The already rapid construction of datacenters in the US is struggling to keep up with the growing needs of artificial intelligence and cloud storage — and now, tariffs are amplifying those hurdles.

Capital expenditures by the five largest cloud infrastructure providers increased by almost 60% in the second quarter of 2024 year over year to $55 billion, according to data from S&P Global Market Intelligence, and most of that spending is directed to the building of datacenters that host AI capabilities.

Market Intelligence data also shows the datacenter footprint is projected to grow more than 58% — to more than 480 million square feet — between 2024 and 2029.

“The datacenter industry has been growing dramatically in the past several years even before the AI explosion two years ago,”said Liu Perkins, S&P Global Market Intelligence's senior research analyst on datacenter infrastructure.“The cloud datacenters have already been in strong growth. But on top of that, this AI growth has triggered a big change, [positioning] the North American market in the leading position, and now the rest of the world is trying to catch up as well.”

S&P Global is the parent company of the Journal of Commerce.

Speaking at the Journal of Commerce Breakbulk and Project Cargo Conference in New Orleans last month, Beth Branch, president and CEO of the Port of New Orleans and CEO of the New Orleans Public Belt Railroad, said that datacenters offered a“really massive”opportunity for breakbulk and project cargo over the coming years. She referenced Meta's construction of a $10 billion AI datacenter in northeast Louisiana — which will be Meta's largest site globally — as an example of the scale of data projects in the pipeline.

Growth has been driven further by government investments and incentivization.

“The speed that datacenters are being built has become condensed,”a project cargo shipper told the Journal of Commerce.“The velocity that projects like this can be done in is amazing when there is government money available that can be thrown at them.”

Dan Thompson, an S&P Global research analyst with a focus on datacenters, said that the industry is“operating behind the curve”in terms of fulfilling demand, and while companies are aggressively building datacenters, questions remain whether they can truly meet the infrastructure needs for emerging technologies such as AI.

Sourcing hurdles could stunt growth

The demand for datacenters has led to a significant manufacturing backlog, which some analysts say could be impacted by tariffs on Chinese products, thus affecting project timelines and import costs.

AI and cloud storage are forcing exponential growth in the datacenter industry, and in an industry where manufacturing is already a significant pain point, tariffs on Chinese-sourced infrastructure components are a taboo topic for many datacenter engineering, procurement and construction companies (EPCs). According to S&P Global Market Intelligence, AI's growing use of graphic processing units (GPUs) consumes as much as 10 times more energy per chip than traditional central processing units (CPUs).

While many of the infrastructure components are subject to manufacturing hurdles, transformer availability continues to pose a critical issue, said Thompson. The datacenter projects — costing about $10 million per megawatt — usually take about 12 to 18 months in the US, with two to three years needed in other countries, he said.

While the US tends to get the projects done faster, there are delays — especially on the manufacturing side. As such, many project managers have begun to get ahead of their procurement schedules, opting to head off any growing backlog.

“We see a lot of providers, operators, try to secure the supply of key parts and equipment way ahead of time — putting orders in a year or two ahead of their schedule to try and secure the supply,” Perkins said.

The main heavy-lift components needed for datacenters are cooling infrastructure — such as large-scale air conditioners — and backup power infrastructure such as batteries, diesel generators and transformers.

“A lot of the delays are in getting these large-scale transformers,”said Thompson.“It seems to me that manufacturing can't keep up. Along with diesel generators — the leads times on those keep getting longer.”

Both Perkins and Thompson agreed that tariffs have exacerbated concerns around sourcing, which was already considered a difficult topic before the Trump tariffs were implemented in April.

“Suddenly, no one wants to get very precise as to where all of this stuff comes from,”said Thompson.“Most companies have some manufacturing in the United States, but it may just be assembly.”

Destination factors a consideration

Much of the datacenter construction is taking place in rural areas of Ohio, Wyoming, Virginia and Texas. That means things such as the availability of power and roads designed to handle heavy cargo pose significant infrastructure challenges.

Much of the equipment is brought in and assembled locally before installation, but large components continue to pose hurdles for the projects. Additionally, beyond the roads, there is a need for terminals designed to handle sufficient pounds per square foot for the heavyweight components being used.

While many of the existing datacenters were constructed near hubs and urban areas, there is a growing need to serve more rural areas with lower network connectivity. As such, the development of centers in those areas often presents a catch-22 — depending on existing utility infrastructure but demanding increased output from that same infrastructure.

“In addition to the traditional hubs, there's also emergent places, especially because of the increase of the need,”said Perkins.“That capacity is driving more demand to the utility system, and the utility is not quick enough to catch up to provide the energy required by these datacenters, forcing datacenters to go where energy is available.”

There are also the accompanying power centers to construct, needed to supply the vast amounts of energy needed for datacenters and offering another cargo avenue for the heavy-lift industry.

“Datacenters do not want to build their own power plants; that's not their goal. They would much rather take power from the grid,”said Amy Groeschel, principal research analyst, costs and technology, at S&P Global Commodity Insights.“So, the siting of those datacenters will become very key.”

And while the growth in datacenter development poses a hurdle for manufacturers, it also offers an opportunity for project cargo vessel operators.

“The time on the water is back to the normal situation,”Perkins said.“It was chaos during the pandemic and for a short time post-pandemic, but I think that the shipment part is [back to normal] already. I think the challenge is specifically with manufacturing capability.”