Enterprises are accelerating the adoption of AI technologies, driven by the success of GenAI products like ChatGPT, to foster business innovation and improve efficiency. However, infrastructure and operation (I&O) leaders often struggle to balance these advancements with environmental sustainability. Many organizations developing AI capabilities tend to overlook the environmental impact of their AI infrastructure.
As AI technologies, models, and datasets grow more complex, the material impact and environmental footprint of AI infrastructure are becoming pressing concerns at the executive level. Ignoring sustainability requirements not only risks undermining organizational sustainability goals but also places I&O leaders under scrutiny. AI infrastructure, different from traditional computing infrastructure, requires specialized hardware, AI-specific software frameworks, and new operational strategies to maximize efficiency while reducing environmental impact.
Environmentally sustainable GenAI infrastructure represents a distinct shift from traditional I&O environmental sustainability practices. This new space requires effective GPU computing power utilization management to maximize efficiency, optimize power consumption, and reduce gas emissions. Additionally, it necessitates the adoption of new skills and operational approaches, such as GreenOps, to fully realize the value of AI.
As GenAI enters the multimodal era, I&O leaders must think strategically and act with agility to navigate changes in both the AI landscape and the evolving environmental, social, and governance (ESG) regulatory environment. To integrate sustainability into their approach, I&O leaders responsible for AI infrastructure initiatives should address the following steps.
Navigating Environmental Trade-Offs to Achieve AI Sustainability
As enterprises transition from piloting to operationalizing AI to accelerate their digital transformation, many are deploying GenAI-enabled applications in production environments. However, they often overlook the environmental trade-offs involved, such as the reusability of assets, the carbon intensity of electricity needed for data processing, and the water required for training and running AI models. I&O leaders face challenges in aligning AI infrastructure strategies with organizational sustainability goals, often mistaking AI infrastructure for general computing infrastructure and ignoring its environmental impacts.
To navigate these trade-offs and achieve sustainable AI, I&O leaders should consider the following steps:
Build/Production of AI Infrastructure: This stage involves the environmental impacts of manufacturing AI hardware and software. Negative impacts include natural resource consumption, greenhouse gas emissions, and power consumption. I&O leaders must review product carbon footprint (PCF) reports to compare the embodied carbon and materials impact of different products.
Delivery/Transportation of AI Infrastructure: Transporting AI hardware generates air pollution, toxic waste discharges, and greenhouse gas emissions. Organizations must use bulk packaging to reduce weight and materials waste. They must also prioritize ocean freight over air freight to lower transportation emissions.
Operations of AI Infrastructure: Operating AI infrastructure involves significant energy consumption and greenhouse gas emissions during data preparation, model training, and deployment. I&O leaders must ensure all systems are Energy Star, EPEAT, or TCO certified. Utilize analytics and telemetry tools to track energy performance in real time.
End of AI Infrastructure: The resource life cycle of AI infrastructure ends with recycling or disposing of e-waste, which has environmental and social impacts. I&O leaders must repurpose and redeploy machines back into the environment, where possible, to maximize asset life. It is also important to ensure that all IT asset disposition vendors are e-stewards or R2 certified to facilitate responsible disposal.
AI/GenAI is complex and requires significant investment in IT infrastructure upgrades and talent upskilling to operationalize it in an enterprise environment. I&O leaders responsible for AI infrastructure initiatives should clarify the environmental sustainability goals for each stage in the AI infrastructure life cycle. This includes setting clear objectives for reducing carbon footprints, optimizing resource usage, and optimizing energy usage and costs.
By addressing these steps, enterprises can better align their AI initiatives with sustainability goals, minimizing environmental impacts while advancing technological innovation. I&O leaders must engage in strategic planning across all facets of AI infrastructure, including procurement, development, deployment, resource allocation, recycling economics, and fostering an eco-culture. By taking a holistic approach, organizations can ensure that their AI initiatives not only drive innovation and efficiency but also align with broader environmental sustainability objectives.
Author bio: Stephen Du, Director Analyst at Gartner
Read next: AI investments fuel shared cloud infrastructure spending, surging 43.9% YoY in Q1 2024 – IDC
