The artificial intelligence boom is putting sustained pressure on memory chip prices, as the data centre build-out required to train and run AI models drives demand for high-bandwidth memory and other specialised components. The trend reflects a broader tightening across the semiconductor supply chain, where capacity that was once absorbed by consumer devices is increasingly being redirected towards AI infrastructure.
Memory chips, including DRAM and high-bandwidth memory, are critical to AI systems because they feed data to graphics processors at high speeds. As hyperscale operators race to expand computing capacity, orders for these components have surged, leaving manufacturers struggling to keep pace. The result has been upward pressure on prices that is now feeding through to the cost of building and operating data centres.
The scale of the infrastructure investment is substantial. Worldwide, data centres are projected to require $6.7 trillion in investment through 2030 to keep pace with demand for computing power, according to McKinsey estimates. That figure underscores how central AI has become to capital spending plans across the technology sector and beyond.
Nvidia chief executive Jensen Huang has described the moment as the «largest infrastructure build-out in human history», a phrase that captures the breadth of construction now under way. The company's processors are among the most visible beneficiaries of the AI surge, but the memory chips that support them are equally essential and increasingly scarce.
The pressure on memory prices is not occurring in isolation. Data centres have become a flashpoint in the wider debate over AI, with critics raising concerns about their impact on local communities, including water use, electricity demand and noise pollution. Those concerns sit alongside the economic benefits that the build-out is generating, particularly in the labour market.
Data centre construction is creating significant demand for skilled trades. Tim Chadwick, Google Data Center Operations Area lead for Ohio and Indiana, has said there are «hundreds of thousands of skilled trade roles open across the U.S. just waiting to be filled», citing jobs such as master electrician, lead pipefitter and project manager. An estimated 2.1 million skilled trades jobs in the U.S. could go unfilled by 2030, according to Department of Education estimates cited in a JLL report.
Yet even with abundant openings, construction firms are struggling to find enough labour. A 2026 survey of construction firms working on data centre projects found that 58 per cent reported increased competition for skilled workers, while 49 per cent reported increased wage pressure. Another 37 per cent cited the availability of workers or subcontractors as their biggest challenge in delivering data centre projects.
The workforce challenge is compounded by the project-based nature of construction work. Building a data centre can create hundreds or even thousands of jobs at a single site, but once a facility is operational, the number of workers needed to maintain it drops significantly to a few dozen. Workers may have to move from one site to another, or even across the country, as projects are completed.
Some of the technology industry's most prominent executives have argued that the skilled trades offer a more resilient career path than some white-collar roles. General Motors chief executive Mary Barra has said the U.S. faces a «societal problem» in treating skilled trades as less desirable than office professions, noting that a technician could earn $80,000 to $90,000, and potentially more depending on hours worked.
Companies including Ford, Carhartt, BlackRock and Google have formed the Alliance for America's Skilled Trades, a partnership focused on building pipelines for vocational skill development. Ford chief executive Jim Farley and other leaders have argued that providing clear pathways into the trades ensures more people can transition into fields offering greater stability and higher wages.
For the semiconductor industry, the immediate challenge is meeting demand for memory without allowing costs to spiral in ways that could slow the pace of AI deployment. The price pressure now visible in memory chips is one of the clearest signals that the AI infrastructure cycle remains in an intense expansion phase, with consequences that extend from fabrication plants to construction sites.