How AI is driving up density requirements and the energy storage needed to support this
At the recent DCD>San Francisco event, we sat down to discuss hyperscale energy storage and sustainability in light of increasing rack-density needs. This seems to be a point of anxiety frequently raised in the data center industry, and unfortunately not one that can be ignored into submission.
One of the technologies that are driving this need for increased density is Artificial Intelligence (AI), a technology reliant upon high-performance computing (HPC), and not showing any signs of slowing down.
Steve Jennings, Senior Vice President at ZincFive Inc initiated the discussion around this high computing demand, and the obstacles it may present.
“We’re seeing server rack power-density climbing steadily across all kinds of application types. We’re also seeing the adoption rate for in-rack energy storage increasing, and of course, the data center footprint, as always, is remaining at a premium.
“We’re trying to get more power into the same amount of space, and that has challenges. How do you upgrade existing facilities to support increased loads? How do you maximize the ROI of high power density racks, versus the need for energy storage infrastructure in the rack? There’s a compromise there.”
A complex debate, but a conversation that needs to be had in order for the industry to keep up.
“Let’s start at the bottom, what the device level trends are that are driving this. So we’ve got new application-specific devices that are driven by the applications in artificial intelligence, machine learning, and deep learning. They’re not only HPC, high-performance computing devices, they’re also HPC, high power consumption devices, relative to traditional server device architectures. So as we see the speed go up, we can also see the power consumption go up.
“A lot of those working in the AI segment of the business are trying to mitigate the power requirement for these workloads, but the numbers are still going up significantly. From the data since 2012, which is considered the modern era in AI, we’ve had 300,000x growth. AI workloads are really driving the rack power density pretty heavily.”
In 2021, the AI industry was worth $327.5 billion, and this is predicted to reach over half a trillion by 2024. This is a prediction that is panic-inducing for a variety of reasons. The potential loss of jobs, the restructuring of entire industries, and the question of how we will successfully power these initiatives without compromising the wellbeing of the planet.
It is important to consider who is going to be impacted by this movement. In many small ways, the global population will be impacted in their day-to-day life. Their decisions will be automated and optimized through algorithms, the articles they read created by AI, rather than written by a journalist. But in terms of business decisions, data centers will take the brunt of the hit.
In light of this, data center operators need to be asking themselves important questions: how are they going to produce the energy required, and just as importantly, how are they going to store it?