Title: Unleashing AI’s Power: Fueling Tomorrow’s Intelligence, One Electron at a Time”
—Unveiling AI’s Energy Appetite: The Electrifying Rise of AI and Its Energizing Impact on Data Centers

3 August 2023, Monica Cai monica.cai@crm.szc-group.com

—1. Introduction—
In recent years, the rapid advancement and widespread application of artificial intelligence (AI) have brought about convenience and transformation in various aspects of our lives. However, as AI technology continues to break new ground and expand its horizons, it is also driving an increasing demand for electricity resources. Astonishingly, the latest reports reveal that an AI server can consume as much energy in a single day as an entire year’s worth of electricity for a small town resident! Even more startling, this energy consumption rate is growing at a staggering pace of 10% annually, leading some to humorously comment that “AI servers aren’t burning chips, they’re
burning electric meters.”

—2. The High Energy Consumption of AI Servers—
The energy consumption of AI servers has long been a focal point of concern within the industry. To facilitate rapid calculations, large-scale data processing, and complex algorithm optimization, AI servers require a substantial supply of electricity. However, recent studies have shown that the power consumption of AI servers has far exceeded expectations and imaginations.

On July 15th, according to the latest report by Digital Information World, the energy consumption generated by data centers for training AI models will be three times that of regular cloud workloads. By the year 2030, the power demand of US data centers is projected to grow at an annual rate of approximately 10%.

Data reveals that training OpenAI’s GPT-3 model consumes 1.287 gigawatt-hours, roughly equivalent to the annual electricity consumption of 120 American households. Yet, this is merely the preliminary power required for model training, accounting for only 40% of the electricity consumed during actual model usage.

In January 2023, OpenAI’s energy consumption in just a month equated to what about 175,000 Danish households would use in an entire year. Google’s AI operations consume an annual 2.3 terawatt-hours, equivalent to the electricity used by all households in Atlanta for a year.

– 3. In the race for power supremacy —
The staggering energy demand is poised to exert even greater pressure on digital infrastructure, pushing existing data centers to their limits as they struggle to keep up with the burgeoning power needs of AI.

Last year, a data center in Northern Virginia faced a moment of crisis, on the brink of a power outage, highlighting the imminent energy challenge.

Addressing the soaring energy consumption of AI, DigitalBridge, a digital infrastructure company, is gearing up to invest billions in constructing and revitalizing data centers tailored to handle the surging workload of generative AI.

Global hyperscale data center capacity 2022 – Statista

Hence, while the potential of artificial intelligence remains boundless, the realms of physics and economics may eventually draw limits. Leading technology companies are tirelessly exploring every conceivable strategy to secure their power needs, laying the foundation for a transformative energy future.

—4. The Energy Composition of AI Servers—
AI servers demand significantly more power due to their specialized functions and capabilities. They are designed to support and execute AI algorithms and models, boasting high-performance processing, substantial memory, and rapid storage to meet complex computing needs. Their primary tasks include large-scale data processing, model training, and inference calculations. AI servers can handle massive data sets, train intricate deep learning models, and efficiently perform real-time inference and predictions.

To support these tasks, AI servers typically employ multi-core processors (such as CPUs or GPUs) for parallel data processing. GPUs, in particular, are more efficient in training and inference processes as they can handle multiple parallel tasks. In addition to processors, AI servers require ample memory to store model parameters and temporary computation results, as well as high-speed storage to accelerate data read-write speeds. Distributed computing is also supported in some AI servers for better scalability and flexibility. Multiple servers can form a cluster, collaborating on training and inference tasks to achieve higher throughput and faster response times.

Furthermore, AI servers need to maintain stable temperatures and efficient heat dissipation for secure operations. This requires additional energy for fans, cooling systems, and monitoring equipment. AI servers also feature power management systems, uninterruptible power supplies (UPS), and more to ensure consistent and reliable operations.

—5. Conclusion —
In conclusion, the rapid evolution of AI technology has brought remarkable innovations and capabilities, but it comes at the cost of substantial energy consumption. Balancing the potential of AI with the growing demands on our energy resources presents a challenge for the future. As AI continues to redefine industries and drive innovation, it is imperative that we explore sustainable solutions and collaborate across sectors to create a harmonious future where AI and energy coexist seamlessly. By striking the right balance, we can unleash the full potential of AI while ensuring a greener and more energy-efficient world.

—Resources—
1. The Energy Crunch: AI Data Centers and the Battle for Power: [https://www.digitalinformationworld.com/2023/07/the-energy-crunch-ai-data-centers-and.html]
2. ‘3 raccoons in a trench coat’: AI boom could actually warm our planet: [https://interestingengineering.com/culture/ai-boom-cause-global-warming]
3. Energy implications of generative AI – A perspective on ChatGPT:
[https://energy.economictimes.indiatimes.com/news/power/energy-implications-of-generative-ai-a-perspective-on-chatgpt/99281787?redirect=1]
4. Share of worldwide hyperscale data center capacity in 2nd quarter 2022, by region https://www.statista.com/statistics/1350992/global-hyperscale-data-center-capacity/
—Literature Review —-
1. A systematic review of Green AI: [https://wires.onlinelibrary.wiley.com/doi/full/10.1002/widm.1507]

2. Trends in AI inference energy consumption: Beyond the performance-vs-parameter laws of deep learning: [https://www.sciencedirect.com/science/article/pii/S2210537923000124]

3. Research and Technologies for next-generation high-temperature data centers – State-of-the-arts and future perspectives: [https://www.sciencedirect.com/science/article/pii/S1364032122008723]