Data centers locked down: New York pauses AI over power costs

The honeymoon between big tech and American states appears to be officially over. While Silicon Valley paints us a future where AI solves all of humanity's problems, reality is knocking harshly at the door in the form of overheated transformers and exorbitant utility bills. New York has become the latest state where politicians are seriously discussing a moratorium on building new data centers. This is not simply a bureaucratic hiccup, but a symptom of a deep systemic crisis that has been brewing for several years.

The problem is that modern artificial intelligence is an incredibly resource-hungry entity. If data centers were once simply large warehouses with servers, the era of generative models has transformed them into industrial facilities with colossal energy consumption. A single rack with the latest Nvidia chips consumes as much electricity as several dozen residential homes. When there are thousands of such racks, the local power grid literally groans under the load. In New York, this was felt especially acutely when plans to expand computing capacity collided with the need to maintain stable voltage in the homes of ordinary citizens.

Lawmakers in Albany are now carefully looking at the numbers and understand that promised jobs in the IT sector do not offset the risks to infrastructure. Let's be honest: a data center is not a factory—it doesn't require thousands of workers. After construction is complete, the facility is maintained by a few dozen engineers and security personnel. Meanwhile, the burden on the environment and the state budget remains colossal. Residents are beginning to ask uncomfortable questions about why they should pay more for electricity just because corporations decided to train a new version of a chatbot in their area.

New York is far from alone in its concerns. Look at Virginia, which for a long time has been considered the "capital of the internet" due to its massive concentration of servers. There, the level of resistance from local communities has reached such a peak that new projects stall for months due to lawsuits and protests. The situation is similar in Georgia and South Carolina. Regardless of party affiliation, politicians have realized that uncontrolled growth of "clouds" could lead to rolling blackouts in residential neighborhoods. This creates a dangerous precedent for the entire AI industry, which has grown accustomed to treating computing power as a practically infinite and cheap resource.

Technology giants are trying to maneuver and find ways out of the energy deadlock. We're seeing them suddenly become the chief advocates of nuclear energy. Microsoft is signing contracts to revive reactors, Google is investing in small modular systems. But this is a perspective of an entire decade, while servers need to be installed and launched today. If states one after another begin to impose moratoriums on construction, we will see a massive migration of computing capacity to countries with a more liberal—or simply desperate—approach to environmental protection and energy.

Adding to this problem is the question of water resources. Data centers not only consume electricity; they literally drain local water bodies to cool their systems. In the context of climate change, when New York is trying to follow strict environmental regulations, the emergence of a new giant water consumer looks like political suicide. We are witnessing a classic conflict between digital progress and the preservation of basic resources, where on one side of the scale is the power of new neural networks, and on the other is the stability of life for millions of people.

The irony of the situation is that AI itself is often advertised as the perfect tool for optimizing power grids and combating the climate crisis. But so far, it has played the role of the main destabilizing factor, one that demands ever more resources. The industry has encountered a physical limit to growth that cannot be overcome by simply updating software code. We either need to radically change chip architecture, making them dozens of times more efficient, or prepare for the cost of "intelligent" computing to rise in line with megawatt-hour prices.

The bottom line: The era of cheap and accessible cloud computing power is coming to an end, running up against the capabilities of aging electrical grids and the patience of taxpayers. Can the AI industry continue to grow if it has to build its own power plants?