AI’s Power Surge Shifts Leverage From Chipmakers to the Grid

Artificial intelligence has encountered a significant electricity challenge. Operating AI requires vast amounts of power; demand in the United States is rising more rapidly than the grid can accommodate, granting substantial leverage to the firms that generate and distribute electricity.

On June 2, the Electric Reliability Council of Texas (ERCOT) voted to overhaul its process for admitting large power users to the grid, addressing a backlog of data centers, cryptocurrency mining operations, and industrial facilities all competing for the same megawatts.

During the same week, lawmakers in Albany, New York, rushed to pass a one-year moratorium on new large-scale data centers, potentially making the state the first in the nation to halt such construction entirely.

Companies developing frontier AI models are encountering a barrier constructed from copper, concrete, and regulatory patience. The beneficiary of this surge in demand is the less glamorous entity at the other end of the wire: the utility, grid operator, or power producer that determines who receives electricity, when, and at what cost.

Electricity becomes the scarcest asset for AI

For most of the previous decade, discussions about AI focused on software, with the primary concern being the supply of advanced GPUs.

The focus has now shifted to industrial economics, where the limiting factors are land, generation capacity, water, high-voltage transformers, and local regulatory boards.

Goldman Sachs projects that US data center power demand will rise from 31 gigawatts in 2025 to 41 gigawatts in 2026 and 66 gigawatts in 2027. This increase will raise data centers’ share of US peak summer demand from 4.1% to 8.5% over the same period.

However, the bank noted that only about 50% to 60% of the capacity scheduled for the next one or two years is likely to be delivered on time due to delays and cancellations. Even when adjusted for these delays, the grid must absorb in two years what it typically takes a decade to add.

The International Energy Agency (IEA) projects that data center electricity use will approximately double by 2030, while demand from AI-focused facilities will triple. The IEA report highlights several bottlenecks, including tightening supply chains for gas turbines and transformers, grid connections that take years to complete, and a rush toward on-site generation that remains largely theoretical.

Power companies now possess significant leverage. A utility collects revenue regardless of which company wins the AI race; it only requires that the race continues to demand more power. Regulated utilities earn returns on approved capital spending, meaning a wave of grid upgrades translates into a wave of rate-based revenue.

Independent power producers sell into a tighter market but at higher prices. Grid operators, controlling a finite amount of connection capacity, become the gatekeepers who determine which projects are viable.

Texas illustrates how gatekeeping evolves into regulation. Under Senate Bill 6, ERCOT is implementing a “pay your own way” model that assigns interconnection costs to large customers and requires them to reduce usage during emergencies. This includes a non-refundable fee of $50,000 per megawatt and steep deposits to filter out speculative claims.

The strain is difficult to overstate, as nearly 200 large users lined up in the first months of 2026 alone, seeking a combined 438 gigawatts—more than five times the entire state’s current consumption.

New York’s proposed pause addresses the same issue from a political perspective, weighing AI data center growth against household bills, water usage, and grid reliability. Electricity has become a rationed input, and the entities managing this rationing now hold the strongest position.

Bitcoin miners anticipated this conflict, and now everyone pays

The Bitcoin market is familiar with this bottleneck because miners experienced it first. Mining built a business model around cheap, interruptible power, utilizing flexible loads that switch off when the grid is strained and absorb surplus energy when prices drop.

This is why Texas designed its new demand-response programs around this model, and why miners spent years pursuing wasted watts in windy plateaus and hydro spillways where energy was often stranded and inexpensive. Some analysts argue further that the grid should embrace this flexibility as a service, given how quickly miners can curtail their usage.

This approach is almost the exact opposite of what AI requires. Hyperscalers desire steady, always-on power and long-term certainty, supported by job creation and national competitiveness arguments that carry significant political weight. When BlackRock warned in January that AI data centers could consume up to 24% of US electricity by 2030, it effectively signaled the end of the cheap-power truce.

A CryptoSlate analysis comparing energy footprints across streaming, AI, and crypto reached a similar conclusion, noting that miners now face a tight squeeze as AI firms bid up the price of firm power supply.

The power company is now arbitrating this conflict and profiting from it regardless of the outcome.

If utilities build out generation and transmission to serve AI hyperscaler demand, ratepayers may end up absorbing part of the cost unless regulators ring-fence those expenses or compel large loads to cover their own share.

The federal forecast already leans in this direction, with the Energy Information Administration (EIA) expecting US power use to set new records in 2026 and 2027. Residential prices have already increased by 5% in 2026, with the sharpest increases occurring along the East Coast.

AI promised abstraction, with intelligence rendered as weightless, infinitely copyable software. Its expansion has made electricity the scarce commodity that determines who gets to scale, who gets priced out, and who collects a check, regardless of which company captures the majority of the market. The tech companies will continue chasing headlines, while the power company maintains a steady hand on the meter.

The post AI’s power race is shifting leverage from chipmakers like NVIDIA to the grid appeared first on CryptoSlate.