The power consumption of a single data center is equivalent to that of a major metropolis. While we're still discussing how many cameras we've upgraded on our phones or how many kilometers our car batteries can cover, another true industrial revolution is quietly underway in the American Midwest. I deeply feel that by 2026, the global technology race will have completely changed. Previously, we competed on who could manufacture chips with smaller processes, then on who could generate larger model parameters, and now, we compete on who can build gigawatt (GW)-level "AI factories" on the land. Recently, ABI Research and SemiAnalysis updated their 2026 global data center rankings. After reviewing this list, my most immediate impression is that the traditional manufacturing era, which judged success by "output per acre," is being replaced by an era where computing power is measured by "power consumption." Below are ten absolute behemoths in this industrial feast. A violent upgrade from "megawatts" to "gigawatts" Let me first explain some background. Two years ago, a data center consuming 200 megawatts of power was already considered a "monster." Back then, 1 million GPUs were simply a pipe dream. However, by 2026, all of this had changed. The unit changed from megawatts to gigawatts (1 gigawatt = 1000 megawatts). What does this mean? A typical coal-fired power plant consumes only a few hundred megawatts. This means that the electricity consumption of these data centers is comparable to or even exceeds that of a small city. First Place: Project Rainier – A True "Chip Rain" Location: New Carlisle, Indiana, USA Power: 2200 megawatts (under construction) This is Amazon's trump card, a collaboration with AI star company Anthropic. It's not only the world's largest AI cluster, but also a high-stakes gamble for Amazon's self-developed chips. The park will deploy 500,000 Trainium 2 chips, with plans to expand to 1 million by the end of the year. In advanced manufacturing, we often talk about mastering "core processes." Amazon is telling Nvidia: the "bread" I make myself can also feed AI models. The significance of this project lies in the fact that it marks the beginning of the cloud computing giant's complete break from dependence on a single supplier and the start of building its own "arsenal." Second and Third: The Fierce Battle Between Microsoft and Meta. Microsoft's Fairwater campus (Wisconsin/Georgia) secured 2,000 megawatts of capacity. Microsoft is connecting these interstate campuses into a giant computer using a dedicated "AI wide area network." Meta goes even further, establishing facilities in Altuna (1,401 megawatts) and Prynville (1,289 megawatts). The established Prynville center, in particular, achieved a PUE (Power Usage Effectiveness) of 1.06. This is an incredibly impressive figure, meaning that almost all the electricity consumed is used for computing power, with no waste on cooling. This serves as a reminder to manufacturing companies: as your equipment power density increases, cooling and energy management are no longer secondary but core competencies. Meta leverages Oregon's cool climate and low-cost hydroelectric power. Campus "Hidden Champions": Switch and Vantage. Besides the self-built projects of internet giants, two professional "contractors" are also worth noting on this list. Switch's Tahoe Reno (Reno, Nevada) not only boasts a staggering 8.09 million square meters (equivalent to 1,130 standard football fields), but its roof is even designed to withstand winds of up to 200 mph. This "fortress-like" design philosophy stems from dual concerns about extreme weather and physical attacks. Vantage's Ashburn campus (Virginia), while ranking tenth (590 megawatts), is strategically located. Ashburn is known as the "Internet Capital of the World" because over 70% of global internet traffic passes through it. Its WUE (water usage efficiency) is close to zero, indicating that it uses almost no water for cooling. Besides the US, what else? While the top ten are dominated by American companies, two places are true "unofficial giants." One is the Lakeside Technology Center in Chicago. Although its total area is only 111,500 square meters, it doesn't rely on size but on "connectivity." It houses over 40 telecommunications service providers, and the core trading system of the Chicago Mercantile Exchange is located here. For the financial manufacturing industry, this is the New York Stock Exchange of the digital world; every microsecond of delay means a loss of real money. Another example is India's Tulip Data Center, located in Bangalore-"India's Silicon Valley." Spanning 84,000 square meters, equivalent to 12 Taj Mahals side-by-side, its most striking feature is its single-rack power density, averaging 9 kilowatts. This is high for traditional data centers, but in the AI era, this figure is being rapidly surpassed. What does this mean for our manufacturing industry? Looking at these behemoths, we cannot simply be spectators. As a researcher in manufacturing, I see at least three clear signals: First, electricity equals computing power, and computing power equals national power. Previously, we measured a country's industrial strength by steel production and electricity generation. In the future, we will look at how many gigawatt-level AI computing power clusters a country possesses. This will drive not only chips but also the explosive growth of a series of advanced manufacturing supply chains, including high-voltage direct current (HVDC), gas turbines, liquid cooling pipelines, and new building structures. Second, the definition of a factory is being revolutionized. Nvidia CEO Jensen Huang has recently been promoting the concept of an "AI Factory." Traditional factory production lines churn out cars and mobile phones. But the "production lines" of these data centers churn out human knowledge, logic, and automated decision-making. Third, extreme energy efficiency is forcing a materials revolution. When PUE approaches 1.0 and single-rack power exceeds 100 kilowatts, traditional fans and air conditioners are obsolete. This compels us to develop new thermally conductive materials, new chip packaging processes, and new liquid cooling and even immersion cooling technologies. This places extremely high demands on materials science and precision manufacturing. In conclusion, looking back from the spring of 2026, humanity is indeed building some incredible things. We are compressing the knowledge accumulated throughout industrial civilization into these enormous boxes covering several square kilometers. They are like behemoths that devour electricity, yet spew out the wisdom that drives this era forward.
