Data centers are seeing unprecedented growth to meet heavy forecasted demand, driven by the AI revolution. Bubbling under the surface of these deals and announcements is a potential disaster for the industry, as supply chains struggle under the weight of enormous pressure.

To meet their goals and obligations, companies investing in AI will need rapid data center build-outs to ensure there is the required compute capacity to meet demand and make a return on their investments.

Global demand for data center capacity is expected to almost triple by 2030, but worldwide resource shortages are conspiring to delay data center projects across the globe. Power infrastructure construction, equipment manufacturing, and pure-play construction materials are all facing shortages and other supply issues.

One reason is that heavy demand has far outpaced existing supply streams, but several compounding factors are adding their weight to the issue. The Covid-19 pandemic disrupted global supply chains due to lockdowns and travel restrictions, followed by workloads shifting digitally and cloud and remote work becoming more commonplace. This was immediately bookended by the meteoric rise of AI workloads.

While the shortages seen in the pandemic-era have somewhat lessened, the US is still suffering from massive supply issues, with tariffs imposed by US President Donald Trump affecting key material and component imports, resulting in construction stalling and significant amounts of capital left tied up in these projects. Though Trump’s so-called “Liberation Day” tariffs were ruled illegal by the country’s Supreme Court, they were reimposed under a different piece of legislation almost immediately.

Even if a data center makes it to its final form in this context, it may still face further hurdles in the form of insufficient power grid infrastructure, which is in need of its own buildout and upgrades. Of course, this requires components and materials that are also in shortage. Without these infrastructure upgrades, many data centers will struggle to be able to supply enough electricity to meet the burgeoning AI demand, forcing them to turn to fossil fuel-powered on-site generation, another expensive avenue.

AI’s growth momentum has caused wounds that were opened during the pandemic to fester, compounded by the Trump administration’s tariff-based strategy to bring manufacturing back to the US. But many US manufacturers are not benefiting from increased demand and shorter imports – instead, they face a rocky road through significant financial distress to meet these ever-growing AI expectations.

Risk versus reward

Data centers and their infrastructure have become the new priority for US builders and manufacturers. And, according to Deloitte, while overall construction spending has declined in 2025, investment in structures is set to grow by 2026, driven by AI-related data center spending. This correlation is depicted across industries, as AI props up worldwide economic growth.

For every announcement made around new compute capacity or a new data center campus, entire industries of manufacturers, importers, distributors, and laborers are invoked to make it happen.

But, under the hood, the companies supporting this growth are struggling.

Charlie Minutella, CEO of RapidRatings, says that these manufacturers and builders are “not in a position to support the increased expectations of them to build these data centers.”

RapidRatings, which provides financial health analysis on supply chains, paints a stark picture of the industry. According to Minutella, 20 percent of companies that support data center construction are already at a “high risk of bankruptcy.”

The RapidRatings’ data shows that both public and private companies across key sectors for data center construction are facing these issues, with between 20 and 30 percent of companies supporting non-residential building construction, utility system construction, semiconductor and other electronic equipment manufacturing, and electric power generation, transmission, and distribution in significant financial distress.

“Our analysis covers not only the tech sector, but also automotive, complex and diversified industrials, and manufacturing,” Minutella says, “we cover the whole span from the very top, all the way to the component part manufacturers.”

“We have a comprehensive view of the private companies that make up this ecosystem. What we're seeing is based on the actual financials of these companies – and our model is highly predictive in determining whether or not a company will go bankrupt.”

Minutella says that these companies wouldn’t be in better shape even in a scenario where they had 50 percent revenue growth.

He argues that those at the top of the chain have not thought through the flow of capital. They’ve set aggressive goals and agreed impressive deals with infrastructure providers, but when these providers experience delays themselves, the domino effect is significant.

Citing a high-profile example, Minutella notes: “CoreWeave has had delays in its data center construction.”

The AI cloud computing company, which is backed by Nvidia, scaled back its annual revenue forecast late last year due to delays at a third-party partner. CoreWeave has announced significant investment in AI-data center capacity in the past two years and has raised significant funding in both equity financing and debt capital commitments.

Elsewhere, OpenAI’s decision not to take additional capacity at the Stargate Abilene Campus is likely to have had a knock-on effect for companies working on additional buildings at the campus, though Microsoft looks set to scoop up the excess planned data center space.

Much of the industry’s modern growth is built on the speculative revenue envisioned by an AI revolution, and Minutella says it’s unlikely that the issues faced by the construction sector will get the attention they deserve until there’s a “real set of defaults,” despite the massive impact they could have on realizing this promise.

“There's definitely kinks in the supply chain. When the data centers get built, the full switch-on isn’t happening,” he argues.

“While private credit has come in and offered financing to these companies, a lot of those financing terms are based on milestones and covenants. So while some of these companies are getting capital to get off the ground, everything needs to line up. Otherwise, there could be instances where they might lose the company, lose a significant portion of the company, or the funding would dry up.”

Roll out and rise up

A key component of data center construction is transformers, which are used in data centers to convert high-voltage electrical power from the grid to lower voltages.

Traditionally, those in need of transformers have been reliant on imports – some 80 percent of transformers have been historically imported into the US, according to Wood Mackenzie. The surge in demand due to the data center buildout, as well as impacts from tariffs, has resulted in slower lead times in acquiring transformers. However, this is only half of the issue.

Of those transformers currently in circulation, many are reaching the end of their working lives. In 2024, the US Department of Energy reported that 55 percent of in-service distribution transformers are older than 33 years, exceeding their service life. Replacing these will compound the already heightened demand and put pressure on upstream suppliers.

The North American Electric Reliability Corporation has said lead times for transformers hit more than two years in 2024, with larger transformers taking up to four years. A supply deficiency of this size, given the age of transformers in operation, is risking more than just data center construction efficiency. The scale of the production capacity requirements leaves it unclear when manufacturing will be able to meet demand.

One upstream supplier Minutella references is Cleveland-Cliffs, the sole US producer of grain-oriented electrical steel (GOES), which is purported to be on the “brink of disaster.”

GOES is a key component of transformers, a specialized alloy with specific magnetic properties that allows for reduced energy losses and improved overall efficiency.

As the only current US manufacturer of this component and other steel products, Cleveland-Cliffs should be benefiting from US steel tariffs, but instead it faces challenges from near-term headwinds and market uncertainty. The manufacturer has seen its share price plummet since the turn of the year, due, in part, to declining revenue.

Cleveland-Cliffs makes many other forms of steel, and GOES is just one small element of its production. According to researchers at The National Interest, the company’s factories do not produce enough volume to fully meet the demand from domestic transformer manufacturers. Cleveland-Cliffs’ GOES is also limited by size and weight, with the company being unable to meet the requirements of larger transformers with designs requiring greater GOES widths.

The volatility of the US economy under tariffs and shortages of the input materials that go into making GOES themselves, as well as a global oversupply of steel, and uncertainty in the profitability of the transformer market, mean that Cleveland-Cliffs is unlikely to expand its production capacity.

Cleveland-Cliffs has, however, seen interest from foreign investment, such as POSCO, which is reportedly taking a 10 percent stake in the steel-maker with the stated aim of localizing its production in the US and dodging steel tariffs.

There are also other US manufacturers investing in GOES production. U.S. Steel, owned by Nippon Steel, is spending billions of dollars to begin mass production of GOES.

Falling short

If the US is unable to scale its construction pipeline, it risks falling behind in the race for AI dominance. Minutella says that hyperscalers involved in AI-data center projects need to take a “proactive view” on how they plan to finance the supply chain.

So far, he explains that it’s been “more about let's make commitments and then let's let those companies figure it out, but they may not have the balance sheet to be able to do that.”

He adds: “The traditional funders, the banks, have stayed pretty far away from this. They're not going to fund the upstream suppliers. The hyperscalers, the ones with the strongest balance sheets, need to use their credit rating to be able to get funding at rates that are reasonable for these companies.”

“I think short of that, there's going to be a lot of missed expectations and finger-pointing.”

Without systemic changes, “whether that’s government-backed lending, or a significant dip in interest rates,” Minutella expects that there could be a significant impact on the expected data center build-out over the next few years.

“Where they have a lot of debt on their balance sheets, if their interest payments go down and cash availability improves, that will help a bit,” he says, “but I think if all remains the same, if tariffs exist, if high interest rates exist, if lack of government intervention subsidies persist, then it’s going to be one of those situations where the flow of capital has a big impact on the success rate of these data center build outs.”