Produced by Zhineng Zhixin

With the rapid proliferation of artificial intelligence (AI), data centers have mushroomed, placing unprecedented demands on the grid. The frequent occurrence of harmonic distortion highlights the severe impact on power quality, even threatening the operational safety of household appliances and aging power equipment.

Drawing insights from Bloomberg's article "AI Needs So Much Power, It’s Making Yours Worse," we delve into two pivotal aspects:

Part 1 focuses on the power quality challenges posed by AI data centers and their respective solutions.

Part 2 examines NVIDIA's strategy to standardize lithium battery BBUs (Battery Backup Units) in the latest generation of servers.

Part 1

Challenges and Solutions to Power Quality in AI Data Centers

The AI boom has propelled the United States to become the world's largest data center operator, with other countries racing to follow suit. Data centers, as significant urban energy consumers, are expanding faster than grid planning can accommodate.

In the U.S., for instance, power demand is projected to surge by nearly 16% over the next five years due to new data centers, more than triple last year's estimate. The grid's power usage has remained relatively stable for decades, and the existing infrastructure struggles to adapt to this explosive growth, leading to tight power supply and voltage stability challenges.

Data centers, typically located near major cities for better grid and fiber-optic access, strain already vulnerable urban grids. Rural areas also experience power distortion near significant data center activities, highlighting the grid's pressure in both urban and rural settings, exacerbated by aging infrastructure, extreme weather, and the rise of electric vehicles.

Data centers' large-scale power consumption and instability (AI energy consumption resembles a sawtooth pattern) can cause grid voltage fluctuations. During peak usage, grid voltage may drop, while off-peak times may see voltage spikes, affecting electrical devices' normal operation, reducing their lifespan, or even causing damage. Voltage-sensitive equipment like precision instruments and medical devices are particularly susceptible to measurement errors, malfunctions, or even scrapping due to instability.

● Direct consequences of harmonic distortion include:

◎ Equipment Damage: Electrical heating, increased noise, or malfunctions, drastically reducing equipment lifespan.

◎ Fire Risk: Voltage fluctuations or harmonic accumulation can spark electrical fires.

◎ Economic Losses: Harmonic-related global losses could reach billions of dollars.

The rapid deployment of data centers outpaces traditional grid planning, imposing unprecedented strain on aging grid infrastructure. Additionally, trends like electric vehicle popularization and increased household electrification further exacerbate power demand, making the grid more fragile and less capable of handling ultra-large loads.

The impact of AI data centers on power quality presents both a technical challenge and an opportunity for grid optimization.

Part 2

Effectiveness and Implications of NVIDIA's Power Quality Solution

NVIDIA aims to standardize BBUs (Battery Backup Units) in its new GB3800 servers, favoring lithium-battery-based BBUs for their competitiveness.

Compared to traditional UPS and diesel generators, BBUs offer faster response times (millisecond level), smaller sizes, and flexible layouts, effectively bridging the gap between supercapacitor startup and diesel generator power supply.

In high-voltage DC power distribution systems adopted by new AI data centers, high-rate BBUs exhibit superior power conversion efficiency, optimizing data center operating costs. Lithium-battery BBUs' 5-10 year lifespan and fast-charging capabilities significantly reduce lifecycle costs.

While China's grid environment is relatively stable, and BBU demand is less urgent than overseas, AI's heightened data security requirements, coupled with BBUs' cost and space advantages over UPS, are gradually steering the domestic market towards BBU adoption. Some predict that with technological advancements, BBUs could become the standard for emergency power supplies.

● Solution Advantages: Lithium-battery BBUs provide fast response times, small sizes, and flexible layouts, enhancing power conversion efficiency in high-voltage DC systems. Their long lifespan and fast charging reduce lifecycle costs and enhance data security.

● Problem Solved: Enhances power supply reliability, preventing data loss or service interruption during outages. Optimizes operating costs by reducing power loss through efficient conversion, lowering electricity bills.

● Design Impact: Transforms power system architecture, integrating supercapacitors and diesel generators for a comprehensive emergency power supply. Alters space and thermal designs for more compact, efficient data centers. Enables intelligent management for real-time monitoring and predictive maintenance.

NVIDIA's harmonic optimization solution introduces AI algorithms for real-time power usage monitoring and optimization. It automatically adjusts data center component power consumption, reducing unnecessary energy use and enhancing overall efficiency, lowering operating costs and environmental impact.

To ease main grid pressure, NVIDIA suggests shifting computational tasks to edge nodes closer to data sources or users, reducing transmission latency, minimizing bandwidth usage, and distributing instantaneous power demand to prevent grid overload.

This distributed power supply strategy promotes localized energy production, such as solar panels and other renewables, enhancing system self-sufficiency and environmental performance.

● Impact on Future Data Center Design:

With technological advancements, future data centers will be more flexible, facilitating rapid service deployment and technological upgrades.

Intelligent management systems enable self-regulation and optimization with minimal human intervention. Amidst growing environmental concerns, data centers increasingly adopt renewable energy like solar and wind, contributing to carbon neutrality goals.

The cloud-edge collaboration concept optimizes computing resource allocation, while regional node deployment ensures service continuity during network failures, enhancing system resilience and user experience.

Comprehensive hardware-to-software optimization improves both physical infrastructure and algorithm efficiency. New filtering equipment combined with enhanced energy efficiency monitoring and optimization algorithms ensure optimal data center performance.

NVIDIA's practices underscore the need for smarter, more resilient grid management. Leveraging AI, electrical engineering, and data analysis can better predict load demand, dispatch energy resources, and ultimately improve power quality, promoting sustainable development. This not only revolutionizes grid management but also heralds a cleaner, more efficient, and reliable power future.

Summary

The rise of AI data centers has placed unprecedented grid pressure, driving technological upgrades in power quality. Zhineng Zhixin believes that diverse technical approaches can enhance power quality, and cross-industry collaboration can facilitate intelligent grid transformation.