AI can be in data centres anywhere in the world. These data centres don’t need to be in financial hubs like London and Singapore. On the other hand, many countries have data sovereignty laws in place.
“A lot of the model training will happen where it’s most efficient, where power costs are less, where land costs are less. The inference side is where people actually start utilising these algorithms. Inferencing AI will have to happen in-country. That’s where location really matters for these facilities,” Dooley says.
Cloud is a very mature business model for data centres, providing owner-operators with the cash flow to invest in AI. “Everyone uses the cloud. The profitability and the business case for cloud are well known. In locations adjacent to data centre hubs such as Phoenix, you have the upside coming through from AI inference and you get downside protection coming through from the cloud,” Dooley says.
With the capex requirements and the building of new data centres, owners of stabilised data centres are likely to seek capital solutions. Dooley points to NTT DC REIT as a successful capital solution for an owner-operator with a newbuild pipeline. Does that mean that more data centre owners are likely to explore capital solutions such as a listing on the Singapore Exchange? NTT DC REIT and other REITs looked to a listing as a solution, as Dooley says.
