Large tech companies' data center power demand drives $15 billion PJM grid tender; nuclear SMRs gain momentum.
The construction of America's AI infrastructure is straining America's physical infrastructure, which, regardless of how advanced the chips are, is unprepared for development at this scale. Furthermore, Washington has made a decision regarding who will foot the bill for AI and related technology development—this "reckoning" has therefore forced companies seeking to develop hyperscale data centers to recognize their role in meeting national infrastructure needs. As a result, the structure of regional electricity markets, the financing of next-generation nuclear power plants, and many other aspects of energy procurement have undergone significant changes, with energy costs becoming a major expense for the world's largest technology companies.
The flashpoint is America's largest regional grid operator—PJM Interconnection's service territory extends from Chicago to Washington, D.C. Over the past three years, the combined effects of booming data center development and surging electricity demand have caused residential energy costs to rise an astounding 38% across PJM's entire coverage area, with an increase of 13% from 2020 to 2025 alone.
All of this has placed enormous pressure on the White House and state governments, prompting PJM to conduct an emergency power auction in January to force large technology companies to fund new capacity needed to meet future electricity demand. This pressure ultimately bore fruit on March 4th, when seven leading hyperscale and artificial intelligence companies—Amazon (AMZN), Google (GOOGL), Meta (META), Microsoft (MSFT), OpenAI, Oracle (ORCL), and xAI—signed the Voluntary Taxpayer Protection Pledge.
While the Voluntary Taxpayer Protection Pledge is not legally binding in itself, it does create legally binding financial structures under U.S. law, with all signatories to the Voluntary Taxpayer Protection Pledge agreeing to negotiate new rate structures for data center loads, in which data center operators will bear all costs associated with grid upgrades and new electricity supply rather than pass these costs to residential consumers or non-technology businesses.
The most critical aspect of the Voluntary Taxpayer Protection Pledge is that data centers will pay for the capacity they agree to purchase under their contractual agreements with PJM, regardless of whether the data center operators actually use all the electricity that PJM must reserve under contract. PJM anticipates a 5.2% capacity deficit by 2027-2028, which means approximately $15 billion in new power plant investments will be required. The emergency power auction must be completed by September 2026, with the goal of having each data center operator sign long-term power purchase contracts, effectively ending implicit public subsidies for AI infrastructure.
PJM's power auction has fundamentally transformed electricity from a business-level detail into a balance sheet item that could delay project development. "For example," a Wedbush analyst wrote in a market report, "financing methods once used to build data centers now must account for additional costs associated with retrofitting the electrical grid, such as the reconfiguration of transmission lines, the relaying of substations, and the costs of installing new connections for data center projects—costs that have historically been embedded in utility companies' rate base." This means that due to the 15-year contracts, data center electricity costs could increase by 30 to 50%, which will ultimately be passed on in the form of higher prices for cloud computing and AI services.
Time is of the essence. At the GTC 2026 conference, Nvidia (NVDA) CEO Jensen Huang stated that the limiting factor for silicon is the ability to power chips, infrastructure, models, or applications; he claimed that all three are resource-constrained—land, electricity, and shell capacity as well. Therefore, the Department of Energy's SPARK program (aimed at increasing transmission capacity by at least 50% through wire replacement on existing transmission lines and the addition of advanced technologies) will not be fully addressed until August 2026, at which point DOE funding will begin providing final funding for the program starting from the end of 2026 and early 2027. It typically takes 3 to 5 years to construct core transmission projects; furthermore, without the development of new reliable power sources, the Department of Energy has indicated that the frequency of blackout events could increase 100-fold by 2030.
Faced with grid inefficiency, the high costs of waiting for the mandatory auction, and a limited timeline to secure new energy supplies, hyperscale cloud providers' response has been entirely consistent with Huang's predictions—they have moved quickly to take action and bring energy supply onto their own balance sheets. This strategic shift is the strongest catalyst for the nuclear energy renaissance we are witnessing today.
Rather than bidding on small modular reactors (SMRs) through auctions and hoping for the best, major players in AI are now writing checks directly to reactor developers, helping transform SMRs from a yet-to-be-regulated concept into fully-funded commercial pipelines.
Some examples in this regard include: Meta has reached an agreement with TerraPower to provide funding to support the development of two new Natrium units, with capacity reaching up to 690 megawatts of firm power, deliverable as early as 2032.
Furthermore, Meta's partnership with Oklo has helped advance the development of new nuclear energy in Pike County, Ohio. This advanced nuclear energy technology park—potentially operational as early as 2030—will directly add up to 1.2 gigawatts of clean baseload power to the PJM market and support the company's operations in the region.
Amazon and X-energy plan to deploy more than 5 gigawatts of power across the United States by 2039, the largest SMR commercial deployment target to date.
Additionally, in pursuing energy supply diversification through nuclear power, Microsoft has expanded its nuclear energy efforts by adding nuclear agreements on top of existing combined-cycle gas turbine and renewable energy agreements. Together, these four companies have already secured or will secure more than 10 gigawatts of nuclear capacity to support renewable energy by 2035, with more new opportunities still being announced.
While the megawatt numbers associated with energy alternatives are important, their impact extends far beyond that. "Bringing a first-rate corporate balance sheet into an industry historically financed through regulated, rate-payer-supported revenue streams has major significance for the industry's overall credit profile," says Shioly Dong, senior analyst at BMI.
Commercial lenders and institutional investors need this credit profile to provide the financing required for construction. "Modular design concepts have smaller scale and shorter construction timeframes; therefore, their upfront capital risk levels are significantly reduced," says Tim Winter, portfolio manager at Gabelli Utilities Fund (GABUX). Furthermore, HSBC analysts note that the technology sector's willingness to accept cost overruns or schedule delays will be a key factor in determining the extent to which these agreements actually facilitate advanced reactor deployment and the expansion of fuel supply chains and advanced reactor technology parks, as these projects begin to attract direct investment, which would be considered unachievable from the perspective of traditional nuclear financing.
While nuclear and solar power cannot meet the continuous service requirements needed for artificial intelligence (AI) workloads due to the inability of nuclear plants to rapidly scale output up or down, long-duration energy storage systems have become a necessary component of AI data center energy supply. Specifically, flow battery systems may be used alongside renewable energy generation and nuclear plants due to their favorable safety characteristics.