GAISSA-Optimizer

Project Name: Cost Optimization of Artificial Intelligence Systems with Sustainable Practices (GAISSA-Optimizer)

Principal Investigator: Silverio Martínez-Fernández
Entrepreneurial scientist: Santiago del Rey
Team members: Vincenzo De Martino, Joel Castaño, Alexandra González
Transfer Manager: Marta Villarroya
Mentor: Andreas Jedlitschka

Context: All the investments in building highly accurate Artificial Intelligence (AI) systems have led to a dramatic growth in AI data volume, models’ size, and infrastructure capacity. The endless pursuit of achieving the highest possible accuracy has led to the exponential scaling of AI with significant energy costs and environmental footprint implications. Training costs—measured by computing resources— have surged 300,000x in six years, doubling every few months. Furthermore, AI companies have reported that inferencing can account for up to 90% of the cost of machine learning work. For instance, ChatGPT costs $700,000 per day in compute hardware costs: a single ChatGPT query uses 2.9 watt-hours of electricity, compared to only 0.3 watt-hours for a Google search, making ChatGPT ten times less efficient. After working with AI companies in our network (see letters of support), with billions of AI queries occurring daily, the overall economic and environmental impact is substantial. This poses a clear industrial need for cost optimization of AI systems.

Improvement: The project aims to tackle these industrial challenges utilizing the GAISSA-Optimizer tool. It consists of a web-based system that enables AI practitioners: (a) to simulate the return on investment (ROI) of implementing sustainable practices to achieve substantial cost savings; and (b) to integrate sustainable practices in the AI workflow of the company, leading to improved and standardized energy efficiency labels of AI systems.

Impact: Our tool significantly reduces up to 75% of operational costs and carbon emissions by optimizing energy use in AI workflows. Pilot implementations with partner companies will showcase measurable cost and energy savings, helping businesses align with sustainability regulations from ISO and UNE and attract investment. This enhances financial performance and drives a more sustainable AI industry.

Contact:  silverio.martinez (at) upc.edu