Powering AI: Could U.S. Brownfield Semiconductor Fabs Fuel the Hyperscale Boom? 

Hyperscale computing – mega data centers with tens of thousands of servers – is all about achieving massive scale in computing, typically for big data or cloud computing. As artificial intelligence (AI) continues to drive a surge in global computing demand, a critical question is emerging: how will we power it all? What if the semiconductor industry held part of the answer? Repurposing existing brownfield wafer fabs – manufacturing facilities already outfitted with substantial power infrastructure and substations – offers a compelling opportunity, but is it really feasible?

To meet the demand for AI compute – the vast hardware resources required to train and deploy AI models – that is growing at an exponential rate, the U.S. semiconductor landscape could evolve toward hybrid facilities that combine chip manufacturing, high-performance computing, and power resilience. The future of AI depends not only on the chips we build, but on where and how we power them. This article explores how brownfield wafer fabs could possibly help meet our insatiable need for power in the accelerating AI era.

The AI power crunch 
AI has quickly become the defining technological force of the 21st century. From generative language models to real-time analytics in autonomous vehicles, the applications are as vast as they are compute intensive. Yet behind the algorithms and neural networks lies an often overlooked but critical concern – electricity.

Despite their sophistication, data centers are not models of efficiency. Most operate at about 75% efficiency when it comes to energy use, with significant losses attributed to cooling, power conversion, and idle server capacity. The industry has made strides toward optimizing power usage effectiveness (PUE), but at hyperscale, even minor inefficiencies become magnified. Moreover, electricity is no longer a simple utility expense – it’s a strategic constraint. With grid congestion, rising energy costs, and environmental regulations tightening across major markets, data center operators are seeking novel strategies to reduce their carbon footprints while maintaining the performance demands of AI applications.

Source: Navitas, I.S.E.S. Power conference, April 2025, USA

The untapped potential of brownfield semiconductor fabs 

In Q4 2024, the data center segment captured an additional 4% of the total semiconductor market, lifting its share to over 38%, up from 14% a couple of years ago. In Q2 2025, the needle moved another 4% of the semiconductor market towards the data center. The data center revenue is fully dominated by the AI workloads.

Source: Ripples and Tsunamis in the Semiconductor Supply Chain, Semiconductor Business Intelligence, May 2025

One of the most underappreciated assets in this equation may lie in an entirely different sector – semiconductors. More specifically, brownfield semiconductor wafer fabs, older 200mm or even some 150mm ones – legacy facilities with mature technologies and not at the forefront of advanced chip production. These fabs still possess valuable existing, ready-to-use infrastructure, and are still in high use. These brownfield sites are often equipped with robust power grids, high-capacity substations, cooling systems, and cleanroom environments that have been meticulously maintained. Could this semiconductor manufacturing capability run in parallel with and include AI data center operations? If there is a bottleneck on AI capacity and underutilized power infrastructure sites that could be augmented or potentially an oversupply of brownfield fabs that is going to waste, we could see an argument for underutilized fabs being retrofit or power infrastructure being further built out into data centers for example.

In the late 2000s, we did see a couple of conversions of brownfield wafer fabs to alternative uses such as data centers to get the power they needed to set up their new data centers. One of the first that happened is the acquisition of the former Qimonda site located outside Richmond, Virginia by QTS in 2010 through a bankruptcy-court auction. When acquired, thought was given to running the semiconductor 300mm fab in parallel to the 200mm fab being converted to a data center. Future semiconductor use could be accommodated on the site by going vertical with multi-story cleanrooms as exist in most of Asia today. The recession hangover didn’t allow for this to come to fruition. The subsequently converted data center campus became Richmond 1, the largest and most connected data center campus in the area. The facility spanning 1.3 million sq. ft. (120,800 sqm) offered 110MW of capacity across three buildings. In July 2024, QTS was confirmed as the company that bought the 620-acre data center campus for a major expansion still in development (Richmond 2 and 3).

Source: QTS data center, Richmond, VA, USA, Google Earth, June 2025

Reimagining these sites for co-located or hybrid use opens the door to a strategic integration of manufacturing and compute. Instead of building greenfield data centers from scratch, could companies co-exist or retrofit brownfield fabs to host AI servers, taking advantage of existing power and facilities infrastructure? This would reduce CapEx, accelerate deployment timelines, and enable closer alignment between chip production and data processing.

Many advantages, but also some limitations

From a financial standpoint, the economics of reusing brownfield fabs are compelling. Land acquisition, permitting, and utility development are among the most expensive and time-consuming aspects of building new data centers. Legacy fabs bypass many of these hurdles. The presence of existing substations, water access, and transportation links could shave years off development timelines and tens of millions off budgets. Time to market is another critical factor. In the fast-moving AI landscape, being first can confer lasting competitive advantage. Companies that can stand up compute capacity quickly – without waiting for multiyear construction projects – will be better positioned to meet surging customer demand and deploy AI capabilities at scale.

One of the most compelling arguments for repurposing brownfield fabs is sustainability. Rather than spending significant time, money, and effort demolishing and decontaminating old sites and building new ones from scratch – a process that generates massive amounts of waste and carbon emissions – retrofitting existing facilities would align with the principles of circular economy and responsible asset utilization. This approach also supports the growing demand for sustainable data center development. Repurposing brownfield fabs would allow these companies to reduce their environmental impact without compromising performance.

Of course, the brownfield fab model is not without its challenges, far from it. Not all legacy semiconductor sites are created equal. Some may require significant remediation to meet modern operational or environmental standards. Zoning regulations, community relations, and local energy grid capacity must also be carefully evaluated. Fabs and data centers are often zoned differently and would require legal changes. Moreover, integrating semiconductor manufacturing and data center operations under one roof would introduce complexity in terms of facility management, cybersecurity, and operational risk. These obstacles would require careful planning and collaboration between traditionally separate business functions.

The U.S. hyperscaling landscape 

When it comes to data center infrastructure, AI compute, and cloud services, the U.S. leads the world in several key areas, but also faces a certain number of challenges.

A call to action for industry and government

To explore the potential of this model further, collaboration across sectors would be definitely needed. Semiconductor firms must be willing to explore new uses for their legacy manufacturing assets and data center operators must broaden their site selection criteria to consider industrial and manufacturing facilities. On their side, public policymakers must create incentives that reward sustainable redevelopment and technological integration. The federal government could play a catalytic role by expanding CHIPS Act funding to support hybrid facilities that combine fabrication, compute, and energy innovation. Local governments, in turn, could offer tax incentives, expedited permitting, and workforce development programs to attract these investments.

The United States has a unique opportunity to lead such a transformation. With dozens of legacy fabs scattered across such states as Texas, Arizona, Oregon, and New York, the infrastructure exists – it just needs to be reimagined. By aligning industrial policy with infrastructure investment, federal and state governments could incentivize the redevelopment of brownfield fabs as hybrid AI hubs. Programs such as the CHIPS and Science Act have already laid the groundwork by directing capital toward semiconductor manufacturing. Expanding that vision to include AI data center integration would multiply the economic and strategic benefits.

Map of U.S. brownfield fab locations

Looking ahead: Building smarter, not just bigger

The convergence of AI, semiconductors, and energy strategies could mark a turning point in how we think about digital infrastructure. As time goes by, we will no longer be able to afford to treat chip fabrication, data processing, and energy provisioning as separate silos. The future lies in integrated systems that are not only technically efficient, but also geographically strategic, economically viable, and environmentally responsible. Brownfield semiconductor fabs could be the catalyst that represents the alignment of these priorities. By leveraging what already exists, the industry can accelerate innovation without sacrificing sustainability or security. As AI reshapes every sector of the economy, the question is not whether we can meet its power demands – but how intelligently and creatively we choose to do so.

In the age of AI, the race is not just for faster chips or larger data centers – it’s for smarter infrastructure. Brownfield semiconductor facilities may offer a powerful and underutilized lever to meet the growing demands of AI compute, while also addressing the energy, cost, and sustainability challenges of tomorrow. By bridging the worlds of chip manufacturing and hyperscale computing, the United States has an opportunity to set a global standard for integrated, resilient, and future-ready digital infrastructure.