Abstract
While the tech industry has prioritized carbon neutrality, biodiversity loss remains a critical yet overlooked planetary boundary in sustainable computing. This paper introduces FABRIC (Fabrication-to-grave Biodiversity Impact Calculator), the first end-to-end modeling framework designed to quantify the biodiversity impact of computing systems across their entire lifecycle. By introducing two new metrics—the Embodied Biodiversity Index (EBI) and the Operational Biodiversity Index (OBI)—we demonstrate that carbon footprints alone are insufficient indicators of ecological health. Our findings reveal that acidification is a dominant driver of species loss and that newer, more efficient hardware significantly reduces biodiversity impact per unit of performance despite higher manufacturing costs.
Introduction
The rapid expansion of AI and high-performance computing has led to a surge in energy consumption and hardware production. However, the environmental discourse has largely centered on "Net Zero" and carbon emissions. Biodiversity loss, characterized by the local extinction of species, is a distinct and equally urgent crisis. Until now, the computing community lacked quantifiable metrics to attribute ecosystem damage to specific hardware or tasks. FABRIC addresses this gap by translating technical specifications—such as kilowatt-hours or wafer layers—into ecological outcomes, specifically the expected fraction of species lost locally per year (species·yr).
The FABRIC Framework
The FABRIC framework bridges Life Cycle Assessment (LCA) with computer systems research. It employs a two-tiered approach to translate environmental stressors into biodiversity impacts.
- Mid-point Impacts: FABRIC tracks stressors including Acidification Potential (AP), Eutrophication Potential (EP), and Freshwater Ecotoxicity (FETP). These are measured in equivalents of $SO_2$, $PO_4$, and toxic units, respectively.
- End-point Impacts: Using the ReCiPe 2016 model, these stressors are converted into a unified metric: species·yr.
The impact is categorized into the Embodied Biodiversity Index (EBI), covering manufacturing, transportation, and end-of-life, and the Operational Biodiversity Index (OBI), which accounts for the impact of electricity generation during use.
Analysis and Key Insights
The study evaluated four generations of AMD EPYC CPUs and five NVIDIA GPUs, utilizing data from sustainability reports and environmental databases.
Manufacturing and Mid-point Stressors
Manufacturing is the primary driver of embodied biodiversity loss, accounting for 78–92% of mid-point impacts across all studied devices. Acidification, driven by $SO_2$ and $NO_x$ emissions, emerges as the "carbon" of biodiversity, responsible for 60–85% of the total endpoint impact.
The Efficiency Paradox
A critical finding is the "Efficiency Paradox." While advanced 5nm chips like the NVIDIA H100 have a higher absolute manufacturing impact than older 12nm T4 chips, their superior performance results in a biodiversity impact per TFLOPS that is approximately 15x lower. This suggests that hardware upgrades can be an effective strategy for reducing ecological footprints in performance-intensive environments.
Operational Dominance and Geospatial Sensitivity
In most operational scenarios, the OBI significantly outweighs the EBI. At 70% load, the operational impact can be 100x higher than the annualized embodied impact. Furthermore, biodiversity impact is highly sensitive to the local power grid. Relocating systems to regions with hydro-rich grids can reduce the OBI by two orders of magnitude, highlighting the importance of geospatial considerations in sustainable data center planning.
Conclusion and Future Directions
FABRIC provides a necessary lens for evaluating the true ecological cost of computing. However, current estimates represent a lower bound due to data limitations in upstream mining and material refinement. Future research will focus on expanding the system boundaries to include these destructive early-lifecycle stages and integrating biodiversity metrics directly into automated system optimization tools. By moving beyond carbon, the computing industry can begin to address its full impact on the planet's biological heritage.
Summary updated