By Jim Shimabukuro (assisted by Gemini)
Editor
Here are some of the most effective and promising innovations for reducing the environmental footprint of AI data centers.
Currently Applied Innovations
- Liquid Cooling: As AI workloads become more power-dense, traditional air cooling is becoming less efficient. Liquid cooling, such as direct-to-chip or immersion cooling, can more effectively dissipate heat directly from the source, reducing the energy needed for cooling systems. This allows for higher server densities, better performance, and significant energy savings. Companies: Asetek, Asperitas, CoolIT Systems, Green Revolution Cooling, LiquidStack, Submer, Vertiv, and Schneider Electric. Microsoft and Google are notable end-users of this technology.
- Renewable Energy Sources: A fundamental way to reduce the carbon footprint of data centers is to power them with clean energy. Many major tech companies are investing in and procuring renewable energy sources like solar, wind, and geothermal to match or even exceed their energy consumption. Some data centers are being built near renewable energy hubs to ensure a consistent clean power supply. Companies: Google, Digital Realty, Equinix, Iron Mountain, and Switch are all major players in powering their data centers with 100% renewable energy.
- Waste Heat Reuse: Data centers produce an enormous amount of waste heat. Instead of simply venting this heat into the atmosphere, companies are now capturing and reusing it. This is often accomplished by using the heat to warm nearby buildings through district heating systems, or for industrial purposes like heating greenhouses or fish farms. This innovation turns a waste product into a valuable resource, improving overall energy efficiency. Companies and Projects: Microsoft, Google, and Amazon are involved in projects that reuse waste heat from their data centers to warm buildings in nearby communities. Notable projects include Stockholm Data Parks in Sweden and Microsoft’s data center region near Helsinki, Finland
- AI for Energy Efficiency: AI itself is being used to make data centers more sustainable. Machine learning algorithms can analyze real-time data from a data center’s infrastructure to optimize energy usage. This includes controlling cooling systems, managing power distribution, and dynamically adjusting workloads to minimize energy consumption. Companies: Google is a pioneer in using AI to optimize its data center cooling systems. Crusoe is another company that uses AI to optimize its data center infrastructure while also using renewable energy sources.
- Efficient Model Architectures & Software Optimization: The software and models themselves are being designed for greater efficiency. This includes using optimized model architectures like Mixture-of-Experts (MoE), which only activate a small part of a large model to respond to a query, and techniques like quantization to reduce the computational and energy cost of running AI models without compromising performance. Companies and Individuals: Google (Geoffrey Hinton), Databricks (DBRX), OpenAI, and Anthropic (Claude 3 family) are developing more efficient AI models and architectures like the Mixture of Experts (MoE) to reduce computational power and energy consumption.
Promising Innovations Under Development
- Hydrogen Fuel Cells: Hydrogen fuel cells offer a promising clean energy solution for data centers. They convert hydrogen gas into electricity with only water as a byproduct, producing zero greenhouse gas emissions during operation. Fuel cells can provide a reliable, on-site power source, reducing reliance on the grid and offering a cleaner alternative to traditional backup diesel generators. Timeline: Within 5 years (Late 2020s): This technology is already being implemented in pilot projects. The primary focus now is on scaling up and reducing costs. Several companies are already providing fuel cell systems for data center power, and the technology is seen as a viable solution for both primary power and backup generation, especially as reliance on traditional diesel generators is phased out. Companies: AFC Energy, Ballard Power Systems, and Bloom Energy are leading companies in the development and manufacturing of hydrogen fuel cell technology for on-site power generation, including for data centers.
- Advanced Nuclear and Fusion Energy: While still in development, small modular reactors (SMRs) and nuclear fusion technology could one day provide a constant, carbon-free source of immense power. This could be a game-changer for large-scale AI data centers, which require a reliable and abundant energy supply to operate. Timeline: 2030 and beyond. Small Modular Reactors (SMRs): Widespread commercial application for SMRs is projected for the early 2030s. The timeline for data center integration will depend on licensing and construction, with the first pilot projects likely to be in the late 2020s. Fusion Energy: True commercial application of fusion energy is a much longer-term prospect, likely in the 2040s and beyond. While significant progress has been made, the technology is still in the research and development phase and is not expected to be a viable power source for data centers in the near term. Companies and Projects: Microsoft and Google are investing in this area. A key partnership includes Google and Kairos Power, which are working to bring advanced nuclear energy to the U.S. electricity grid to power data centers. Other companies in this space include TAE Technologies, Helion Energy, and Oklo.
- Ocean-Based Data Centers: Submerging data centers in the ocean offers a natural and cost-effective cooling solution. The cold seawater can be used to cool servers, reducing or eliminating the need for energy-intensive mechanical cooling systems. This innovation also allows for data centers to be co-located with offshore renewable energy sources like wind or tidal power, further enhancing their sustainability. Timeline: Within 5-10 years (Late 2020s to early 2030s): While Microsoft’s Project Natick was a research initiative, the successful results demonstrated the viability of the concept. Several companies are now pursuing commercial floating data centers in ports and other controlled environments. Commercial deployment of submerged data centers will require further development in subsea engineering, but the foundational technology and its benefits are proven. Companies and Projects: Microsoft’s Project Natick and Nautilus Data Technologies are notable players in this space. Project Natick was a research initiative for undersea data centers, while Nautilus focuses on floating data centers on barges.
- Modular and Prefabricated Data Centers: This approach involves building data centers in a controlled factory environment in pre-fabricated modules. This not only speeds up construction but also allows for the use of more sustainable materials, such as green concrete and renewable diesel. This reduces the environmental impact associated with traditional, on-site construction. Timeline: Already in application and growing rapidly. This innovation is not a future projection but rather an existing and rapidly growing segment of the data center industry. The market for modular data centers is already valued in the tens of billions of dollars and is projected to see significant growth from now through the late 2020s, driven by the need for faster deployment and scalable solutions to support edge computing and AI. Companies: Dell Technologies, Hewlett Packard Enterprise (HPE), Schneider Electric, and Vertiv are some of the top manufacturers providing prefabricated and modular data center solutions.
- Photovoltaic Cooling Technology: Research is being conducted on new ways to use solar power for more than just electricity generation. For example, some technologies are being developed that can use solar thermal energy to power cooling systems directly, potentially offering a sustainable, integrated solution for both power and cooling. Timeline: Within 10 years (Early to Mid-2030s). While photovoltaic (PV) technology for electricity generation is mature, the direct use of PV for cooling is an area of ongoing research. It will require advancements in materials and system design to integrate it efficiently into data center infrastructure. The widespread application will likely follow the successful development of scalable and cost-effective prototypes. Companies: While a search for “photovoltaic cooling technology for data centers” returned results for general data center cooling companies, it did not identify specific companies or individuals leading a distinct “photovoltaic cooling” innovation. The companies listed in the search results, such as Vertiv and Schneider Electric, are broad cooling solution providers. It’s possible that this technology is a nascent field or integrated into broader cooling solutions.
Filed under: Data Center |
Educational Technology and Change Journal 
Leave a comment