An AI-driven infrastructure race is sweeping across the United States. According to a Barclays research report dated October 31, the total capacity of currently planned large-scale data center projects in the US exceeds 45 gigawatts (GW), with this construction frenzy expected to attract over $2.5 trillion in investment.
The report highlights that the primary drivers of this expansion are OpenAI's Stargate project, hyperscalers like Amazon, Meta, and Microsoft, as well as Elon Musk's xAI. These companies are rapidly planning and building computing clusters at an unprecedented pace to train and operate increasingly complex AI models.
This is not just a computing arms race among tech giants but also poses unprecedented challenges to the US power infrastructure. Surging electricity demand is colliding with the "power wall" of the existing grid. Insufficient grid capacity, permitting delays, and supply constraints are forcing these tech companies to adopt "Bring-Your-Own-Power" strategies.
**Key Players: Stargate, Hyperscalers, and xAI Lead the Charge** Barclays' tracking reveals that a handful of tech giants are at the core of this 45GW construction boom:
- **OpenAI's Stargate Project**: Targeting 10GW capacity and $500 billion in investment by late 2025, with 7GW already committed across Texas, Wisconsin, and other locations. Partners include Oracle, SoftBank, and data center developers Vantage and Crusoe.
- **Meta**: Advancing multiple "Titan Clusters," including the 1GW Prometheus project in Ohio and plans to expand Louisiana's Hyperion project to 5GW.
- **Amazon**: Added 3.8GW globally in the past 12 months, with capacity expected to double again by 2027. Barclays estimates Amazon could add ~13GW in the US alone during 2026-2027.
- **Microsoft**: Building a 900MW AI factory in Wisconsin, with similar projects planned elsewhere in the US.
- **xAI**: Expanding its Memphis, Tennessee, data center to 1.4GW to train its Grok model.
**Sky-High Costs and Capital Intensity** The report notes that data center construction costs (excluding IT equipment) now exceed $17 million per megawatt. For instance, OpenAI's 7GW Stargate project represents over $400 billion in committed investment—translating to ~$57 million per MW (including IT equipment)—underscoring the massive capital density of AI infrastructure.
**Grid Bottlenecks Spark "Bring-Your-Own-Power" Trend** Grid limitations pose the most critical challenge. Even with approved grid connections, project developers are increasingly building on-site generation to accelerate timelines and ensure reliability. For example, Stargate 1—despite securing 1.2GW of grid access—plans to deploy ~350MW of on-site natural gas capacity to speed up activation and replace diesel backup.
To handle AI workloads' millisecond-level power fluctuations, hybrid solutions are emerging. Meta's Prometheus project combines gas turbines, engines, and diesel generators for baseload, load-following, and emergency starts—a model becoming industry-standard.
**Funding the Trillion-Dollar Boom** Behind the massive investments lie complex financing structures and rising costs. Beyond tech giants' own capex, private equity (e.g., Blue Owl Capital's $15B JV with Crusoe for Stargate 1) and infrastructure funds play key roles. The "Energy-as-a-Service" (EaaS) model is also gaining traction, with firms like Williams investing billions to build dedicated power plants under long-term agreements—including $2B for Meta's Prometheus and $3.1B for another unnamed client.
**Supply Chain Strains** Exploding demand is straining power equipment supply chains. Heavy gas turbine prices have surged 50% in under two years, with lead times ballooning. Manufacturers like GE Vernova and Caterpillar face parts and labor shortages despite capacity expansions. Some firms are bypassing queues by acquiring used or "new-in-box" equipment—e.g., Fermi America secured unused Siemens turbines from an LNG project.