Ex-Hugging Face, Salesforce leaders launch AI sustainability firm

BetaKit reports that the Sustainable AI Group (SAIG) launched on Wednesday, led by Sasha Luccioni, formerly Hugging Face's AI and climate lead, and Boris Gamazaychikov, formerly Salesforce's head of AI sustainability. BetaKit says the Montreal-based research and advisory firm plans to work with businesses to help measure and report AI's environmental footprint and to liaise between buyers and large AI providers. BetaKit quotes Gamazaychikov: "Part of what we're trying to do is de-risk AI in general." BetaKit also cites the International Energy Agency estimate that asking ChatGPT a question can use up to 10 times the electricity of a Google search, and cites Goldman Sachs research projecting AI-driven data centre energy consumption could grow 160 percent to reach 3-4 percent of global electricity by the end of this decade. BetaKit reports Luccioni co-authored a paper this month addressing variation in how an "AI life cycle" is defined.

Industry-pattern observations: Enterprises and vendors face real measurement challenges when quantifying compute-related emissions. Public reporting often mixes different scopes and lifecycle boundaries, so metrics such as kWh per inference, model training hours, data centre PUE, and chip manufacturing impacts are not consistently comparable across providers. For practitioners, that inconsistency complicates benchmarking, procurement, and internal carbon accounting.

Reporting by BetaKit places SAIG's launch into a broader trend where specialized consultancies and research groups emerge to translate technical compute metrics into corporate sustainability signals. Increased scrutiny on cloud energy use and ESG disclosures, together with growing enterprise AI adoption, is raising demand for independent verification, standardized reporting frameworks, and clearer lifecycle definitions for models and infrastructure.

• •Whether SAIG or other groups publish reproducible methodologies or lifecycle frameworks that others adopt.
• •Uptake of standardized metrics from cloud and AI providers, such as per-query kWh or carbon-intensity indexing at region and time granularity.
• •Third-party benchmarks or audits for model training and inference energy use from research groups or advisory firms.
• •Regulatory or investor-driven disclosure requirements that reference compute-specific emissions accounting.

Editorial analysis: Teams tracking sustainability should expect ongoing debate over scope definitions and measurement methods, and will benefit from documenting workload characteristics, region/time-of-day energy profiles, and model lifecycle stages. Independent frameworks or third-party advisory reports can reduce vendor lock-in in sustainability reporting and improve comparability across procurement options.