Yakov Danilevskiy, Vice President Strategy, Channels and Marketing, Schneider Electric, shares insight into why the rapid emergence of AI workloads has proved a timely lesson for the broader consideration of future-proofing data centres for the needs of tomorrow.
Digital infrastructure is accelerating due to demand, the impact of new technologies and the widespread adoption of Artificial Intelligence (AI) and Machine Learning (ML).
Estimates from IDC indicate that in 2023, enterprises worldwide spent US$166 billion on AI solutions, covering software, hardware and services. However, the analyst has estimated growth of 27% per year to US$423 billion by 2027.
Data centre design used to be about providing a scalable capacity to meet demand that was cost-effective, resilient and secure. Now, AI demand is having a specific impact on rack design, power density and cooling, according to our research.
Today, there are many more elements to consider encompassing speed, scalability, resilience and reliability, and the specific needs of AI and other future compute-intensive workloads.
Strategic considerations
A data centre strategy must incorporate minimising operational costs, carbon footprint and overall environmental impact, even as it provides the essential facilities of reliability and resilience.
Energy security and resilience means greater use of renewable energy sources (RES) from the outset is critical, to include on-site generation. Wind, Solar photovoltaic (PV) and hydro are all being seen in various geographies, with other facilities such as Battery Energy Storage Systems (BESS), also being deployed. These contribute to greater resilience, allowing operators to reduce grid reliance where needed.
The impact of other resource use, such as water, is also of prime consideration.
The triple considerations of power (PUE), water (WUE) and carbon emissions must figure highly, while also thinking about how individual facilities will integrate into the local environment from an economic and social perspective, such as in district heating systems.
And, when all of these considerations are met, future proofing must be considered.
Modular architectures – created with digital design tools, leveraging best practices in design and the most efficient hardware from the outset – give designers the tools to create scalable facilities that embody all these principles. Initial design models can evolve into Digital Twins that can serve as useful sandbox environments. Here, new configurations, permutations and developments can be tested for impact and effectiveness before being implemented in a live environment.
Sustainability and energy efficiency
Despite rising electricity consumption in data centres attracting much media coverage, there is still strong evidence to suggest that relentless innovation has kept efficiency at a very high level. Data centre capacity globally has grown at around 17% annually in recent years, but energy consumption represents less than 2% of total global energy demand.
Maintaining that record of capacity and efficiency, data centre operators can scale with the right strategy, while prioritising sustainability and business goals.
To help with this, Schneider Electric was a signatory of the recent iMasons open letter advocating for the widespread adoption and implementation of Environmental Product Declarations (EPD) across the industry. This allows designs and operators to know which items of equipment are the most efficient and low carbon intensity.
Increasingly, data centre operators and energy grid operators are entering into partnerships. The stable and continuous need for power from data centres allows them to sign long-term power contracts that can underpin the development of more renewable power. Power Purchase Agreements (PPA) give certainty to the RES market allowing more capacity to be brought safely online.
Future workloads driving density
Even as new workloads emerge, such as the growing adoption of AI, others such as the likes of blockchain-based supply chains and next-gen Fintech, are likely to continue density trends associated with compute-intensive workloads, with similar implications for space, power and cooling.
According to Schneider Electric research, we estimate that in 2023, AI consumption was of the order of 4.5GW globally, but that will grow to between 14-18.7GW by 2028. AI power consumption was 8% of total data centre power in 2023 but will grow to represent between 15-20% in the same period.
AI workloads were 95% centrally deployed and 5% Edge in 2023, by 2028, we expect this to be on the order of 50% central and 50% Edge deployed. All of these developments will impact how data centres are designed, deployed and operated.
Power supplies to racks and rows, as well as liquid cooling, are significant considerations for the transformation of existing facilities, and for the design of new ones. We have conducted extensive research in this area and produced a white paper with our findings that provides guidance on the four key AI attributes and trends that underpin physical infrastructure challenges in power, cooling, racks and software management.
The paper gives recommendations for assessing and supporting the extreme rack power densities of AI training servers, and guidance for achieving a successful transition from air cooling to liquid cooling to support the growing thermal design power (TDP) of AI workloads.
Complexity risks
Gartner projects that by 2025, 85% of infrastructure strategies will integrate on-premises, colocation, cloud and Edge delivery options. Added to this, AI at the Edge has some specific benefits regarding latency and bandwidth.
Simplifying operations will be a necessity, requiring a paradigm shift toward agility, elasticity and cognitive capabilities, which entails rethinking architectures, particularly network architectures. These must enable seamless scale-out while unlocking deeper correlations and insights into system performance and behaviour. In managing the complexity, there are great benefits for optimisation from the insights available, especially for Edge deployments.
Partnerships and alliances
In this world of many permutations and mixed architectures, the reality is that no one vendor or service provider will be able to do everything in this fast-moving and highly competitive market. Only through strong, strategic partnerships and alliances can a technology partner provide the comprehensive value offering that will be needed by enterprises tomorrow.
These partnerships and alliances, as the base of a strong ecosystem, will be what allows businesses to thrive in the highly competitive marketplace, win customers, and engender loyalty.
Our partnerships with market leaders such as NVIDIA and NTT Data characterise our collegiate approach and have allowed us to create a comprehensive industry portfolio, that is backed up by long-standing expertise, broad research and development and cutting-edge implementations.
Best practice
Many lessons have been learned from the current iteration of technology developments, such as AI and the impact on data centre design and operations.
- Future-proof design – It is clear that new approaches must be taken to account for rack and row configurations, power density, cooling measures and overall resource consumption, in catering for new and emerging workloads. All of this must also be achieved while keeping within sustainability commitments and delivering the core capabilities of resilience, availability and security.
- Agility – Today’s design and management tools enable us to establish the state-of-the-art for data centre design and operations, while providing a foundation for tomorrow that can evolve for developing needs, while mitigating downtime, improving security and providing more granular information for sustainability efforts.
- Resources and partnerships – Best practice, combined with experience and insights, allows us to provide a strong base for informed partnerships with customers. Extensive research and publishing allow us to put the right information and resources in the hands of business and technology leaders to build Digital Transformation strategies today that will drive competitive advantage in the future.
Our digitally connected world demands a dynamic approach to building a future-proof strategy that must encompass an international perspective tempered with regional knowledge.
