AI giants are learning the language of power grids — the hard way. The companies building the world’s most advanced artificial intelligence models have suddenly realized that their massive data centers cannot plug into the U.S. electric grid on demand. This realization has forced a crash course in federal energy regulation, a world dominated by utilities, state commissions, and a powerful agency known as the Federal Energy Regulatory Commission.
The relationship between Silicon Valley and Washington, D.C., has always been complex. But the relationship between Big Tech and the nation’s energy regulators has proven to be a far more difficult language to master. The stakes are enormous, the timelines are tight, and the vocabulary is entirely foreign to most software engineers and executives.
Why Are AI Companies Struggling to Connect to the Grid?
The core problem is not a lack of willing power plants or available renewable energy. The bottleneck is the process of physically connecting a new, enormous load — a data center consuming as much electricity as a small city — to the high-voltage transmission network that spans the country. This process is governed by a thicket of rules, interconnection queues, and cost-allocation studies managed by regional grid operators and approved by FERC.
Laura Swett, chairman of the Federal Energy Regulatory Commission, delivered a blunt assessment at a major industry conference earlier this spring. She stated that the giants of artificial intelligence were not coming to see her. When they did show up, they simply did not understand the system they were trying to navigate.
“The hyperscalers, when they do come speak to us, they don’t speak FERC,” Swett said. Her remarks highlighted a fundamental disconnect. The companies that dominate cloud computing and AI development had spent years mastering federal policy on issues like data privacy, antitrust, and content moderation. But the electric grid operates under a different set of rules, one that has evolved over a century and involves a complex dance between federal authority and state jurisdiction.
Swett’s comments set off a two-month sprint for AI developers to engage FERC. The urgency stems from a simple reality: frontier AI models require staggering amounts of computing power. Microsoft, Google, Amazon, Meta, OpenAI and Anthropic are each spending hundreds of billions of dollars on AI infrastructure. Without a faster connection process, those investments face years of delays.
Hyperscalers’ complaints about utilities, according to Swett, show a lack of understanding of how the grid functions. The grid is not a simple extension cord. It is a finely balanced machine where adding a large load can destabilize voltage, overload transformers, and require new transmission lines that take a decade to permit and build. Tech companies accustomed to rapid software deployment are finding that electrons do not move at the speed of code.
What Is the Most Controversial Option FERC Is Considering?
FERC approaches a June release for a proposal to bring data centers onto regional electric grids faster. The commission is weighing several paths forward, but one option stands out as the most politically fraught. That option is a federal takeover of the connection process.
Under this scenario, FERC would assert broad federal control over how quickly to link data centers to the grid. Federal officials would determine how to allocate costs for transmission upgrades and utility service. This approach bypasses state regulators who traditionally oversee electricity sales and set the rates that households and businesses pay.
A federal takeover of the connection process is opposed by state officials. They argue that local utilities understand their own grids better than Washington bureaucrats. They also worry about cost shifting — if data center connections are fast-tracked, the expenses for new transmission lines could fall on ordinary ratepayers rather than on the tech companies themselves. The tension between national speed goals and local control is the central political battle in this debate.
Energy Secretary Chris Wright asked FERC in October to draft a rule asserting federal jurisdiction over grid connections for data centers. This request from the Department of Energy adds significant weight to the federal takeover argument. It also signals that the White House sees data center power demand as a matter of national economic competitiveness, not just a utility issue.
How Did Google Prepare for This Challenge?
Among the tech giants, one company moved early and strategically. Google got a jump on other tech companies after Trump returned to the White House. The company hired energy and regulatory experts in-house and through outside counsel. This investment in human capital gave Google a distinct advantage in understanding the arcane world of FERC filings and interconnection procedures.
Brad Simmons, Google’s Washington-based energy and sustainability policy manager, made the company’s position clear at a recent conference. FERC’s approach to the Trump administration’s “speed-to-power” push for data centers has a “direct impact on our infrastructure build-out,” he said. Google recognized early that hiring people who speak the language of utilities and regulators was not optional — it was essential to keeping their data center construction timelines on track.
The company’s preparation goes beyond lobbying. Google has filed detailed comments in FERC proceedings, participated in technical conferences, and built relationships with grid operators across the country. This groundwork positions Google to influence the upcoming FERC proposal in ways that align with its business needs, rather than reacting to rules written without its input.
Other tech companies are now scrambling to catch up. The two-month sprint that followed Swett’s remarks has involved a flurry of meetings, filings, and hiring. The learning curve is steep, but the financial incentives are enormous. A delay of even one year in connecting a major data center can cost hundreds of millions of dollars in lost compute capacity.
What Warning Did Nvidia’s Director Give About Energy Infrastructure?
Nvidia, the dominant maker of chips for advanced AI models, has a unique perspective on this challenge. The company does not build data centers itself, but its GPUs power the vast majority of AI training and inference workloads. If data centers cannot get connected to the grid, Nvidia’s chip sales suffer.
You may also enjoy reading: Jeff Bezos Close to Making a Good Point.
Levi Patterson, Nvidia’s director of energy and AI infrastructure policy, delivered a stark warning at an April conference in Washington. “I don’t think Nvidia has ever filed comments before FERC,” he said. “Nvidia is new to D.C. and trying to figure it out. We need to figure out a better way to really communicate and drive our positioning here.”
Patterson’s admission underscores how unfamiliar the energy regulatory landscape is for even the most powerful tech companies. Nvidia has never filed comments before FERC, according to Patterson. This is a company that dominates its industry, yet it had zero institutional knowledge of the agency that controls the pace of data center grid connections.
The warning from Patterson went beyond regulatory inexperience. He painted a picture of what happens if the energy infrastructure gap is not closed. “Eventually, there’s a world where, if we don’t deploy enough infrastructure, the most powerful [AI] models are out of reach for a lot of the average user,” Patterson continued. This is not a theoretical concern. The computational requirements for frontier models are doubling every few months. Without commensurate growth in data center capacity, the benefits of AI will concentrate among those with access to scarce compute resources.
Nvidia CEO Jensen Huang accompanied President Donald Trump to China earlier this month, a sign of the company’s growing political engagement. But even that high-level access does not solve the fundamental problem of grid interconnection. The solution requires technical expertise, regulatory filings, and patience — none of which come naturally to a company that moves at the speed of silicon.
Who Asked FERC to Draft a Rule on Grid Connections?
The push for federal action on data center grid connections did not originate solely within FERC or the tech industry. A key external actor accelerated the timeline. Energy Secretary Chris Wright asked FERC in October to draft a rule asserting federal jurisdiction over grid connections for data centers.
This request from the Department of Energy carries significant weight. It signals that the executive branch views data center power demand as a matter of national importance, not just a utility industry issue. The Energy Secretary’s involvement also creates political cover for FERC to take more aggressive action than it might otherwise consider.
The request has implications for the balance of power between federal and state regulators. State officials who oppose a federal takeover now find themselves arguing against not just FERC but also the Department of Energy. This alignment of federal agencies makes it more likely that the upcoming FERC proposal will include strong federal preemption of state authority over data center connections.
FERC’s internal discussions, as Swett has indicated, are focused on creating national solutions to the challenges of pulling large-load data centers onto the grid. The commission is aware that piecemeal state-by-state approaches will not work for facilities that consume as much power as entire towns. A national framework, while politically difficult, offers the only realistic path to the speed that AI companies demand.
Frequently Asked Questions
What does it mean for a tech company to “speak FERC”?
Speaking FERC means understanding the language of electric utility regulation, including interconnection procedures, cost allocation methodologies, and the legal frameworks that govern wholesale electricity markets. Tech companies that speak FERC can file effective comments in regulatory proceedings, negotiate meaningfully with grid operators, and anticipate how FERC rulings will affect their data center construction timelines.
How does a federal takeover of grid connections differ from the current process?
Under the current process, state regulators and local utilities control most decisions about connecting new data centers to the grid. A federal takeover would shift that authority to FERC, giving the federal agency power to set timelines, allocate costs, and approve interconnection requests. State officials oppose this shift because it reduces their ability to protect local ratepayers from cost increases associated with data center expansion.
Why are AI data centers so difficult to connect to the existing power grid?
AI data centers consume enormous amounts of electricity — often hundreds of megawatts per facility, equivalent to the demand of a small city. Adding this load to the grid requires upgrades to transmission lines, substations, and transformers. The interconnection process must also ensure that the new load does not destabilize the grid, which involves complex engineering studies and coordination with multiple utilities and grid operators. This process currently takes five to ten years, far too slow for the pace of AI development.
