NextFin News - The physical architecture of the artificial intelligence revolution is no longer a matter of silicon alone, but a massive industrial undertaking that is reshaping the American energy and real estate landscape. Inside CoreSite’s NY3 facility in Secaucus, New Jersey, the scale of this shift is measured not in square feet, but in the hundreds of megawatts required to sustain the next generation of large language models. As U.S. President Trump’s administration continues to emphasize domestic infrastructure and energy independence, the race to secure power has become the primary bottleneck for the world’s largest technology firms.
Big Tech’s commitment to this infrastructure is staggering. Amazon, Alphabet, Meta Platforms, and Microsoft are collectively on track to spend at least $608 billion this year on data center buildouts and AI hardware. This capital expenditure reflects a fundamental change in how these companies operate; they are transitioning from software-first entities into some of the world’s largest owners and operators of heavy industrial equipment. Juan Font, CEO of CoreSite—a subsidiary of American Tower—likens these facilities to digital shopping malls where multiple tenants share the same cooling and power infrastructure to avoid the prohibitive costs of building proprietary sites.
The demand for electricity has fundamentally altered the development timeline for these projects. Michael Wall, executive vice president of product delivery at Prime Data Centers, noted that just three years ago, clients typically requested 10 to 30 megawatts of capacity. Today, that requirement has scaled into the hundreds of megawatts, with some projects now seeking multiple gigawatts. This surge has exhausted existing grid capacity in many regions, forcing developers to look toward "behind-the-meter" solutions. GE Vernova has emerged as a critical beneficiary of this trend, providing fuel cell technology and natural gas solutions that allow data centers to generate their own power without relying solely on a strained public utility grid.
Inside the server racks, the dominance of Nvidia and Broadcom remains absolute. The Secaucus facility serves as a showcase for the "five-layer cake" of AI infrastructure described by Nvidia CEO Jensen Huang, where energy forms the base layer, followed by the physical chips, networking, software, and finally the AI applications themselves. Broadcom’s networking silicon provides the essential "fabric" that allows thousands of GPUs to communicate as a single cohesive unit, a requirement for training models that now involve trillions of parameters. Without this specialized networking hardware, the raw computing power of the chips would be lost to latency and data bottlenecks.
However, the sheer cost of this buildout has invited skepticism from some corners of the market. While the "Magnificent Seven" continue to pour billions into these facilities, the timeline for a tangible return on investment remains a point of contention. Critics argue that the current spending pace assumes a linear growth in AI demand that may not materialize if enterprise adoption slows. Furthermore, the rising cost of commodities continues to pressure the margins of these capital-intensive projects. For instance, spot gold (XAU/USD) is currently trading at $4,533.25 per ounce, while Brent crude oil stands at $110.42 per barrel, reflecting a broader inflationary environment that increases the cost of everything from specialized wiring to the diesel generators used for backup power.
The next phase of this expansion will likely be defined by the entry of OpenAI and Anthropic into the public markets. Both companies are spending at a furious rate to keep pace with their larger rivals and are expected to pursue initial public offerings as early as the end of this year. Their arrival as public entities will provide more transparency into the unit economics of AI training, but it will also increase the competition for the very power and cooling resources that CoreSite and Prime Data Centers are currently rushing to provide. The "arms race" is no longer just about who has the best algorithm, but who can secure the most electricity and the most efficient cooling systems in an increasingly crowded field.
Explore more exclusive insights at nextfin.ai.
