The decisive factor in global artificial intelligence competitions has officially shifted from chip delivery cycles to power and grid infrastructure. As the power demand of hyperscale data centers surges exponentially, electricity has not only become the core of hundreds of billions of dollars in capital expenditures by tech giants such as Microsoft and Amazon, but also triggered strategic layouts at the U.S. national security level due to the severe shortages and geopolitical risks surrounding key equipment like transformers. In the AI era, computing power is equivalent to national strength, and the sole prerequisite for guaranteeing computing power is robust and uninterrupted power grid support. This computing power-driven energy scramble has also reshaped the strategic status of the global power supply chain.
Electric power has become the top capital expenditure
For a long time, the biggest challenge facing the technology industry has been the lead time for hardware and chips. However, entering 2026, the development bottleneck has shifted from servers to substations. Fueled by the rapid expansion of the large-scale data center industry, electricity has become the primary growth constraint between 2023 and 2026. According to forecasts by the International Energy Agency (IEA) and Data Center Dynamics, global data centers consumed approximately 460 terawatt-hours (TWh) of electricity in 2022, with the figure projected to peak above 1050 TWh by 2026. Artificial intelligence and high-performance workloads are growing far faster than power grids can accommodate. The unbreakable rule governing modern data center development is simple: no power, no project.
Meanwhile, the latest report released by JLL in 2026 clearly points out that the power density required by AI training facilities is ten times that of traditional data centers. It is estimated that by 2030, AI workloads will account for more than half of the total global data center capacity, doubling compared with 2025. Artificial intelligence has risen to become a core national strategy, prompting countries worldwide to compete in investing in sovereign infrastructure to strengthen domestic capabilities, with an estimated capital investment of up to 8 billion US dollars driven by 2030. To power these energy-intensive facilities, major US hyperscale data center operators including Microsoft, Alphabet, Amazon, Meta and Oracle plan to invest nearly 700 billion US dollars in capital expenditure for artificial intelligence from 2025 to 2026, the vast majority of which is aimed at ensuring the stability of infrastructure and power supply networks.
Transformer Shortages and Grid Interconnection Bottlenecks
The essence of this crisis lies in the fact that infrastructure development fails to keep pace with technological iteration. A report by venture capital firm Bessemer Venture Partners lays bare the harsh reality: the physical construction of a data center can be completed in 12 to 18 months, yet connecting it to the existing power grid takes five to seven years. This massive timeline gap has forced more than a quarter of new data center projects to be delayed due to power supply and permitting issues. Reports released by multiple regional grid operators across the United States in early 2026 all warn that the queue for grid interconnection of large power loads is growing exponentially. In Texas and the U.S. Midwest alone, the capacity of ultra-large data centers waiting for grid connection is projected to soar to 173 gigawatts by 2030, far exceeding the current grid load limit.
However, even with sufficient power generation, the lack of transformers—devices that convert high-voltage electricity into voltages usable by data centers—can bring all computing power progress to an immediate halt. Competition for transformers has reached a fever pitch, and lengthy delivery lead times are dictating the progress of power engineering and the development of AI infrastructure. According to a report by Wood Mackenzie, from 2019 to 2025, U.S. demand for Generator Step-Up (GSU) transformers surged by 274%, while demand for power transformers rose by 116%, with supply gaps standing at 6% and 30% respectively. Large power transformers that once required only around 50 weeks for delivery now have an average lead time of over 120 weeks, with the waiting period for some high-spec equipment extending to several years. The market imbalance has driven a sharp rise in transformer delivery times and prices, prompting U.S. power transformer buyers to scramble for imported products and factory production slots, fearing that tens of billions of dollars in AI infrastructure deployment will stall at the final stage.
Recognizing the significance of data center construction for national security and U.S. competitiveness in the global artificial intelligence race, the U.S. government took action in April 2026. U.S. President Donald Trump invoked Section 303 of the Defense Production Act to officially designate large-scale power grid infrastructure as a national defense necessity, and authorized emergency federal funding to expand domestic supply of key components in this supply chain.
Power grid upgrading faces challenges in Chinese supply chain and cybersecurity.
Due to the United States’ lack of domestic manufacturing capacity for ultra-high-voltage transformers, while 70% to 90% of the global production capacity of key power grid components is highly concentrated in China, amid the dual pressures of global geopolitical competition and tariff barriers, major U.S. tech companies have actively relocated server production lines out of Asia yet face an extremely fragile supply chain situation for core power transmission equipment.
To meet the massive power demand driven by AI data centers, the United States is comprehensively advancing power grid expansion and modernization projects. Supported by federal and state government initiatives, U.S. utility companies are actively deploying Advanced Metering Infrastructure (AMI), AI analytics systems, Battery Energy Storage Systems (BESS), and Distributed Energy Resource Management Systems (DERMS). These digital technologies transform traditional one-way power transmission systems into intelligent, real-time responsive networks, enabling more efficient resource management and laying a rapid expandable power foundation for data centers required for advanced AI applications.
However, while grid digitalization boosts efficiency, it also introduces new cyber risks. Unencrypted communication protocols and persistent remote access functions in grid equipment can easily become vulnerabilities for hackers to modify settings, disrupt services, or inject erroneous data. Compounding these risks, 70% to 90% of critical U.S. power grid equipment is manufactured by Chinese producers, and the U.S. lacks manufacturing capabilities for key assets such as ultra-high-voltage transformers. This leaves the U.S. confronting severe supply chain and national security risks while advancing grid digital transformation. To mitigate such hazards, the U.S. federal government has recently enacted legislation establishing the **Foreign Entity of Concern (FEOC) restrictions, which mandate that projects must meet a minimum threshold of non-FEOC component usage to qualify for tax credits. This places utility companies in a dilemma: they must rapidly expand infrastructure to accommodate the surging power demand of data centers, while striving to diversify supply chains and comply with stringent cybersecurity and regulatory requirements amid a shortage of viable alternative components.
To address the soaring demand for transformers, major U.S. power grid equipment manufacturers are ramping up production capacity. According to forecasts by Hitachi Energy, U.S. demand for power grid infrastructure will continue to grow for at least the next decade. Hitachi Energy has announced investments of over 1 billion U.S. dollars to build a new factory in South Boston, scheduled to commence operations in 2028, and an additional 106 million U.S. dollars to construct a transformer plant in Alamo, Tennessee. Siemens also increased its investment in its North Carolina manufacturing facility from 150 million U.S. dollars to 421 million U.S. dollars in February 2026, including a new transformer plant in Charlotte set to start production within the year.
Unlike standardized electronic products, large high-voltage transformers typically require customized design with long manufacturing lead times, and their production is highly dependent on specialized materials and professional manufacturing expertise, making rapid capacity expansion extremely challenging. Furthermore, as countries worldwide simultaneously promote the development of AI data centers, electric vehicles, renewable energy, power grid upgrades, and industrial electrification, a global competition for power grid equipment has emerged, and power grid development has become a key component of national competitiveness.
According to the 2026 latest report released by McKinsey & Company, global investments of up to 7 trillion U.S. dollars will be required for data center-related infrastructure by 2030 to meet soaring computing power demand, with a large proportion directly allocated to energy infrastructure such as power generation and cooling systems. Infrastructure funds and private equity firms have also begun direct investments in power plants and large-scale acquisitions of renewable energy projects. In the United States, an increasing number of hyperscale cloud providers are considering building self-owned power generation and energy storage systems to secure stable power supply. Tech giants including Microsoft, Amazon, Meta, and Google have invested heavily in securing capacity from conventional nuclear power plants. For instance, Microsoft signed a long-term agreement with Constellation Energy to restart unit operations at the Three Mile Island Nuclear Generating Station and comprehensively advance the commercialization of Small Modular Reactors (SMRs). Meta signed nuclear power agreements totaling over 6 gigawatts (GW) in early 2026, while Amazon has invested in X-energy and targeted a future SMR deployment capacity of 5 GW. As of early 2026, major U.S. tech companies have accumulated tens of billions of U.S. dollars in nuclear power procurement orders, elevating power procurement from a basic facility management task to a top-tier strategic decision at the corporate board level.
The Next Arms Race in Computing Power: Grid Warfare
With ultra-large data centers successively adopting self-developed processors, customized chips are expected to capture a 15% market share by 2030. Emerging technologies such as neuromorphic computing are also poised to reduce infrastructure demands and improve energy efficiency. The sole prerequisite for guaranteeing computing power is robust and uninterrupted power grid support.
Over the past two decades, the core of global technological competition has centered on the chip war. In the next decade, the focus of competition may shift back to energy flow. Artificial intelligence will no longer be merely a software industry, but a capital-intensive industry highly reliant on physical infrastructure. This means power grids, transformers, power distribution equipment, energy storage systems and power management facilities will no longer be categorized as traditional industrial assets, but will evolve into new-generation strategic technological assets. Multinational corporations and tech giants must classify heavy power equipment for infrastructure construction as top-tier strategic materials and even directly engage in upstream industrial layout to secure a stable supply chain.
The decisive factor in global artificial intelligence competitions has officially shifted from chip delivery cycles to power and grid infrastructure. As the power demand of hyperscale data centers surges exponentially, electricity has not only become the core of hundreds of billions of dollars in capital expenditures by tech giants such as Microsoft and Amazon, but also triggered strategic layouts at the U.S. national security level due to the severe shortages and geopolitical risks surrounding key equipment like transformers. In the AI era, computing power is equivalent to national strength, and the sole prerequisite for guaranteeing computing power is robust and uninterrupted power grid support. This computing power-driven energy scramble has also reshaped the strategic status of the global power supply chain.
Electric power has become the top capital expenditure
For a long time, the biggest challenge facing the technology industry has been the lead time for hardware and chips. However, entering 2026, the development bottleneck has shifted from servers to substations. Fueled by the rapid expansion of the large-scale data center industry, electricity has become the primary growth constraint between 2023 and 2026. According to forecasts by the International Energy Agency (IEA) and Data Center Dynamics, global data centers consumed approximately 460 terawatt-hours (TWh) of electricity in 2022, with the figure projected to peak above 1050 TWh by 2026. Artificial intelligence and high-performance workloads are growing far faster than power grids can accommodate. The unbreakable rule governing modern data center development is simple: no power, no project.
Meanwhile, the latest report released by JLL in 2026 clearly points out that the power density required by AI training facilities is ten times that of traditional data centers. It is estimated that by 2030, AI workloads will account for more than half of the total global data center capacity, doubling compared with 2025. Artificial intelligence has risen to become a core national strategy, prompting countries worldwide to compete in investing in sovereign infrastructure to strengthen domestic capabilities, with an estimated capital investment of up to 8 billion US dollars driven by 2030. To power these energy-intensive facilities, major US hyperscale data center operators including Microsoft, Alphabet, Amazon, Meta and Oracle plan to invest nearly 700 billion US dollars in capital expenditure for artificial intelligence from 2025 to 2026, the vast majority of which is aimed at ensuring the stability of infrastructure and power supply networks.
Transformer Shortages and Grid Interconnection Bottlenecks
The essence of this crisis lies in the fact that infrastructure development fails to keep pace with technological iteration. A report by venture capital firm Bessemer Venture Partners lays bare the harsh reality: the physical construction of a data center can be completed in 12 to 18 months, yet connecting it to the existing power grid takes five to seven years. This massive timeline gap has forced more than a quarter of new data center projects to be delayed due to power supply and permitting issues. Reports released by multiple regional grid operators across the United States in early 2026 all warn that the queue for grid interconnection of large power loads is growing exponentially. In Texas and the U.S. Midwest alone, the capacity of ultra-large data centers waiting for grid connection is projected to soar to 173 gigawatts by 2030, far exceeding the current grid load limit.
However, even with sufficient power generation, the lack of transformers—devices that convert high-voltage electricity into voltages usable by data centers—can bring all computing power progress to an immediate halt. Competition for transformers has reached a fever pitch, and lengthy delivery lead times are dictating the progress of power engineering and the development of AI infrastructure. According to a report by Wood Mackenzie, from 2019 to 2025, U.S. demand for Generator Step-Up (GSU) transformers surged by 274%, while demand for power transformers rose by 116%, with supply gaps standing at 6% and 30% respectively. Large power transformers that once required only around 50 weeks for delivery now have an average lead time of over 120 weeks, with the waiting period for some high-spec equipment extending to several years. The market imbalance has driven a sharp rise in transformer delivery times and prices, prompting U.S. power transformer buyers to scramble for imported products and factory production slots, fearing that tens of billions of dollars in AI infrastructure deployment will stall at the final stage.
Recognizing the significance of data center construction for national security and U.S. competitiveness in the global artificial intelligence race, the U.S. government took action in April 2026. U.S. President Donald Trump invoked Section 303 of the Defense Production Act to officially designate large-scale power grid infrastructure as a national defense necessity, and authorized emergency federal funding to expand domestic supply of key components in this supply chain.
Power grid upgrading faces challenges in Chinese supply chain and cybersecurity.
Due to the United States’ lack of domestic manufacturing capacity for ultra-high-voltage transformers, while 70% to 90% of the global production capacity of key power grid components is highly concentrated in China, amid the dual pressures of global geopolitical competition and tariff barriers, major U.S. tech companies have actively relocated server production lines out of Asia yet face an extremely fragile supply chain situation for core power transmission equipment.
To meet the massive power demand driven by AI data centers, the United States is comprehensively advancing power grid expansion and modernization projects. Supported by federal and state government initiatives, U.S. utility companies are actively deploying Advanced Metering Infrastructure (AMI), AI analytics systems, Battery Energy Storage Systems (BESS), and Distributed Energy Resource Management Systems (DERMS). These digital technologies transform traditional one-way power transmission systems into intelligent, real-time responsive networks, enabling more efficient resource management and laying a rapid expandable power foundation for data centers required for advanced AI applications.
However, while grid digitalization boosts efficiency, it also introduces new cyber risks. Unencrypted communication protocols and persistent remote access functions in grid equipment can easily become vulnerabilities for hackers to modify settings, disrupt services, or inject erroneous data. Compounding these risks, 70% to 90% of critical U.S. power grid equipment is manufactured by Chinese producers, and the U.S. lacks manufacturing capabilities for key assets such as ultra-high-voltage transformers. This leaves the U.S. confronting severe supply chain and national security risks while advancing grid digital transformation. To mitigate such hazards, the U.S. federal government has recently enacted legislation establishing the **Foreign Entity of Concern (FEOC) restrictions, which mandate that projects must meet a minimum threshold of non-FEOC component usage to qualify for tax credits. This places utility companies in a dilemma: they must rapidly expand infrastructure to accommodate the surging power demand of data centers, while striving to diversify supply chains and comply with stringent cybersecurity and regulatory requirements amid a shortage of viable alternative components.
To address the soaring demand for transformers, major U.S. power grid equipment manufacturers are ramping up production capacity. According to forecasts by Hitachi Energy, U.S. demand for power grid infrastructure will continue to grow for at least the next decade. Hitachi Energy has announced investments of over 1 billion U.S. dollars to build a new factory in South Boston, scheduled to commence operations in 2028, and an additional 106 million U.S. dollars to construct a transformer plant in Alamo, Tennessee. Siemens also increased its investment in its North Carolina manufacturing facility from 150 million U.S. dollars to 421 million U.S. dollars in February 2026, including a new transformer plant in Charlotte set to start production within the year.
Unlike standardized electronic products, large high-voltage transformers typically require customized design with long manufacturing lead times, and their production is highly dependent on specialized materials and professional manufacturing expertise, making rapid capacity expansion extremely challenging. Furthermore, as countries worldwide simultaneously promote the development of AI data centers, electric vehicles, renewable energy, power grid upgrades, and industrial electrification, a global competition for power grid equipment has emerged, and power grid development has become a key component of national competitiveness.
According to the 2026 latest report released by McKinsey & Company, global investments of up to 7 trillion U.S. dollars will be required for data center-related infrastructure by 2030 to meet soaring computing power demand, with a large proportion directly allocated to energy infrastructure such as power generation and cooling systems. Infrastructure funds and private equity firms have also begun direct investments in power plants and large-scale acquisitions of renewable energy projects. In the United States, an increasing number of hyperscale cloud providers are considering building self-owned power generation and energy storage systems to secure stable power supply. Tech giants including Microsoft, Amazon, Meta, and Google have invested heavily in securing capacity from conventional nuclear power plants. For instance, Microsoft signed a long-term agreement with Constellation Energy to restart unit operations at the Three Mile Island Nuclear Generating Station and comprehensively advance the commercialization of Small Modular Reactors (SMRs). Meta signed nuclear power agreements totaling over 6 gigawatts (GW) in early 2026, while Amazon has invested in X-energy and targeted a future SMR deployment capacity of 5 GW. As of early 2026, major U.S. tech companies have accumulated tens of billions of U.S. dollars in nuclear power procurement orders, elevating power procurement from a basic facility management task to a top-tier strategic decision at the corporate board level.
The Next Arms Race in Computing Power: Grid Warfare
With ultra-large data centers successively adopting self-developed processors, customized chips are expected to capture a 15% market share by 2030. Emerging technologies such as neuromorphic computing are also poised to reduce infrastructure demands and improve energy efficiency. The sole prerequisite for guaranteeing computing power is robust and uninterrupted power grid support.
Over the past two decades, the core of global technological competition has centered on the chip war. In the next decade, the focus of competition may shift back to energy flow. Artificial intelligence will no longer be merely a software industry, but a capital-intensive industry highly reliant on physical infrastructure. This means power grids, transformers, power distribution equipment, energy storage systems and power management facilities will no longer be categorized as traditional industrial assets, but will evolve into new-generation strategic technological assets. Multinational corporations and tech giants must classify heavy power equipment for infrastructure construction as top-tier strategic materials and even directly engage in upstream industrial layout to secure a stable supply chain.
