The Network That Can’t Keep Pace
On June 2, 2026, a data center in Middenmeer, Netherlands, began operating with a power capacity of 350 megawatts, straining an already stressed electrical grid. This is not an isolated incident: in Northern Virginia, the demand for grid connection has reached a level where it is impossible to provide access for new projects within 14 years. The congestion is not a problem of a lack of energy, but of distribution. According to a report from June 4, 2026, the European Union launched the AI.grids initiative, involving 48 partners, including grid operators and research institutions, to develop sovereign AI models for managing energy flows. The crucial data point is that in the USA, the queue of projects awaiting interconnection exceeds 2,600 gigawatts, more than double the country’s current operational capacity. This means that the demand for power is not only growing, but is outpacing the physical system that must supply it. Consequently, the expansion of the digital world is no longer limited by the availability of electricity, but by the capacity of the grids to transmit it.
The crisis is not technical, but operational. Data centers, such as those of Nvidia, require a completely new energy architecture, not only more powerful, but also more intelligent. The collapse is not imminent, but is already underway: the waiting time for a project increases from 24 to 72 months, a delay that directly impacts the competitiveness of companies. In fact, the growth of AI is no longer a matter of power, but of access. The event is not a blackout, but a flow blockage. The system does not shut down, but becomes blocked. The problem is not generation, but connection. This implies that optimization can no longer be an afterthought, but must be integrated into the core of the project.
The interconnection node as a critical element
The electrical interconnection node is the point of maximum tension between demand and capacity. In the PJM region, which serves 65 million people, the energy demand for data centers exceeds the available network capacity by 6.6 gigawatts. This deficit is not caused by a drop in production, but by an overlap of requests in a fixed system. The problem is not a lack of generation, but a lack of transmission infrastructure. In Texas, ERCOT recorded a queue of 410 gigawatts, a value that is not only high, but growing. Companies like Enphase Energy and NextEra Energy are investing in storage solutions, but these systems do not solve the problem of congestion, which is a flow constraint, not a volume constraint.
The operating mechanism is simple: to connect a data center to the grid, an agreement with the network operator is required, which assesses the residual capacity. If the system is already at its limit, the project is put in a queue. The waiting time is not fixed: it can vary from 24 to 72 months, depending on the location and complexity. In Virginia, the delay is 14 years, a time that cannot be addressed with traditional models. The repair time for a fault in a transmission line is a few days, but the waiting time for a new connection is years. This implies that the system is not only under pressure, but is chronically behind. The bottleneck is not physical, but temporal. The infrastructure exists, but it is not usable in a timely manner.
Who Pays and Who Profits in the Power Grid Congestion?
The costs are not distributed equally. Companies operating in areas with congested networks, such as Northern Virginia, see their projects delayed by years, with a direct impact on return on investment. A 24-72 month delay adds an additional cost of over $500 million for a single data center, according to estimates from Enel Green Power. Conversely, companies that already have access to the grid, such as Microsoft and Amazon, are increasing their profit margins, as they can offer services at fixed prices for long periods. The advantage is no longer in power, but in access. Energy management companies, such as S&C Electric and ABB, are seeing an increase in sales of demand-side control systems, with a 40% increase in the first quarter of 2026.
The consequences extend beyond the digital sector. Electricity is a primary input for industrial production. In Pennsylvania, semiconductor factories have reduced production by 15% due to the lack of dedicated power for data centers. The cost of transporting natural gas has increased by 22% in Ohio, as regasification projects have been postponed to make way for data centers. In Europe, the Union has imposed a 10-12.5% tax on products from countries with forced labor practices, a measure that directly affects suppliers of electronic components. This is not just an increase in costs, but a paradigm shift: the demand for energy is no longer a factor of production, but a factor of access.
Shutdown: The System That Stops Pretending
The euphoria assumed that the growth of AI was limited only by the availability of chips and power. Data shows that the limit is the flow. The system hasn’t shut down, but it’s starting to show its limitations. Efficiency is no longer a goal, but a condition for survival. The solution is not to increase generation, but to optimize the flow. The next indicator to monitor is the average waiting time for a connection to the grid in North America: if it exceeds 48 months, the system is in collapse. The other indicator is the number of projects in the queue for interconnection: if it exceeds 3,000 GW, the system is in structural crisis. The real challenge is not to produce energy, but to deliver it. AI must not only consume, but also manage. The paradox is that to save the grid, we must use AI itself.
Photo by İsmail Enes Ayhan on Unsplash
⎈ Content generated and validated autonomously by multi-agent AI architectures.
> SYSTEM_VERIFICATION Layer
Verify data, sources, and implications through replicable queries.