OpenAI will not disclose GPT-5’s energy use. It could be higher than past models – The Guardian

Report on the Environmental Impact of Advanced Artificial Intelligence and its Implications for Sustainable Development Goals

The recent development of increasingly powerful Artificial Intelligence (AI) models, such as OpenAI’s GPT-5, presents a significant challenge to global sustainability commitments. The escalating energy and resource consumption required to train and operate these models directly conflicts with several key United Nations Sustainable Development Goals (SDGs), particularly those concerning energy, responsible production, and climate action.

Analysis of AI Energy Consumption and Sustainability

Escalating Energy Demand of New AI Models

The progression from earlier AI models to the latest generation is marked by a substantial increase in energy consumption. This trend raises critical questions about the environmental viability of the current trajectory of AI development.

• A query to a mid-2023 model consumed approximately 2 watt-hours of electricity.
• Experts estimate that a comparable query to the new GPT-5 model could consume up to 20 times that amount.
• Independent research from the University of Rhode Island indicates an average consumption for GPT-5 of 18 watt-hours per medium-length response, with a potential peak of 40 watt-hours.
• Aggregated over billions of daily requests, GPT-5’s potential electricity usage could equal the daily demand of 1.5 million U.S. households.

Conflict with SDG 7 (Affordable and Clean Energy) and SDG 13 (Climate Action)

This massive energy footprint places a direct strain on energy grids and undermines efforts to transition to sustainable energy sources. The development of energy-intensive AI technologies runs counter to the objectives of SDG 7, which aims to ensure access to affordable, reliable, and sustainable energy for all. Furthermore, as a significant portion of the electricity powering data centers is generated from fossil fuels, this surge in consumption directly contributes to greenhouse gas emissions, impeding progress towards SDG 13 (Climate Action).

Corporate Responsibility and the Call for Transparency

A Deficit in Environmental Reporting

A primary obstacle to assessing the full environmental cost of AI is the lack of transparency from major technology firms. Most leading AI companies, including OpenAI, have not released official data on the power consumption or resource footprint of their latest models. This opacity prevents comprehensive analysis and accountability.

1. OpenAI has not published official energy usage figures for any model since GPT-3 in 2020.
2. Undocumented figures provided by corporate leadership lack the necessary verification for credible environmental assessment.
3. In contrast, disclosures from other firms, such as Mistral AI, confirm a “strong correlation” between a model’s size and its energy consumption, reinforcing the need for industry-wide standards.

Aligning with SDG 12 (Responsible Consumption and Production)

The current paradigm of developing ever-larger AI models without public disclosure of their environmental impact is inconsistent with the principles of SDG 12. This goal calls for promoting resource and energy efficiency and encouraging companies to adopt sustainable practices and integrate sustainability information into their reporting cycles (Target 12.6). Researchers and academics are now calling on developers to:

• Commit to full transparency regarding the environmental costs of their models.
• Publicly disclose data on energy consumption, water usage, and carbon emissions for new AI systems like GPT-5.
• Collaborate on establishing a global environmental standard for AI to guide responsible development.

Technological Factors and Infrastructure Sustainability

Drivers of AI Resource Consumption

The environmental footprint of an AI model is determined by a combination of factors, highlighting areas for potential mitigation and the need for sustainable innovation.

• Model Size: The number of parameters in a model is a primary determinant of its energy use. GPT-5 is believed to be significantly larger than its predecessors.
• Task Complexity: Advanced reasoning and multimodal capabilities (processing text, images, and video) are far more energy-intensive than simple text generation.
• Hardware and Architecture: While more efficient hardware can reduce consumption, the use of streamlined architectures like “mixture-of-experts” is crucial for mitigating the energy demands of large-scale models.

Implications for SDG 9 (Industry, Innovation, and Infrastructure)

While AI is a driver of innovation (SDG 9), the infrastructure supporting it must be sustainable. The current trend of scaling up AI models without prioritizing energy efficiency challenges the objective of Target 9.4, which aims to upgrade infrastructure and retrofit industries to make them sustainable. Achieving truly sustainable innovation requires a paradigm shift where environmental impact is a core consideration in AI design, not an afterthought.

SDGs Addressed in the Article

1. SDG 7: Affordable and Clean Energy

   • The article’s central theme is the massive and increasing electricity consumption of AI models like GPT-5. It directly addresses the challenge of energy demand, which is a core component of SDG 7. The comparison of GPT-5’s potential daily energy use to that of “1.5m US homes” highlights the scale of this energy challenge.

2. SDG 12: Responsible Consumption and Production

   • This goal is highly relevant as the article scrutinizes the resource footprint (consumption) of AI technology (production). It questions the sustainability of producing increasingly powerful AI models without regard for their energy and water use. The call from researchers for “full transparency” and for companies to “publicly disclos[e] GPT-5’s environmental impact” directly relates to promoting responsible corporate practices.

3. SDG 6: Clean Water and Sanitation

   • The article explicitly mentions water consumption as part of the resource footprint of AI. It cites a figure from OpenAI’s CEO of “0.000085 gallons of water per query.” This connects the operation of data centers for AI to the sustainable management of water resources, a key aspect of SDG 6.

4. SDG 13: Climate Action

   • Although not explicitly mentioning “climate change,” the article’s focus on massive electricity consumption has direct implications for SDG 13. Such a large energy demand, if met by fossil fuels, would result in significant carbon emissions. Therefore, understanding and mitigating the energy footprint of AI is a crucial step in climate action.

Specific SDG Targets Identified

1. Target 7.3: By 2030, double the global rate of improvement in energy efficiency.

   • The article directly engages with the concept of energy efficiency in AI. It notes that newer models like GPT-5 are less efficient on a per-query basis, consuming “significantly more energy than GPT-4o.” However, it also mentions technological efforts to improve efficiency, such as the “mixture-of-experts” architecture, which “will likely cut its energy consumption.” This discussion aligns perfectly with the goal of improving energy efficiency.

2. Target 12.2: By 2030, achieve the sustainable management and efficient use of natural resources.

   • The entire article is an analysis of the use of natural resources—specifically electricity and water—by AI models. Researchers are attempting to “benchmark the energy and resource usage of AI models” to understand if their use is sustainable and efficient. The finding that GPT-5 could consume up to “40 watt-hours of electricity to generate a medium-length response” points to an inefficient use of these resources.

3. Target 12.6: Encourage companies, especially large and transnational companies, to adopt sustainable practices and to integrate sustainability information into their reporting cycle.

   • This is a central point of the article. It highlights that OpenAI “has released no official information on the power usage of its models since GPT-3” and that existing figures “have no supporting documentation.” The article concludes with a direct call to action: “We call on OpenAI and other developers to use this moment to commit to full transparency by publicly disclosing GPT-5’s environmental impact.” This is a clear demand for the corporate sustainability reporting outlined in Target 12.6.

4. Target 6.4: By 2030, substantially increase water-use efficiency across all sectors.

   • The article’s mention of water consumption, specifically the figure of “0.000085 gallons of water per query,” directly relates to the water use of the technology sector. Analyzing and benchmarking this consumption is the first step toward increasing water-use efficiency as mandated by this target.

Indicators for Measuring Progress

1. Energy Consumption per AI Query (in watt-hours)

   • The article provides specific, quantifiable data points that serve as direct indicators. It mentions that a query on an older model might take “2 watt-hours,” while researchers found GPT-5’s average consumption is “just over 18 watt-hours” and can be as high as “40 watt-hours.” This indicator can be used to measure progress (or regression) toward Target 7.3 (Energy Efficiency).

2. Water Consumption per AI Query (in gallons)

   • An indicator is explicitly stated in the article, referencing Sam Altman’s blog post: “0.000085 gallons of water per query.” Although the article notes this figure lacks documentation, it establishes a benchmark that can be used to measure water-use efficiency (Target 6.4) for different AI models.

3. Public Disclosure of Environmental Impact Data

   • The article implies this as a crucial qualitative indicator. The fact that “OpenAI, like most of its competitors, has released no official information on the power usage of its models since GPT-3” is presented as a problem. Therefore, the number of AI companies that publicly and transparently report on the energy consumption, water usage, and carbon emissions of their models serves as a key indicator for measuring progress toward Target 12.6 (Corporate Sustainability Reporting).

4. Total Electricity Consumption of AI Services

   • The article provides a macro-level indicator to illustrate the scale of the issue by estimating that “the total consumption of GPT-5 could reach the daily electricity demand of 1.5m US homes.” This aggregate figure is a powerful indicator of the overall environmental footprint of the technology, relevant to SDG 7 and SDG 13.

Summary Table of SDGs, Targets, and Indicators

Source: theguardian.com