The evolution of computing power begins with AI, is realized through energy, and is perfected through collaboration.

Author and source: New Energy Industry Home

Editor's Note:

The transformation of computing power is achieved through energy.

AI data centers are evolving from mere "major electricity consumers" into collaborative builders of the new power system.

This transition from "passive power supply" to "compute-power synergy" represents not only a shift in technological paradigm but also the core essence of the national "15th Five-Year Plan" new infrastructure strategy.

The author of this article, Liu Yuankun, graduated from Tsinghua University’s Department of Electrical Engineering. He was a persistent entrepreneur and innovator in the energy and power sector, and now serves as Senior Vice President at Century Link, a leading AIDC service provider in China.

This rare cross-disciplinary experience in “new energy + AI computing power” has given him a foundational understanding of modern power systems and a sharp, hands-on perspective on AIDC.

He believes: "The great energy technology companies of the future may emerge from the field of computing and power synergy."

In this comprehensive article, he will address: How should the power grid become more resilient in the era of high-power-density, high-capacity AIDC? How will computing power reciprocate and support the new power system? And where are the opportunities for new energy and energy storage?

The following is the main text:

The explosive advancement of generative AI is fundamentally reshaping the underlying logic of the digital economy and energy systems.

AI technologies, represented by large language models and open-source agents, are rapidly evolving, driving exponential growth in computing power demand. AI data centers have transformed from traditional information infrastructure into a new type of electrical load characterized by high power density, strong variability, and high controllability potential.

Becoming the third-largest electricity consumer after industry and commerce and residential users, profoundly transforming the entire chain of power system planning, operation, markets, and services.

In 2026, "computing-power and electricity coordination" was officially incorporated into the national new infrastructure strategy and the 15th Five-Year Plan.

Signifies an industry transformation in which computing power and electricity shift from unidirectional supply to bidirectional integration, and from passive adaptation to active collaboration, injecting new meaning into the construction of a next-generation power system.

01 The Tide of the Era: Explosive Growth in Computing Power Is Reshaping the Power System Surging Demand for Computing Power Is Restructuring the Power Load Landscape

According to data from the International Energy Agency, global data center electricity consumption reached 4,150 billion kilowatt-hours in 2024, accounting for approximately 1.3% of global total electricity use;

Institutions such as BloombergNEF predict that by 2030, this figure will exceed 18,000 billion kilowatt-hours, accounting for 6%–8% of total consumption, with computing power electricity use becoming a key factor reshaping the global power landscape.

AI computing centers differ fundamentally from traditional data centers: rack power consumption has increased from 7–10 kW to 30–100 kW, with some high-end racks reaching 120 kW and even megawatt levels;

Compute centers have grown from a few megawatts in a single building to tens of megawatts, and now to gigawatt-scale compute facilities, with future superprojects reaching up to 10 GW.

Large-scale deployment of heterogeneous chips such as GPUs and TPUs results in clusters operating at full capacity during training and experiencing random load fluctuations during inference, with millisecond-level power spikes capable of causing grid impacts of hundreds of megawatts.

From a full lifecycle perspective of computational power, model training exhibits sustained peaks and high-frequency oscillations, parameter fine-tuning shows intermittent fluctuations, and online inference is characterized by strong bursts and frequent alternations between peaks and troughs, completely disrupting the traditional pattern of stable and orderly load operations.

This new load characteristic presents unprecedented challenges to traditional power system functions such as load forecasting, dispatch control, grid planning, and market pricing, rendering the conventional "source follows load" model inadequate for accommodating the dynamic nature of computing power loads.

A typical case occurred in July 2024 in Virginia, USA, where a lightning strike caused multiple voltage sags on the transmission line, triggering protective shutdowns simultaneously across dozens of large-scale computing and data centers in the region. This resulted in approximately 1,500 megawatts of load being disconnected from the grid in a short time, causing significant fluctuations in grid frequency and voltage.

The grid avoided instability only because the dispatch authority urgently reduced power generation, highlighting the severe challenge posed by high-density computing loads to the safe operation of the power grid.

Mismatch between power supply and demand requires urgent coordination.

China's computing power demand and electricity supply exhibit significant regional mismatches:

Eastern regions such as the Yangtze River Delta, Beijing-Tianjin-Hebei, and the Pearl River Delta account for over 80% of the nation's computing power demand, but suffer from scarce energy resources and tight power supply; in contrast, western regions are rich in renewable energy sources like wind, solar, and hydropower, yet face challenges of insufficient consumption and curtailment of wind and solar power.

The national "East Data, West Computing" strategy guides the deployment of computing hubs toward western regions. As of the first quarter of 2025, the total computing capacity of the eight national computing hub nodes reached 215.5 EFlops, with intelligent computing accounting for 80.8%. However, computing and power scheduling remain relatively independent, and a market-driven collaborative mechanism has yet to be fully established.

Globally, the surge in AI computing power is driving profound changes in electricity supply and demand.

Data from the 2026 Cambridge Energy Week shows that global data center electricity consumption grew at an annual rate of 18% from 2023 to 2026, with AI computing's share rising from 15% to 35%, reaching a total electricity demand of 1,050 terawatt-hours in 2026—equivalent to Germany’s annual electricity consumption.

In U.S. power markets such as PJM and Texas ERCOT, the concentrated deployment of AI data centers has caused supply-demand imbalances, leading to soaring capacity market prices and frequent grid congestion, compelling accelerated reforms to grid connection rules and market mechanisms.

A typical example is Ireland, where the concentrated commissioning of AI computing centers has led to electricity consumption accounting for 22% of the nation’s total usage, saturating regional grid capacity and causing insufficient spacing. This prompted a temporary halt in approving new computing projects for grid connection, forcing some cloud providers to delay project implementation due to inability to connect on schedule, highlighting the acute conflict between explosive growth in computing power and inadequate grid supply.

Led by national strategy, coordinated computing and power has become an industry consensus.

From the conservation phase of PUE optimization in the "Green Computing Center" to the power-switching phase of "Green Power Computing" supported by green certificates and power purchase agreements.

By reaching the stage of "computing-power and electricity synergy," where computing power and the power grid engage in bidirectional interaction and coexist, China has completed a three-level leap in its computing power energy transition.

The government work report explicitly proposes "implementing new infrastructure projects such as ultra-large-scale intelligent computing clusters and computing-power-electricity coordination."

The 15th Five-Year Plan explicitly promotes the coordinated development of green power and computing power. The integration of computing and power has evolved from an industry-level topic to a national strategy and has become one of the key directions in building a new power system.

Currently, integrating AI and embracing intelligent transformation has become an inevitable trend for the energy and power industry to adapt and develop.

For power companies, this is a valuable opportunity for transformation and upgrading—allowing them to proactively embrace societal change and steadily evolve from traditional power suppliers into enablers of computing infrastructure.

Leveraging diversified pathways such as customized power supply, green power trading, flexible scheduling, and energy storage integration, we are building an integrated “power–computing–storage–carbon” service system to steadily embark on a new journey of high-quality, digital, green, and coordinated development in the power industry.

Looking to the future, the development of the power industry is not merely about incremental functional upgrades, but also presents a unique opportunity to achieve profound architectural transformation and an upgrade of its value logic.

Both the IT engineering and scientific research fields are fully adapting to the AI-native paradigm, and the power grid industry is now entering an excellent window of development, with the potential to evolve into an AI-native power grid and seize opportunities to achieve value enhancement in the wave of computing-power synergy.

For example, Fluence, centered on grid-scale energy storage systems, provides megawatt-level flexible regulation and backup power support, smoothing grid fluctuations and accelerating the grid connection of computing centers.

Emerald AI also uses the Conductor intelligent platform to dynamically align AI workloads with grid conditions, transforming rigid computational loads into flexible resources.

Together, they build an "energy storage + intelligent scheduling" system that effectively resolves the core conflict between surging electricity demand at AIDC and grid supply capacity.

02 Global Perspective: Overseas Compute-Power Struggles Reveal Core Contradictions and Practical Insights

As a leading frontier for the integration of AI computing power and electricity markets, the U.S. has seen regional grids such as PJM and ERCOT experience early conflicts between computing and power demands, becoming a global testing ground for key issues in computing-power coordination.

Storage costs have surged, increasing the burden on users.

In the PJM region, the concentration of AI data centers has caused capacity market prices to surge from $28.92 per megawatt-day in 2024–2025 to $269.92 per megawatt-day in 2025–2026, resulting in capacity bills totaling $16.1 billion—costs ultimately passed on to end users, sparking public and industrial controversy.

IMF research shows that, against the backdrop of lagging grid expansion, AI data centers could raise U.S. electricity prices by 8.6% and increase carbon emissions by 5.5%, placing dual pressure on electricity equity and the green transition.

Grid connection process is congested, with approval efficiency as a bottleneck

Power hubs such as Northern Virginia, USA, face AI data center grid connection wait times of 5 to 7 years, with severe backlogs in the grid access process.

ERCOT in Texas defines loads above 75 MW as "large loads." The total grid connection queue exceeds 10 GW by 2030, far surpassing the grid’s theoretical capacity. The individual project review method is inadequate for handling bulk applications, and procedural standards cannot resolve systemic bottlenecks.

Flexibility Controversy: The Clash Between Regulatory Fiction and Technical Feasibility

The Independent Market Monitor (IMM) considers the "load flexibility" of AI data centers to be a "regulatory fiction," as high-value training tasks are difficult to actively reduce during grid emergencies;

PJM has raised the ELCC rating for demand response resources to 92%, recognizing the regulatory potential of computing loads.

Real-world testing in Phoenix has demonstrated that, through software scheduling, data centers can achieve a 25%–40% load reduction without compromising core performance, debunking the myth of "illusory flexibility." However, the quantifiable and trustworthy nature of this "flexibility" remains uncertain.

Technical risks are becoming more apparent, putting pressure on grid stability.

AI data centers are highly electrified, with millisecond-level power fluctuations causing voltage flicker, frequency deviations, and even triggering subsynchronous oscillations.

ERCOT observed a 23 Hz load oscillation with a peak-to-peak amplitude of 50 MW, while Dominion's grid experienced a 14.7 Hz regional oscillation—both caused by coupling between data center UPS systems, server control logic, and grid dynamics. Traditional load models cannot accurately simulate these phenomena, posing a direct threat to grid physical security.

In summary, the overseas computing power and electricity competition reveals three core principles:

First, computing power is not merely a power load but a new type of flexible resource; it must shift from "passive power supply" to "active coordination."

Second, technical feasibility must be matched with well-designed mechanisms; the flexibility of computing power must be realized through market incentives, grid connection standards, and scheduling rules.

Third, coordinated computing and power management must balance efficiency and security, ensuring the development of computing power industries while maintaining the stability of the power grid.

This provides a practical reference for China in building a computing-power and electricity coordination system with distinctive Chinese characteristics.

03 Preparing for the Future: Core Issues and Development Strategies for China's Computing-Power Coordination

Faced with similar physical shocks and supply-demand imbalances, China is forging a unique path of development.

Governments at national and local levels, as well as enterprises, are aligning their consensus, starting from strategic coordination, institutional innovation, industrial collaboration, and business models.

Collaboratively resolve the conflict between the rapid development of computing power and the safe, efficient operation of the power grid, achieving sustainable computing power, green transformation of electricity, and shared value for all stakeholders.

Compared with overseas regions, China's coordinated development of computing power and electricity is fully leveraging its institutional advantages and resource characteristics, aiming to build a computing-power-friendly power ecosystem through end-to-end collaboration, full-scenario adaptation, and market-wide activation.

Strategic Coordination: Optimize the spatial and resource allocation of computing and power

Leveraging the national strategy of "East Data, West Computing," promote co-location planning of computing hubs and green power bases to enable local consumption of green electricity in the west and efficient support for computing demand in the east.

Establish a cross-regional computing and power scheduling system that aligns power supply and demand patterns by leveraging the spatiotemporal mobility of computing tasks.

Mechanism innovation: Integrating the entire process of grid connection, dispatching, and market operations

Some provinces are establishing dedicated grid connection channels for AI computing load, streamlining approval processes and clarifying rules for sharing the costs of grid upgrades;

At the power trading level, tech companies are integrating computing loads into virtual power plants for unified control, enabling coordinated scheduling between the power grid and computing resources.

Build a coupled market system for "computing power–electricity–carbon," enabling computing power to participate in electricity spot markets, ancillary services, and green power trading.

Facility Upgrade: Building a Resilient and Compatible Power Support System

Power grid construction is a traditional strength of China, and power grid companies are currently advancing the development of highly adaptable distribution architectures to meet the high-power-density power supply demands of computing infrastructure.

Relevant industry stakeholders are also promoting the integration of backup power, energy storage systems, and computing facilities to enhance grid regulation capabilities.

Multiple parties collaborate to build an integrated source-grid-load-storage base, enhancing the coordinated supply capacity of green electricity and computing power.

04 A Promising Future: The Synergy of Computing and Power Paves the Way for a New Era of Digital Energy and Reshapes the Industrial Landscape—from One-Way Supply to Two-Way Empowerment

In the future, the synergy between power and computing will completely transform the one-way model of “power supplying computing, computing consuming power,” creating a symbiotic ecosystem where power supports computing and computing reciprocates by enhancing the power grid.

The power system is no longer just a "support function" for computing power; it is the core competitiveness of the computing power industry;

Mining centers are no longer a burden on the power grid, but rather a core flexible resource of the new power system, with their deep integration forming a shared foundation for the digital economy and the energy revolution.

Technological paradigm shift: Full-stack collaboration has become an industry standard.

Chip design, computing power scheduling, power distribution architecture, grid control, and market trading will achieve full-stack collaboration, with compilers and schedulers becoming core tools of energy infrastructure.

Large-scale deployment of solid-state transformers, grid-forming energy storage, and integrated liquid-cooled power distribution equipment; AI algorithms reverse-optimize grid dispatching;

The renewable energy absorption rate and grid operational efficiency have significantly improved, achieving the optimal state where every kilowatt-hour of green electricity supports computing power, and every unit of computing power serves the grid.

Market ecosystem enhanced: Integrated market for computing, electricity, and carbon fully implemented.

A robust cross-regional computing power and electricity collaborative trading mechanism enables unified quantification and trading of green electricity consumption, carbon emissions, and flexibility value, with electricity traders and computing power schedulers working in tandem.

The integrated "Electricity-Computing-Carbon" service has become a core business for power companies, with computing demand in the east actively aligning with green power supply in the west, fully establishing a market-driven closed loop for the "East Data, West Computing" initiative.

Elevating Industry Value: Power Companies Become the Core Pillar of the Digital Economy

Relevant power companies are also undergoing a historic transition from "utility providers" to "digital economy infrastructure providers."

Deeply participate in the construction of the national computing power network through customized power supply, green energy services, energy storage solutions, and computing power scheduling.

Play an irreplaceable role in ensuring energy security, promoting the dual-carbon goals, and supporting the digital economy, helping China maintain a leading position in global AI competition and energy transition.

05 Conclusion

In the United States, Google and Tesla, along with multiple energy and technology companies, have jointly formed the Utilize Grid Utilization Alliance, with the core goals of improving U.S. grid utilization and reducing electricity costs to address the growing power strain caused by AI data centers and electrification.

The alliance leverages Tesla's energy storage, virtual power plant, and distributed energy technologies, combined with Google's intelligent scheduling and data center load management capabilities, to unlock idle capacity through demand response, load shifting, and grid optimization.

Drive the power grid’s transition from extensive expansion to efficient reuse, helping users and businesses reduce electricity costs while providing more stable and cost-effective power support for the digital economy and new energy development.

The core of the U.S.-China power competition lies in the battle for computing power, and behind this battle is the ultimate contest in energy innovation capabilities.

In China, the collaboration between computing power technology companies and energy utilities is timely and well-positioned; together, they will surely forge an innovative development path that surpasses Europe and the United States.

The evolution of computing power begins with AI, is realized through energy, and is perfected through collaboration.

The computing power revolution is reshaping the form, function, and value of power systems; coordinated computing and power is moving from concept to practice, from pilot projects to large-scale implementation, becoming an inevitable path in building a new-type power system.

The new era of bidirectional synergy between computing power and electricity has just begun…