Supply chain constraints are curbing US data center development

Data centers are specialized facilities housing the digital infrastructure needed to store, process, and transmit data. The supply chain for these facilities spans critical minerals, advanced manufacturing, energy infrastructure, and skilled labor (see figure 1). Delays at any upstream stage of the supply chain can stall deployment downstream.

As demand for computing capability accelerates, particularly from AI workloads, these constraints are no longer theoretical; they now impose binding limits on how fast data center capacity can be built.

Upstream: The critical minerals bottleneck

The supply chain begins with critical minerals such as copper, aluminum, silver and rare earth elements, which are essential for servers, power systems, and cooling equipment (see Table 1). Any disruption or shortage at this stage cascades through manufacturing timelines, inflating costs, and extending equipment lead times.

Source: USGS, Center for Strategic & International Studies (CSIS), RaboResearch 2026

Manufacturing capacity strains increase shortages

Manufacturers transform critical minerals into semiconductor chips, networking hardware, power distribution equipment, and cooling systems – components that manufacturers produce in highly specialized facilities with limited spare capacity, and geographic concentration. Because many of these components share inputs and suppliers, shortages add up rather than remain isolated.

Installation and operations face labor, power, and water constraints

Once developers procure equipment, its deployment depends on skilled labor for construction, installation, and operations, as well as access to water and reliable 24/7 power. These site‑level constraints mean that supply chain challenges do not end at the factory gate but persist all the way through commissioning and operation.

Rapid growth in data center demand and the accelerated need for developing clean technologies is straining an already overwhelmed global supply chain. The IEA has tracked the critical minerals needed for clean technologies and forecasts that the total mineral demand for those technologies may grow by 28%-41% by 2040 (see figure 2). This estimate excludes demand from data centers, which could materially increase projected mineral demand growth. The surge in demand for minerals has created a significant supply chain challenge, impacting data centers with long lead times for equipment, product shortages, and higher costs. A single country – China – controls many of these materials, which adds a layer of geopolitical risk to the equation.

The looming copper shortage

Copper is a critical mineral in data center infrastructure. It is used in servers, cooling systems, networking systems, transmission cables, and power connections. However, the surge in demand for copper, paired with depleting mine reserves, raises concerns about future availability. By 2035, a BNEF forecast indicates that copper supply could fall short of demand by roughly 6 million metric tons[1] (see figure 3).

Copper is emerging as a constraint because multiple growing sectors, such as electrification and renewable energy, rely on it. Supply of the critical mineral cannot keep up with rising demand due to long development timelines. An industry expert[2] forecasts 80 new mines are needed by 2040 to keep up with demand. There are at least 200 mines[3] planned globally, but a new copper mine could take up to 17 years to reach production. The existing demand from legacy industrial use, coupled with growing demand from electrification and expansion of data centers, poses a major supply chain risk. Copper prices have already absorbed major shocks with record-high pricing of USD 14,500 per metric ton in January 2026 due to supply disruptions and surging demand. For data centers, these constraints translate into uncertainty across construction timelines and copper procurement, driving higher costs and supply risks for copper intensive equipment, which may dampen the pace of new capacity development.

[1] Data sourced from BloombergNEF’s Tracking Copper in Data Centers 2025-2035 report.

[2] Statement from Anglo American Chile’s CEO, reported by Fastmarkets in April 2025.

[3] This is according to the 2021 ICSG Directory of Copper Mines and Plants.

Déjà vu – semiconductor chip crisis

Copper, used in semiconductor chips, is not the only factor constraining chip supply. Manufacturers use semiconductor chips in all facets of modern technology, including smartphones, tablets, smart home devices, and medical devices – making demand for the product persistent and broad across multiple sectors of the economy.

The semiconductor industry experienced significant disruptions during the Covid-19 pandemic, and again during the 2022 blockage of the Suez Canal, one of the world’s busiest trade routes connecting Europe and Asia. However, constraints today are not just limited to temporary shocks but also include structural demand growth and geopolitical risks.

The acceleration of AI adoption and data center development has added a layer of demand to an already tight market. High-performance chips used in servers, like Graphics Processing Units (GPUs) and networking equipment, now compete with traditional end markets. With little margin for error, developers are stockpiling their inventory, leaning on a just-in-case inventory strategy, in anticipation of a shortage in 2026.

For data centers, constrained access to advanced chips directly limits server availability, which extends equipment lead times and delays commissioning schedules.

Constraints amplified within servers

Data center servers are comprised of multiple components, including memory chips and data storage, that are subject to supply chain disruptions. The constraints in sourcing critical materials and the availability of semiconductor chips converge when it comes to server production. As a result, server assembly and delivery are highly sensitive to upstream disruptions that may translate into longer server lead times.

Servers rely on Central Processing Units (CPUs) and GPUs to manage and accelerate computational workloads. While CPUs and GPUs have historically been used across applications, including gaming PCs, smartphones and automobiles, demand has surged with the rapid adoption of AI. Ongoing supply constraints have created lead times of up to 12 weeks for CPUs for some manufacturers and even longer for GPUs (36-52 weeks) (see figure 4).

Data storage issues further compound these challenges. Servers typically rely on solid state drives (SSD) or hard disk drives (HDD) to store data. While HDDs are the lower cost option, longer lead times and reportedly sold-out supply through 2027 are forcing developers to rely more on SSDs despite higher costs. Developers may choose to trade cost efficiency for speed, as component availability starts dictating deployment timelines.

Network infrastructure labelled as a commissioning bottleneck

Network infrastructure is a component within server racks and core to connectivity and security for data processing. The core hardware for this infrastructure – switches, routers, and printed circuit boards – have emerged as another downstream bottleneck in the supply chain. As demand outpaces production capacity, industry reports indicate lead times of up to 52 weeks for networking equipment, while printed circuit board lead times have doubled from 8-12 weeks to 20-30 weeks.

For data centers, even if servers are delivered on site, facilities cannot be fully commissioned or brought online without the networking layer in place, impacting deployment times.

Water consumption and availability are a growing challenge

Data center water consumption varies by type, scale, cooling technology (air vs liquid) (see Table 2), and location. Water plays a critical role in the operational efficiency of data centers by removing the excess heat created by the IT equipment. As workloads become more computationally intensive, especially with AI, cooling requirements and thus water use rise accordingly. The growing importance of water availability for data centers is now a key factor for data center site selection.

Data center operators track water efficiency using the Water Usage Effectiveness (WUE) metric. While useful for benchmarking efficiency, the WUE does not capture the gravity of water consumption. A 100 megawatt (MW) and 10 MW facility could report the same WUE (ex. 0.5), but the larger facility may consume up to 10 times more water (see figure 5). Additionally, the WUE may only track direct, onsite water, leaving out approximately 75% of water consumption from indirect sources like electricity generation and upstream supply chains. Although efficiency improvements matter, the size of facilities remains critical in overall water consumption.

A report by Ceres finds that water use associated with US data center cooling operations could increase by 870%, from 385m gallons per year to more than 3.7bn gallons per year, which is enough to supply a city the size of Flagstaff, Arizona for almost two years. The immense water usage is one factor fueling the record pushback from communities on data center buildout. In 2025, communities blocked or delayed over USD 150bn in data center development because of concerns like the impact on local water. Some cities have already set moratoriums on data center development because of concerns around issues such as the impact of water usage.

Operators are exploring ways to minimize freshwater withdrawals and avoid local depletion. Microsoft is already looking at new liquid cooling technologies that utilize recycled water. Amazon has moved to use recycled water in more than 120 of its data centers across the US, with expectations to save over 530m gallons of freshwater annually.

The human element – labor shortages

Skilled labor is critical to the data center value chain, from site development through to operations. Construction spending for data centers reached USD 3.57bn in December 2025, surpassing spending for offices for the first time – illustrating the rapid growth in the sector. As development progresses, access to skilled labor emerges as another binding constraint.

Today’s labor shortage is demographically driven. A wave of retirements is reducing the available workforce for electricians, HVAC technicians, and facility operators at a pace currently unmatched with new entrants. Industry estimates suggest the construction industry would need to recruit roughly 500,000 additional workers in 2026 to keep up with demand and that amount could rise to 1.4m by 2030. One projection estimates that for every five workers who retire, only one new worker enters the trade workforce (see figure 6).

Specialized skill requirements compound the issues in data center construction and operations. Managing expensive and fragile data center equipment leaves little margin for error. Even when replacements are available, correcting mistakes can introduce delays, particularly when critical components are subject to long backorders. Labor shortages have tightened the market, pushing wages for skilled trade workers 25%-30% above norms and increasing project and execution risk.

The elephant in the room – power procurement

The biggest challenge in the data center industry right now is access to reliable power. Data centers require continuous electricity from the grid or from behind-the-meter assets, such as wind, solar, battery storage, geothermal, nuclear, or natural gas turbines. However, securing reliable power for data centers has become difficult.

Transmission bottlenecks and interconnection queues already constrain grid infrastructure across the US, delaying new large-load connections. Developers are seeking to bypass grid limitations with behind-the-meter assets but are facing shortages in critical generation equipment. Lead times for natural gas turbines, one of the few options capable of meeting data centers’ 99.99% uptime requirements, now extend up to seven years, creating a bottleneck in the development timeline.

For data centers, access to affordable and reliable power now determines which projects advance and which stall. Developers with sufficient capital to secure on-site generation or priority grid access can move ahead, effectively introducing a pay-to-play model for new capacity deployment. RaboResearch covers power for data centers in more detail in a separate piece.

Geopolitical disruptions magnify existing supply chain bottlenecks. The data center supply chain is global, leaving it heavily exposed to geopolitical risks such as trade policy shifts, tariffs, and armed conflict. These risks can translate into higher costs, longer lead times, and greater uncertainty for developers.

Tariffs reel in higher costs and uncertainty

Data centers are already among the most capital-intensive infrastructure assets. In March 2026, Rystad Energy estimated total global data center capex at roughly USD 770bn, surpassing upstream oil and gas investment. While hyperscalers like Google and Meta with deep balance sheets can often absorb incremental cost increases, tariff-driven price inflation disproportionately impacts smaller developers, affecting their ability to compete.

Copper

In the first half of 2025, the US imported 85% of its refined copper from four major markets: Chile, Congo, Canada, and Peru, totaling over 730,000 metric tons (see figure 7). Refined copper is primarily used for electrical applications in the US, including for data centers. As a result, the global nature of the US copper supply chain could expose developers to tariff risks.

On July 30, 2025 the US administration announced a 50% tariff on all imported, semi-finished copper products, effective August 1, 2025. The tariffs cover products like wires, tubes, and plates, which are used in the power distribution, networking infrastructure, and cooling systems for data centers. The 50% tariff did not apply to refined copper, but there is a proposed phased-in tariff (see figure 8). The tariffs push developers to strategically plan inventories. In the near term, a copper shortage exists, but over time it can resolve through increasing domestic production.

Steel

The Trump administration also applied tariffs to steel, essential for load-bearing structures, server racks, cooling systems, and structural frames within data centers. In February 2025, the administration imposed a 25% tariff on all imports of steel and aluminum, which were then raised to 50% a few months later. Further guidance issued in April 2026, extended tariffs to full customs value.

While only 25% of the steel used in the US is imported, primarily from Canada, Mexico, and Brazil, tariff uncertainty still introduces volatility into sourcing decisions. Developers can pivot toward domestic suppliers but often at a higher cost.

Recent war implications

Geopolitical conflict adds another layer of risk to the data center supply chain by introducing sudden disruptions to highly specialized supply chains. The US-Iran war has heightened concerns around helium availability – a critical component for semiconductor chip manufacturing.

Following disruptions related to the war, QatarEnergy’s Ras Laffan LNG facility, a major helium production site, sustained damages which could impact more than a tenth of global helium supply in 2026. The impact is uneven across semiconductor manufacturing hubs. Outside of China, the key centers for semiconductor fabrication are in South Korea and Taiwan. Together, these three centers control about 36% of the global chip fabrication capacity (see figure 9). South Korea sourced about 65% of its helium from Qatar in 2025, leaving chipmakers in that market potentially vulnerable to a supply crunch. Taiwan is a bit more diversified, sourcing roughly 30% of its helium from Qatar.

While the diversification limits immediate systemic risk, exposure remains concentrated at critical nodes. The Taiwan Semiconductor Manufacturing Company (TSMC) produces about 90% of the world’s most advanced logic chips, many of which rely on helium-intensive processes, and supplies key US-based companies including Apple, Nvidia, Google, Amazon, and Qualcomm. Although TSMC’s diversified helium sourcing reduces the likelihood of an abrupt production halt, any sustained disruption to helium supply could tighten capacity.

The rapid expansion of data centers is colliding with supply chains that cannot scale at the pace of AI and data center demand. Constraints across power, water, critical materials, skilled labor, and geopolitically-exposed inputs compound, limiting how quickly new capacity can be delivered.

As a result, data center buildout in the US is likely to slow to some extent and become more selective over the next couple of years, with timing and site viability increasingly determined by access to physical resources rather than demand alone. Developers with the balance sheet flexibility to secure power, equipment, and materials upfront are more likely to advance projects within the desired time, while smaller developers could face longer delays or stalled developments. These dynamics place the greatest pressure on the more than 900 US data centers currently in the planning phase (see figure 10) some of which may face prolonged timelines or cancellations.