AI’s impact on data center power requirements
We’ve entered the era of AI, and everyone is interested in tapping into its potential – individuals and enterprises alike. It’s easy to imagine that mass adoption of AI could change the world in dramatic ways.
AI, however, isn’t some standalone technology that can be leveraged on its own. To meet current and future demands for AI, the world’s technological infrastructure will have to undergo some major changes. That’s in part because, compared to other existing digital tools, the power draw of AI is immense.
To fully understand the power demands of AI, consider the hypothetical scenario posited in a research paper recently published in the scientific journal Joule: If Google replaced its current search engine with ChatGPT-like technology, its power consumption would soar. In 2021, Google’s total electricity consumption was 18.3 TWh, with AI accounting for 10-15 percent of this total, researcher Alex de Vries noted. But with ChatGPT-like functionality, it would take 29.3 TWh per year just to power Google Search. That’s as much electricity as the entire country of Ireland typically consumes in a year.
To be clear, Google won’t be making this kind of move, for several reasons. Even so, de Vries noted, Nvidia is projected to ship 1.5 million AI server units per year by 2027. Those servers would consume somewhere between 85.5 to 134 TWh of electricity annually.
This brings data center operators to an inflection point. To keep up with modern demands for generative AI, companies will have to either build greenfield data centers or rip and replace their existing infrastructure.
Data Center Modernization: Building for Power Density
Read PostWhat’s more, businesses have to consider exactly how to power their new servers – when building out new power infrastructure for the data center, you don’t want to rely on yesterday’s technology.
In the era of AI, power density in the data center is more important than ever. This should prompt data center architects to consider emerging options like nickel-zinc (NiZn) batteries, an innovation that delivers industry-leading power density.
Every data center has to include a battery backup system (BBU) – either distributed in the server racks or a centralized uninterruptible power supply (UPS) with batteries – to ensure that in the event of a power outage, critical systems keep running and vital data is preserved.
Traditionally, UPS systems have relied on lead-acid batteries. While sticking with the status quo has its advantages, NiZn batteries have twice the power density. For the same level of backup power, NiZn is about half the size and half the weight.
More modern UPS and BBU systems also leverage lithium-ion batteries as an alternative to lead-acid batteries. Lithium-ion batteries are indeed more efficient and denser than lead-acid batteries. As far as data center battery options go, lithium-ion offers the highest energy density.
In other words, lithium-ion batteries will slowly trickle out energy over time but are unable to meet the high power density requirements of modern AI infrastructure.
By comparison, NiZn batteries have a similar energy density to lithium-ion batteries but offer a much higher power density. This means NiZn batteries will safely discharge higher levels of power than other technologies in a smaller space, making it optimal for a power-dense AI infrastructure backup power system that’s expected to leap into action quickly and keep mission-critical systems running, just milliseconds after a power outage.
All told, a power-dense technology amounts to a smaller battery, which means smaller backup systems and a smaller overall footprint – allowing data center operators to power AI solutions without having to add real estate at an untenable pace.
A recent forecast from Synergy Research Group shows how the largest data center operators are already putting more value on power density. According to the research firm, the average capacity of new hyperscale data centers to be opened over the next six years will soon be more than double that of current operational hyperscale data centers.
Higher power density also allows data center operators to better control peak loads – an increasingly important issue in the AI era.
It’s Time to Rethink Data Center Power
Read PostThe power draw of AI compute isn’t a constant. Instead, its power requirements come in ebbs and flows, with higher power draws during training runs or when enterprise-grade models are put into production.
There are several factors aside from power density that will impact the performance and sustainability of data centers in the AI era. When it comes to power generation, heat generation, cooling, and thermal volatility are major considerations.
These are areas where NiZn once again offers notable benefits. With no thermal runway, NiZn batteries can operate over a wider temperature range than other batteries. That means NiZn requires less cooling technology. NiZn batteries also require less safety-related infrastructure than lithium-ion alternatives, which are chemically volatile.
Investing in the infrastructure to support generative AI will take forward-looking strategic decision-making. It’s never easy to invest in emerging technologies, but a revolutionary innovation like generative AI deserves the most future-proof infrastructure.
Previously published with Data Center Dynamics
