■ Han Xue
The demand for AI (artificial intelligence) models continues to drive the construction of data centers, and the scale and computing power of data centers are growing rapidly, and the total scale of data centers in mainland China has ranked second in the world. In the future, with the acceleration of data center construction and the increase of computing power requirements, the power consumption of data centers will show a rapid growth trend, and the competition for computing power is evolving into the competition for energy resources.
With the rollout of large-scale models, it is expected that by 2030, the power consumption of data centers will reach 3%-4% of global electricity consumption. The China Academy of Information and Communications Technology estimates that in 2022, the electricity consumption of the digital industry in China will be 370 billion kWh, and the electricity consumption of data centers in use will be 76.6 billion kWh, accounting for 4.5% and 0.9% of the total electricity consumption of the whole society, respectively. It is estimated that by the end of the "14th Five-Year Plan", the electricity consumption of data centers will increase by 2-3 times compared with 2022, reaching a scale of 150 billion to 200 billion kWh, accounting for 1.5%-2% of the total electricity consumption of the whole society, and close to 400 billion kWh by 2030. Based on the average emission factor of the national power grid, carbon dioxide emissions from data centers currently account for about 0.4% of all greenhouse gas emissions, and it is expected to increase to about 1% of all greenhouse gas emissions by the end of the 14th Five-Year Plan.
Low-carbon operation data centers have become an inevitable choice to alleviate the huge energy consumption demand of AI. On the one hand, low-carbon operation of data centers is an inevitable choice for the sustainable development of the digital economy in mainland China. On the other hand, the decarbonization of data production is also an important aspect to ensure the digital competitiveness of the mainland. The core of low-carbon operation of data centers lies in reducing energy consumption, access to green electricity, and load regulation.
1
Data centers are actively improving through energy efficiency
Reduce carbon emission intensity and rapidly improve energy efficiency
The carbon emissions of the whole life cycle of data centers are mainly in the operation stage, and the carbon emissions in the construction and operation stages account for about 10% and 90% respectively. For a long time, the carbon reduction of data centers has mainly improved energy efficiency and reduced carbon emissions by improving cooling methods, optimizing rack design, and reducing chip energy consumption. The energy consumption of the cooling system of the data system is rapidly decreasing, and the energy consumption of IT equipment will become the core of energy efficiency management. On the basis of the existing mature technology, the space for reducing the energy consumption of IT equipment can be further explored by using low-power chips, all-flash storage, all-fiber network, and all-lossless Ethernet.
Compared with the improvement of energy efficiency, the low-carbon energy supply structure has a more significant effect on reducing the carbon emissions of data centers. And with the maturity of new energy and energy storage technology, green power is gradually gaining economic advantages.
From the perspective of regional layout, the layout of the data center to the west and north is more in line with the distribution of wind and solar resources, and the power load brought by the data center is also conducive to improving the local power load level in Northwest China and improving the stability of the local power system, thereby improving the power grid's access to new energy.
From the perspective of operational economics, data centers can reduce operating costs by purchasing green electricity. According to the survey of typical projects, electricity expenditure accounts for about 60%-70% of the operating cost of data centers, and the purchase of green electricity with the support of large power grids and supporting small-scale energy storage, heat energy cascade utilization and heat pumps can effectively reduce the operating costs of data centers.
2
Mining the elasticity of the power load of the data center itself
is the key to achieving zero-carbon operations
The volatility of renewable energy generation and the variation in the power load of data centers make it difficult for data centers to be powered by renewable electricity alone. If the data center cannot provide the adjustment capacity, the increased power consumption of the data center will need more fossil energy power generation to meet the demand, and the greater carbon emission reduction effect cannot be achieved. Therefore, it is necessary to further optimize the operation mode of data centers, and better adapt to and match the characteristics of renewable energy generation by releasing the elasticity of electricity load and configuring energy storage facilities.
The elasticity of the power load can be achieved by adjusting the layout of work task time or space in the case of uneven occupation of resources by various components of the data center. At the component level, the power consumption of the server in the no-load state can be optimized by setting a power cap, hibernation or shutdown, etc., but different methods will also affect the speed at which the server can return to normal working conditions. At the business level, optimizing business processes and distribution, and reducing component redundancy can effectively reduce the power load. At the task level, latency-tolerant tasks can be used to shift and expand computing power requirements with the help of breakpoint resuming and adjusting parallel computing nodes. Latency-sensitive tasks can rely on high-speed networks to achieve optimal redistribution of tasks across multiple data centers.
Image source/Xinhua News Agency
3
Difficulties and challenges in promoting low-carbon operation of data centers
There is still a large mismatch between the regional layout demand of data centers and renewable energy resources. Although the strategy of "Eastern Data and Western Computing" has strengthened the layout of data infrastructure at key nodes in the western region, the gap in the number of racks between the east and the west is still relatively large. The construction of data centers in the western region also faces obstacles such as insufficient technical personnel and poor operation and maintenance level. The difference in the distribution of data centers and renewable energy resources makes it more difficult to improve the overall low-carbon level of data center operations, and data centers in the central and eastern regions are likely to face a lack of local green power.
The imperfect green power trading market system and certification mechanism also make it more difficult for data centers to obtain low-carbon electricity. Although the mainland has recently significantly improved the policy mechanism related to green electricity and green certificate trading, there are still some problems. First, the current inter-provincial and inter-regional power trading is still dominated by intergovernmental contracts, and inter-provincial and inter-regional trading of green electricity is still relatively difficult. Second, the current relationship between green power and green certificates, the carbon market and carbon accounting is still unclear, and the carbon reduction benefits of green power and green certificates need to be further clarified. Third, the traceability of green electricity and green certificates is poor, and there is still a lot of room for improvement in market information transparency, green certificate verification, and third-party certification.
For data centers with green power sources, there are multiple obstacles such as technology, economy, awareness, and mechanism to improve the consumption level of green electricity and achieve a high proportion of green power supply. In the absence of a clear time-of-use price signal, the addition of additional energy storage devices will greatly increase the electricity cost of the data center, and it is difficult to promote. In addition, data center operations have a variety of business models, which may involve multiple entities, making it difficult to coordinate computing loads. In addition, there are great differences in different computing tasks, and the occupation of IT equipment is also different, which cannot be used as a standard to measure the power load, which makes it difficult to calculate carbon emissions.
4
Policy recommendations
Accelerate the research and development of relevant technologies and systems to provide a technical foundation for low-carbon operation of data centers. Optimize the design from the aspects of devices, racks, systems, and tasks to further reduce the operating energy consumption per unit of computing power in the data center. Strengthen the research and development of high-efficiency cooling technology and supporting materials and equipment, waste heat recovery and utilization, multi-time scale prediction of renewable energy power generation, data processing task identification and packaging technology, and system integration of multi-field technologies. Speed up the education and training of relevant personnel in the western region.
Promote the construction of the electricity market to provide reliable sources of green electricity and effective price incentives for data centers. To further improve the green electricity and green certificate market, on the one hand, it is necessary to strengthen market information disclosure and certificate life cycle management, and improve the traceability of green attributes; On the other hand, it is necessary to innovate trading varieties, optimize the market access mechanism, and improve the availability of green electricity and green certificates in various places. Improve the flexibility of inter-provincial and inter-regional electricity trading, promote inter-provincial and inter-regional trading of renewable energy, strengthen the tracking and disclosure of electricity carbon information by power grid companies, and try to establish standard rules for zero-carbon electricity accounting. Construct a multi-subject renewable energy power generation forecasting system of "power grid-region-power plant", accelerate the establishment and improvement of the electricity spot market, improve the supporting electricity prices for the retail market and agency power purchase, and provide price guidance for the investment in power supply and demand regulation capacity of data centers.
Improve incentive mechanisms such as green data centers, and guide data centers to accelerate the pace of carbon reduction. Increase the weight of indicators such as PUE value, waste heat utilization rate, and renewable energy use in the green data center index system, and guide the westward layout of data centers with low latency requirements. In the data infrastructure planning and renewable energy development planning of governments at all levels, the supply of green electricity and the electricity demand of data centers should be considered as a whole. In the approval process for the construction of large and medium-sized data centers, the requirements for PUE value, waste heat utilization methods, and supporting renewable energy project construction will be strengthened. Guide the industry to build low-carbon data service specifications, and promote data center owners, builders, operators, and service providers to form a consensus and reach a consensus in a variety of business forms.
(The author is deputy director of the Research Office of the Institute of Resources and Environmental Policy, Development Research Center of the State Council)
Chief Producer丨Wang Hui and Che Haigang
Producer丨Li Piguang, Wang Yu, Liu Weimin
Editor-in-Chief丨Mao Jinghui Editor丨Gu Yun