In recent years, a surprising race towards power generation has captivated the attention of tech giants in Silicon ValleyCentral to this ongoing saga is Sam Altman, the co-founder of OpenAI and the chairman of Oklo, a startup focused on nuclear fission technologyThe company's stock has skyrocketed since its public listing on May 11 last year, witnessing an astonishing near 900% increase in just five months, marking it as one of the highest-gaining stocks in the U.S. market.
Altman isn't alone in his pursuits; a host of notable figures from the tech and investment landscapes, such as Bill Gates, Jeff Bezos, Peter Thiel, Larry Ellison, and Cathie Wood, have also placed their bets on the renewables landscape, especially nuclear energy.
The staggering energy demand related to artificial intelligence (AI) operations, such as those powering ChatGPT, has prompted tech giants to significantly scale up their investments in electricity to meet the soaring requirementsIt's reported that a single ChatGPT interaction consumes nearly ten times the electricity of a traditional Google search, fostering the urgency to bolster energy infrastructures.
One of the most alarming statements comes from Jensen Huang, co-founder of NVIDIA, who pointed out, "NVIDIA helps to improve computational efficiency and reduce energy consumption, but without faster compute speeds, we might need 14 planets, three galaxies, and four suns to fuel all of this." His remarks underscore the intricate relationships among technology, energy resources, and their implications for the future of AI.
This discourse on energy primarily revolves around the supply side; however, a complete power system encompasses infrastructure from electric grids and distribution lines to various electrical equipment, including generators and transformersTransitioning to renewable sources such as nuclear, wind, and solar requires a suite of supporting facilities—uranium fuel, nuclear machinery, and energy storage systems—representing a multifaceted challenge.
Many Western countries grapple with outdated electrical grid infrastructures, much of which remains severely neglected
Advertisements
In fact, most transformers in the United States rely heavily on importsThus, nations outside of China must face the reality of needing massive investments to address these long-standing deficits in energy infrastructure.
The dawn of what might be termed a historical cycle in energy acceleration is firmly upon us, spurred by advances in AI technology.
The notion of energy constraining the development of Artificial General Intelligence (AGI) has emerged as a concernSam Altman has claimed that breakthroughs in energy are imperative for the future of AI, emphasizing that energy consumption will likely outpace current expectationsElon Musk has similarly warned of impending shortages, stating that in the next year, “the electrical supply will not meet the demands of all chips.”
Each interaction with generative AI, alongside the underlying data centers, computational power, and the chip industry, consumes an enormous quantity of electricityAccording to some estimates, ChatGPT currently responds to approximately 200 million requests daily, consuming upwards of 500,000 kilowatt-hours of energy, equating to the power used by 17,000 American households.
The costs associated with training AI models have also risen to staggering heights; it’s estimated that training the GPT-3 model incurs costs of around $1.4 million per session, with larger language models costing between $2 million and $12 millionAlarmingly, about 60% of those costs stem from electricity alone.
Data centers, often described as massive energy consumers, further exacerbate the issueFor instance, a large Amazon data center can consume an entire medium-sized city’s annual electricity supplyThe U.SDepartment of Energy's reports indicate that from 2014 to 2023, electricity consumption by data centers surged from 58 terawatt-hours (TWh) to a shocking 176 TWh, representing about 4.4% of the country's total energy usageProjections suggest this figure could balloon to between 325 and 580 TWh by 2028, accounting for as much as 12% of America's overall electricity consumption.
Today, generative AI remains in its infancy
Advertisements
The proliferation of advanced AI models, akin to DeepSeek and ChatGPT, is poised to create an unprecedented demand for electricityThis raises significant questions regarding how future power needs will be met, particularly given the prevailing reliance on traditional fossil fuels which are costly and incompatible with the ideals of a low-carbon futureConversely, while renewable sources like wind and solar are critical, their unreliability further complicates energy strategies, positioning nuclear power as an attractive option for many tech heavyweights.
Among these options, Small Modular Reactors (SMRs)—noted for their shorter construction times, lower costs, higher safety profiles, and greater site adaptability—have garnered particular attention from AI companiesIn the U.S., government subsidies for SMRs make investments in smaller nuclear installations more lucrative compared to larger power plantsStatistical insights reveal that the cost of electricity from SMR projects hovers around $180 per megawatt-hour (MWh) but can diminish to approximately $100/MWh after subsidies, a figure competitive with wind and solar generation.
Since October last year, there’s been a rush among tech giants in Silicon Valley to amass nuclear power options:
- Microsoft has partnered with Constellation Energy in a $1.6 billion initiative to reactivate Reactor 1 at the Three Mile Island plant, with an eye toward powering Microsoft’s data centers by 2028.
- Google has signed an agreement to procure electricity from seven reactors at the nuclear startup Kairos Power, totaling 500 MWh.
- Amazon has followed suit, investing $500 million in three separate agreements with Dominion Energy to develop an SMR project, financing four advanced SMRs at Energy Northwest and establishing a data center adjacent to Talen Energy’s site in Pennsylvania.
- Lastly, Larry Ellison of Oracle has announced plans for a nuclear power data center to be powered by three small reactors producing over 1,000 megawatts.
However, despite these ambitious initiatives, the output of the projected small nuclear plants still only scratches the surface when compared to the astronomical electricity demands required globally
Advertisements
Effectively addressing this energy crisis will necessitate a colossal wave of investment across the board.
As various nations and corporations craft grand strategic blueprints for new energy developments, the limitations posed by outdated grids and electrical equipment have become glaring hindrances to making meaningful progress towards renewable energy integrationFor instance, the UK government vowed to achieve 100% clean power by 2035, yet commitments to various wind and solar projects have been thwarted by an inability to connect new energy sources to the grid promptly.
The antiquated nature of existing grid infrastructures means that vast amounts of renewable energy remain stranded in power plants, with some wind and solar projects waiting over a decade to connect to the gridUK National Grid anticipates a five-fold increase in high-voltage transmission lines needed by 2030 compared to the past 30 yearsChallenges faced by the U.K. resonate with similar realities across Europe; as of late 2022, countries such as the UK, France, and Spain had nearly 596 gigawatts (GW) of waiting wind and solar capacity, with the average wait for these projects to connect to the grid ranging from three to seven years.
Most European high-voltage grid systems were constructed between the 1950s and 1980s, many of which have surpassed 70 years of service; similarly, American power systems and transformers, built predominantly in the 1950s to 1970s, have seen approximately 70% in operation for over 25 years, with 60% of circuit breakers lasting more than 30 years and transformers designed for a lifespan of only 35 to 40 years.
The predicament many countries face regarding aging electrical infrastructures reveals a dire need for upgradesWithout swift interventions, new wind, solar, and nuclear installations could face obsolescenceRecognizing this, the U.SDepartment of Energy has initiated the Grid Resilience Innovation Partnership (GRIP) aiming to invest $10.5 billion in grid modernization over five years
Meanwhile, the European Union has put forth the “European Grid Action,” planning to inject 584 billion euros in investments before 2030. Additionally, the UK National Grid has committed to a £60 billion investment in network infrastructure over the next five financial years, nearly doubling the funds allocated in the previous five years.
The resulting wave of investment is already manifesting favorable impacts on the stock performance and earnings of electrical companies, with traditional giants seeing their stocks outpace even AI market starsGeneral Electric (GE), despite its former complexities and restructuring, has witnessed its spinoff company, GE Vernova, shine brightly since its independence in April last year, with a remarkable stock appreciation exceeding 165%. Reporting a net profit of $1.55 billion for 2024, its profits have surged by over 454% year-on-year.
Siemens Energy, emerging as a spinoff from the original Siemens, has enjoyed a staggering 331% stock increase over the past year, while enduring demand for grid modernization has resulted in a record backlog of orders reaching 123 billion eurosEven as their new manufacturing plant in the U.S. is still under construction, their future two-year capacity has already been sold out, showcasing robust demand dynamics.
Long-standing electrical manufacturers are also enjoying rejuvenated fortunes; for example, Hitachi has seen its stock climb by 67.96%, with Schneider Electric and ABB similarly registering increases of 42.66% and 40.69%, respectivelyIn contrast, during the same period, the renowned AI chip maker NVIDIA recorded a stock increase of “only” 171% and a profit surge of 190.64% (Q3 2024).
Parallel to the historic push for electricity stemming from industrialization and urbanization, the current AI boom has initiated a new era of power investmentScott, the CEO of GE Vernova, has articulated that we are witnessing the onset of a “super cycle” in electricity investment.
Elon Musk has been vocal about the impending shortages of transformers in the U.S., arguing that with the confluence of electric vehicles and AI requiring power, a tremendous demand surge for energy equipment and generation is unfolding.
A comprehensive electricity ecosystem involves not only ample power supply but also streamlined electrical grids, along with supporting gear such as transformers, meters, energy storage, and smart control centers
In fact, transformers have emerged as a critical component concerning the upgrading of U.S. power structuresMuch like photolithography systems in the Chinese semiconductor industry, transformers lack essential materials like oriented silicon steel, leading to a precarious reliance on imports for 80% of the country’s electrical transformers.
Oriented silicon steel, vital for transformer cores, was first discovered by N.PGoss in the 1930s and remains crucial todayHowever, its scarcity now hampers the modernization of U.S. electrical grids, which could profoundly impact the trajectories of AI developmentComprising a considerable component of overall transformer costs, oriented silicon steel is predominantly produced in East Asia, particularly in China, Japan, and South Korea, while domestic production levels in the U.S. are significantly limited.
Transformer imports into the United States have surged over the past five years, totaling approximately $5.8 billion in 2023, a 48.7% year-on-year increaseNotably, imports reached a staggering $7.32 billion in the first 11 months of 2024, reflecting a 41.16% growth and setting a record high for transformer imports.
With high transport efficiency and minimal losses, ultra-high voltage systems are becoming a global trend in grid modernizationTransformers function as the critical “magician” that adeptly adjusts voltages, enabling efficient long-distance electricity transmissionPower generated from generators is boosted through large transformers to extreme voltages (ranging from 161 kV to 765 kV) to facilitate this transmission before being reduced to 15-34.5 kV and finally dialed down to 240 V for consumer and industrial use.
The rising costs of oriented silicon steel and copper, compounded by disruptions in maritime transport, have created an imbalance in supply and demandConsequently, transformer prices have surged by 60%-70% between 2020 and 2023, with indications of further increases
Presently, the delivery times for transformers stretch between 50 and 150 weeks, with worst-case scenarios extending up to five years.
While China is a primary producer of transformers, in light of growing trade barriers, the U.S. primarily sources their transformers from Mexico, South Korea, Brazil, and Canada, with most products coming from manufacturers like Hitachi, GE Vernova, Siemens Energy, SGB-SMIT, and Hyosung.
It’s also critical to note that the export of transformers from China represents only around 8% of the total market, disputing the narrative that China is deliberately obstructing U.S. power equipmentEssentially, this market operates on principles of competition, differing markedly from the technology and trade monopolies evident in high-end semiconductor industries.
Nonetheless, transformers are merely a reflection of the challenges faced in the broader scope of electrical modernization in the WestBeyond transformers, the issues extend to smart meters, relays, capacitors, sensors, inverters, generators, and other essential infrastructure, coupled with the need for a smarter, digitized grid systemAll the mounting stressors are the legacies of decision-making made over several decades, largely aimed at de-industrialization.
In conclusion, sociologist Kate Crawford poignantly articulated, “Artificial Intelligence is both embodied and material, composed of natural resources, fuels, labor, infrastructure, logistics, histories, and classifications—all carrying costs.” Her insights bridge the seemingly virtual domain of AI with the tangible energy realm.
The shift of power dynamics reflected in data and technology necessitates recognition, as more individuals acknowledge that “AI is the oil resource of the 21st century.” Yet, the cherished yet crumbling energy systems of the West have not adequately prepared for such monumental shifts in historyOver the decades, monopolistic enterprises have failed to align themselves with market demands.
Within this vast electricity investment cycle, differing narrative paths unfold for China, the U.S., and Europe
Advertisements
Advertisements