OpenAI sustainability score: 25/100

OpenAI has published zero verified Scope 1, 2, or 3 emissions figures as of March 2026. Stanford Foundation Model Transparency Index found no disclosed energy usage, emissions, or water consumption. Third-party estimates place GPT-4 training at 12,000–15,000 tCO2e and annual GPT-4o inference at 391,000–463,000 MWh. Total operational emissions exceed 100,000 tonnes with no disclosure.

OpenAI has no published Scope 3 data or supply chain emissions assessment. Supply chain spans NVIDIA GPU manufacturing via TSMC, multi-state data center construction, and partnerships with Microsoft, Oracle, SoftBank. Microsoft's Scope 3 emissions (66% of 17.1 million tonnes in 2023) reflect OpenAI's outsized impact. Semiconductor fabrication involves water-intensive processes and mining with zero OpenAI disclosure.

ChatGPT users grew from 100 million (Jan 2023) to 800 million weekly (April 2025). GPT-4 training consumed 50x more energy than GPT-3. Stargate targets 10 GW by 2029. No emissions baseline exists, but user growth, model complexity, and infrastructure expansion all point to rapid escalation with no disclosed reduction pathway.

OpenAI claims 50% renewable via Microsoft Azure RECs (unverified). First Stargate facility in Texas uses on-site natural gas plant as primary power. Stargate Norway planned with hydropower. 100% renewable goal by 2027 relies entirely on Microsoft RECs with no direct additionality. Not a RE100 member or EU Climate Neutral Data Centre Pact signatory.

OpenAI has no published biodiversity policy, TNFD disclosure, or nature impact assessment. Data center construction in Texas, Wisconsin, Michigan, New Mexico involves land use and ecosystem impact with no disclosed environmental impact assessments. Supply chain includes water-intensive semiconductor manufacturing in water-scarce regions. Zero evidence of nature awareness in public communication.

OpenAI has no published waste strategy, circular economy targets, or resource use data. AI infrastructure involves rapid hardware turnover—GPUs and servers have limited lifespans. 10 GW expansion by 2029 includes no disclosed hardware lifecycle management. No evidence of supplier sustainability requirements or material stewardship framework.

OpenAI has published no water consumption figures. Academic estimates: GPT-3 training consumed ~700,000 litres; inference consumes substantially more. CEO Sam Altman dismissed water concerns as 'fake' at India AI Summit (Feb 2026). 56% of data centers globally still use evaporative cooling. Stargate Norway planned with closed-loop cooling; Texas facility details unclear.

OpenAI states carbon neutrality by 2030 and 100% renewable by 2027, but these are informal, unverified claims with no published pathway or interim milestones. Not registered with SBTi. 100% renewable goal relies entirely on Microsoft Azure RECs, not direct PPAs. Stargate expansion (10 GW by 2029) has no disclosed lifecycle carbon analysis. No offset strategy published.

OpenAI does not publish annual sustainability reports, ESG data tables, or third-party audited emissions. No sustainability governance, dedicated role, TCFD/CSRD disclosure. Stanford Foundation Model Transparency Index found zero disclosed key environmental metrics. DitchCarbon score: 23/100. Mission Alignment team disbanded (Feb 2026); corporate restructuring removed 'safely' from mission statement.

Global Witness (Jan 2025): ChatGPT produces greenwashing content on fossil fuels. Beyond Fossil Fuels (Feb 2026): 74% of big tech AI climate claims unproven, naming OpenAI. Altman dismissed water concerns as 'fake.' Donated $1M to Trump inauguration; reversed military policy, partnering with DoD. Lobbying spend increased 7x to $1.76M in 2024, focused on energy permitting and reduced AI regulation.