The energy crisis puts the brakes on data center expansion

The end of the energy free-for-all

The artificial intelligence industry, which has operated under the premise of near-infinite exponential growth, has hit an inescapable physical limit: the capacity of the electrical infrastructure. PJM Interconnection, the operator that manages the power grid for 13 U.S. states and serves 67 million people, has formalized an unprecedented request to the Federal Energy Regulatory Commission (FERC). The proposal seeks the authority to prioritize the disconnection of new large-scale data centers (exceeding 50 MW) over residential homes during critical energy stress situations. This move marks the end of the era where energy was a resource taken for granted, transforming it into the most contested and strategic asset in the tech sector.

The structural imbalance: a planning crisis

The measure requested by PJM is not a random event, but a response to a severe structural mismatch. According to the operator's data, the projected demand for 2038 amounts to an additional 70 GW, driven almost exclusively by the energy voracity of generative AI and high-performance computing. In contrast, since 2022, the system has seen the retirement of 15 GW of generation capacity, mainly due to the closure of old coal and gas plants that have not been replaced quickly enough by renewable or nuclear sources. PJM's latest capacity auctions have failed to secure the supply needed to satisfy this construction frenzy, demonstrating that the speed of AI deployment has outpaced the response capacity of national energy systems.

Historically, this scenario recalls the infrastructure crisis of the late 90s, when the expansion of the Internet required data transmission capacity that the telecommunications networks of the time could not sustain. However, unlike fiber optics, which could be installed with relative agility, electrical infrastructure requires years of permitting, substation construction, and high-voltage line installations, making this 'bottleneck' much more rigid and difficult to circumvent.

Who is affected and what does the regulation mean?

The measure, called Interim Resource Adequacy Service, introduces a critical distinction: it will not affect all facilities equally. It will apply to those new data centers that do not demonstrate their own generation capacity or demand management mechanisms. In practice, tech companies planning data centers of more than 50 MW starting in June 2027 will have to act as their own energy providers. This implies integrating microgrids, investing in large-scale battery storage, or even financing the construction of renewable or modular (SMR) power plants to ensure their operability. Companies that do not meet this self-sufficiency will be first on the disconnection list if the grid enters an emergency state, an operational risk that, for critical AI services, could translate into millions of dollars in losses per minute of downtime.

Impact on the future of work and investment

This regulation profoundly alters the business model of hyperscalers (such as Amazon, Microsoft, or Google). To date, the location of a data center was decided based on latency, access to technical talent, and tax incentives offered by local governments. From now on, energy availability and the ability to integrate microgrids will be the determining factors, above any other variable. This geoeconomic shift could displace investment toward regions with energy surpluses, leaving behind areas that were previously attractive but now lack the necessary electrical robustness.

From a market perspective, this shortage will force a consolidation in the sector. Only those companies with enough capital to internalize energy infrastructure costs will be able to stay in the race for training foundational models. Smaller startups, which depend on third-party clouds, could face significant increases in computing costs as providers pass on energy investments to the final price of the service. It is a reasonable speculation, supported by current trends, that we will see an increase in 'corporate energy sovereignty,' where Big Tech will become, de facto, energy companies, buying or managing power plants directly.

The electrical infrastructure has become the invisible bottleneck of the AI revolution. The era of unlimited availability has ended, giving way to resource management based on resilience and energy autonomy.

In conclusion, the tech sector is facing a reality that demands a mandatory symbiosis between digital innovation and electrical engineering. Those who ignore the physics of the grid in favor of model deployment speed will find an insurmountable barrier. Resilience, and not just computing capacity, will be the factor that defines the leaders of the next decade.