As AI promises to revolutionize our lives, it is crucial that we proactively integrate sustainability measures into all AI rollouts from day one.
Annit Lalla, Sustainability Technical Officer – Secure Power International, Schneider Electric
As organizations race to integrate AI into their operations, the data centers that host these AI workloads are undergoing a significant transformation. According to Schneider Electric’s latest white paper, Partnering for AI-Ready Data Centers, global data center capacity is projected to triple between 2022 and 2030. This growth is driven by the demand for accelerated computing for Artificial Intelligence (AI).
The Rise of AI Data Centers
AI data centers are notably more power-dense, with GPUs becoming the backbone of this technological revolution. Just a couple of years ago, the standard was around 10 kW per rack for traditional IT workloads; now, this has skyrocketed to over 100 kW per rack—a tenfold increase in power requirements! With this exponential rise in power demand, it is imperative that carbon reduction initiatives are not put on the back burner (pun intended!).
Acknowledging Multiple Stakeholders
Before diving deeper into sustainable data center practices, it’s essential to recognize the various stakeholders involved—both within and outside the organization. Central to this ecosystem is the IT department, responsible for providing AI platforms that serve business functions and customers. They must ensure high availability of AI workloads, which necessitates resilient electrical protection and precise temperature and humidity control to keep servers, storage, and networks operational.
To support the IT department’s objectives, the engineering team is tasked with redesigning infrastructure to meet AI requirements. Engineers often adopt a “worst-case” scenario approach to ensure availability and resiliency while hoping to achieve efficiency and a low Power Usage Effectiveness (PUE).
Additionally, the facility or data center operations team manages day-to-day operations, aiming for seamless and zero-downtime operations to meet Service Level Agreements (SLAs) committed to the IT department.
Given the high-stakes nature of the environment, these stakeholders understand that everyone’s job is on the line. Consequently, a conservative approach—prioritizing resiliency over efficiency—often prevails.
Embracing a Flexible Strategy
However, with advancements in concurrently maintainable designs, fail-safe processes, and intelligent digital support systems monitoring critical environments, this mindset can evolve. A Internal more flexible strategy that equally emphasizes sustainability is essential. Key to this strategy is the inclusion of other important stakeholders throughout the data center lifecycle.
Beyond the three primary stakeholders mentioned, it’s crucial to engage procurement, supply chain, partners, and customers. Organizational sustainability goals should be adjusted to ensure that the data center receives adequate focus, budget, and accountability in reducing carbon emissions.
Understanding Scope 3 Emissions
Scope 3 emissions constitute a significant portion of overall emissions for any data center, especially with the introduction of new AI infrastructure. Imagine a river of carbon fed by upstream suppliers, contributing to an organization’s embedded carbon and flowing downstream to partners and customers. Achieving net-zero ambitions requires collaboration across the entire ecosystem, including suppliers, the organization, and its partners and customers, to close the tap on carbon emissions.
The Five Tenets of a Holistic Sustainability Strategy
Schneider Electric believes that integrating the following five tenets creates a comprehensive environmental sustainability strategy for data centers:
- Setting a Bold and Actionable Strategy
- Establish a holistic, data-driven sustainability strategy that minimizes direct and indirect environmental impacts (including Scope 1, 2, and 3 emissions, biodiversity, raw materials, land, and water) by leveraging circular economy principles.
- Set measurable targets to develop actionable programs and establish transparent external reporting.
- Implementing Sustainable Data Center Designs
- Prioritize sustainability as a key parameter in data center designs, focusing on longevity, serviceability, and lower embodied carbon.
- Develop efficient, repeatable global data center designs that can be localized worldwide, selecting innovative products that are low carbon, environmentally compliant, and circular-ready.
- Driving Efficiency in Operations
- Utilize monitoring and data analytics to enhance operational efficiency, optimize system performance and longevity, and reduce operational expenditures (OpEx).
- Digitize aging infrastructure, adopt predictive maintenance practices, and use software tools to identify and manage inefficiencies across the data center.
- Buying Renewable Energy
- Address Scope 2 emissions by decarbonizing the energy mix through a comprehensive renewable procurement strategy.
- Meet long-term goals with an evolving mix of procurement options, including onsite solutions, offsite solutions, Energy Attribute Certificates (EACs), and offsets.
- Decarbonizing Supply Chains
- Tackle Scope 3 emissions by establishing a comprehensive supply chain decarbonization program that incorporates circular economy models, renewables, emerging technologies, and other reduction levers.
- Identify the size of supply chain carbon emissions, establish decarbonization programs and targets, and engage with partners who support these initiatives.
Conclusion: A Call to Action for Sustainability
By the time we recognized the impact of fossil fuels on the climate, they had already powered the world for decades—leaving us with no choice but to focus on damage control. As AI promises to similarly revolutionize our lives, it is crucial that we proactively integrate sustainability measures into all AI rollouts from day one. These measures must consider the entire ecosystem contributing to the AI revolution. Let’s remember: while our natural world may be indifferent to the AI revolution, the opposite can be very well true.