Omdia view
Summary
The recent event DCD Connect London was focused on helping future-proof data centers. The atmosphere was optimistic, and most talks were at full capacity. A discussion about how data centers can adapt their infrastructure for the incoming artificial intelligence (AI) workloads stood out. The speaker analyzed the upcoming challenges of managing computationally intensive AI workloads and proposed solutions to the infrastructure-based constraints that currently constrain data centers from being able to manage them.
Optimizing data center performance for generative AI workloads
On the second day, a talk by Adam Levine, chief commercial officer at DATA4, entitled “Impact of AI on Data Center Design” underlined what data centers will need to make provision for to be commercially and environmentally successful when dealing with the accelerating number of AI workloads.
Levine started by emphasizing that data centers must undergo a fundamental change to support AI workloads, because traditional cloud and colocation data centers cannot support the power density needed for AI. The computing power required to support AI workloads is growing exponentially and it will mean an increasing supply of power and land to house the servers that run those workloads. As a result, he concluded that newer data centers will be built outside cities, despite resistance from local people, with green energy generation built nearby to meet the centers’ demands. Before AI, analysts predicted that data center power consumption would double by 2030. However, with the current projections for AI power consumption, this is expected to quadruple, leading to a global temperature increase of 0.5 degrees.
He believes that as AI experiences its rapid growth, the demand for and services of traditional cloud and colocation data centers will remain unaffected, and these established data centers should sustain an annual growth rate of approximately 10%. Omdia also forecasts this over the long term, despite a sharp dip in demand in 2023. Adam Levine stated: “The rise of AI is not just another trend in the data center industry, and it is poised to become an asset class of its own.” AI’s computationally demanding workloads necessitate expensive and exceptionally powerful servers, setting them apart from other data center workloads. Its distinctive server requirements and substantial power and cooling needs differentiate AI from the rest of the data center landscape, prompting a unique and specialized approach to AI management and infrastructure.
The final portion of Levine’s talk detailed that the pinnacle of AI silicon, the NVIDIA H100 GPU, is a significant leap in computational power compared to its predecessors, with substantially increased energy consumption. The higher energy demands present a pressing challenge for the data center industry, necessitating a dedicated focus on sustainability. In order to address this issue effectively, data centers must confront many hurdles, including the current power grid, recruiting specialized talent, resolving supply chain constraints, and mitigating bottlenecks in the quest for greater efficiency. He argued that retrofitting older data centers with AI technology is difficult, because these facilities often lack the infrastructure required to support the heightened power needs and heavier equipment. He finished by stating that regulation plays a pivotal role in mitigating the environmental impact of this thirst for power, emphasizing the imperative that data centers be designed and constructed with a net-zero approach in mind.
Omdia view
Omdia believes a proactive approach is crucial to ensure the longevity and competitiveness of data centers in a rapidly changing landscape, with generative AI models evolving and becoming more sophisticated. In the future, for some segments of the data center market, data centers may be able to outsource workloads to AI servers more efficiently than by using general-purpose servers. Therefore, data centers must anticipate the upgrading of their infrastructure to support AI servers.
Appendix
Further reading
Blockchain Technology and Adoption Trends (December 2019)
“Blockchain is good for more than just Bitcoin” (September 2019)
“CenturyLink goes ‘colorless’ and takes on the edge cloud” (February 2020)
Service Provider Routers & Switches Market Tracker – Q4 2019 (February 2020)
Li You, “Tech-savvy Hangzhou tries out new ‘City Brain’,” China Daily (retrieved June 17, 2021)
