Data dilemma: Microsoft wants to have its sustainability cake and eat it

Ines Lin, Taipei; Samuel Howarth, DIGITIMES Asia 0

Credit: AFP

Following the announcement of several data center expansion projects, Microsoft reiterated its commitment to sustainability.

Microsoft said the deployment of Artificial Intelligence (AI) applications is accelerating and resource intensity is increasing environmental impact. Given the company's recent emphasis on Small Language Models (SLM) and energy efficiency, energy management appears to be the biggest challenge.

As OpenAI continues to advance Large Language Model (LLM) research, Microsoft, one of the company's major investors disclosed multiple data center expansion projects. The expansion encompasses projects in the UK, Germany, and Spain.

Recent media reports suggest that Microsoft plans to invest billions of dollars in establishing a new AI data center. It is slated to be operational by 2028 and will include a supercomputer for OpenAI's use.

Compared to OpenAI's rumored US$7 trillion fundraising drive for multiple chip factories, Microsoft's plans seem more conservative. However, estimates suggest Microsoft's costs could end up one hundred times greater than those of existing major data centers.

Microsoft has made several sustainability commitments. Some of its pledges include achieving negative carbon emissions and positive water impact by 2030 and removing all carbon emitted since its founding by 2050.

However, with continuous data center expansion, it remains to be seen how the company will achieve these goals. A further unknown is what will happen if it fails to meet them.

Working on it

One of Microsoft's strategies for achieving sustainable AI is optimizing data center resource efficiency. Microsoft identifies energy management and water resource consumption as the two major challenges in data center operations.

The company plans to expand the use of renewable energy, reduce peak electricity usage, and collect unused energy, improving server utilization efficiency. Microsoft will employ intelligent management, consider energy consumption when deploying virtual machines, and aim to enhance the efficiency of chips and code.

In terms of renewable energy, Microsoft currently has over 135 renewable energy procurement projects globally. The company aims to achieve 100% zero-carbon electricity as it designs and constructs data centers.

Irish innovation

During local peak electricity demand periods, the company aims to feed data center power back into the local grid. In Ireland, teams install energy storage batteries in wind turbines to feed excess electricity back into the local grid while in Denmark operate projects for heat recovery and community feedback.

The primary use of data centers is for cooling, followed by indirect electricity generation. Measures to reduce water consumption and increase recycling and reuse include using air instead of water for data center cooling, collecting rainwater, and reusing other reusable water sources.

Microsoft is also focusing on the use of low-carbon building materials. The company said that among data center building materials, steel, and cement contribute the most to carbon emissions, accounting for approximately 13.5% of global carbon emissions.

Green steel and low-carbon cement are in development. For now, though, market demand is still in its infancy.

Microsoft has procured solutions with Swedish company H2 Green Steel that are estimated to reduce carbon emissions by 95% compared to traditional steel production. There are experimental projects using alternative materials such as bio-limestone, fly ash, and slag to reduce cement usage.

The company hopes to improve the energy efficiency of AI and cloud services. To reduce energy consumption for AI computation, Microsoft emphasizes "green software engineering" and ensures cloud users receive necessary information without wasteful cloud configurations.

Microsoft recently introduced the Small Language Model (SLM) Phi series, claiming performance comparable to or surpassing larger models, including a model 25 times larger. Other AI models are also used for accelerating battery material research.

IEA warnings

The International Energy Agency (IEA) report said that in 2022, global data centers, cryptocurrencies, and AI-related demands collectively consumed 460 TWh of electricity, which is estimated to increase to 620-1,050 TWh by 2026, depending on improvements in energy efficiency. Currently, computation and cooling equipment each account for 40% of data center electricity usage.

IEA data shows that renewable energy generation accounts for 41.6% of global energy use. Renewable energy is estimated to surpass coal as the primary global electricity source by 2025.

Though revealing, the IEA's figures are averages, and renewable energy development varies by country. In Taiwan, for example, Taipower's data for 2023 includes 44.1% natural gas, 34.1% coal, 9.9% renewable energy, 7% nuclear, 2.4% cogeneration, and 1.2% fuel oil and pumped storage hydropower.

In addition to walking the tightrope of technological development and environmental protection, major cloud providers must evaluate and address the issue of surplus energy expansion.