Goldman Sachs research predicts that by 2030, data center power demand will surge by 160%, primarily driven by the rapid development of artificial intelligence. A simple query using ChatGPT consumes up to 2.9 watt-hours of power, far exceeding the 0.3 watt-hours used for a typical Google search, highlighting the immense computational demand of AI. This growth will pose challenges to the power market, leading to increased electricity costs, particularly affecting residents near data center areas. By 2028, it is expected that artificial intelligence will account for a significant portion of the total power demand in data centers.

With the rapid development of artificial intelligence (AI), the increase in electricity demand has attracted a massive influx of investors into utility stocks. In May and June of this year, U.S. utility funds attracted over $1.7 billion in inflows, marking the highest record in nearly two years. An additional $1.1 billion inflow is expected in July, primarily into the Utilities Select Sector SPDR (XLU) exchange-traded fund. Investors are turning their attention to utility stocks in search of AI investment opportunities that can outperform 'Big Tech' stocks in a strong bull market.

["A recent report by the International Energy Agency highlights that by 2026, the contributions of data centers, artificial intelligence, and cryptocurrencies to global power demand may double, causing a significant impact on global energy reserves.", "The IEA estimates that these three sectors consumed approximately 460 terawatt-hours (TWh) of electricity globally in 2022, accounting for 2% of total global demand.", "The report also notes that the widespread adoption of artificial intelligence in large existing applications could significantly increase electricity consumption."]

1. A 100 billion parameter model requires 266 8-card A100 servers with a single card computing efficiency of 44%. 2. To improve the performance of large models, it is necessary to optimize aspects such as frameworks, IO, and communication. 3. Compared to GPT-4, domestic large models have discrepancies in computing power, algorithms, and data.