Tworzenie zrównoważonej sztucznej inteligencji w oparciu o analizę cyklu życia

Artificial intelligence (AI) is transforming industries at breakneck speed. From the large scale of data centers to the tiny, yet powerful, semiconductors that enable machine learning, the infrastructure behind AI demands a closer look.

Life Cycle Assessments (LCAs) help clients navigate the complexities of sustainability, offering a comprehensive review of environmental impacts across a product’s life cycle, and serving as strategic tools for companies aiming to meet ambitious climate goals.

One company that is looking to meet its climate goals is NVIDIA, and its GPUs (graphic processing units) are at the heart of today’s AI revolution. Its major tech clients also have climate goals and need product-level carbon data for Scope 3 reporting, which tracks emissions across their entire supply chain.

NVIDIA is actively evolving its sustainability strategy. The company has accelerated supplier engagement for Scope 3 emissions and invested in energy efficiency to improve its emissions reduction practices.

WSP recently collaborated with NVIDIA to deliver product carbon footprints (PCFs) for its NVIDIA HGX H100 and NVIDIA HGX B200 GPU baseboards. These LCAs incorporated supplier-specific data to enhance accuracy and enabled NVIDIA to publish public-facing PCF summaries for the HGX H100 and HGX B200 GPU baseboards.

Engaging directly with suppliers was critical for building granular models that reflect real-world environmental impacts, and working independently with different hyperscalers helped WSP identify common needs and emerging trends, enabling the development of solutions that are scalable, while also being sensitive to the client’s needs.

A recent article in Nature reinforces the importance of expanding the impact lens beyond carbon. AI’s environmental footprint includes energy, water, materials and land use, and the article highlights the need for a more holistic approach to improving LCA efficacy, especially as AI scales globally.

Strategically applied, LCAs help prioritize sustainability investments and conduct scenario modeling to evaluate trade-offs. Combined with environmental product declarations and open-access data sharing, regionalized LCA models will further improve transparency and comparability.

The future of LCAs in AI lies in system-level assessments. Moving from component-level footprints to full data center and AI workload evaluations will require consensus on functional units, system boundaries and improved data fidelity. These advancements will enable more meaningful insights and better alignment with policy and procurement strategies.

LCAs will increasingly inform regulatory compliance, supplier engagement and corporate responsibility reporting. One way we actively shape these standards is through participation in forums like the Open Compute Project Global Summit (OPC Global Summit) and by developing tools showcased in peer-reviewed publications and sharing them publicly in open-source platforms. We also gave a talk about “Empowering Sustainable Data Center Design with Microsoft’s Open Source Cloud LCA Tool.”

Ongoing efforts to refine assessment methods, such as establishing consensus on functional units and improving data accuracy, will be essential for meaningful system-level evaluations. Part of the challenge in these evaluations is that data centers often perform multiple AI workloads simultaneously, making it difficult to precisely attribute environmental or operational impacts to specific tasks. These resource burdens also increase exponentially as input token counts grow over time.

Participation in technical forums and development of transparent tools are contributing to a more standardized approach, helping organizations better understand and manage the environmental impacts of AI infrastructure.