Greenly Lifts Lid on ChatGPT-4 & DeepSeek's Sustainability

AI is often touted as the technology that could solve the climate crisis, yet, as it expands, so does its carbon footprint.

Google, Meta and Microsoft have all failed to keep track with net zero goals because of investments in AI and the emissions associated with running its associated technology.

This includes supercomputers, data centres and cooling systems.

A study conducted by Greenly, a specialist in enterprise carbon accounting, has compared the sustainability performance of two global AI platforms, OpenAI’s ChatGPT-4 and DeepSeek, a Chinese AI-powered chatbot.

The study raises urgent questions about the climate impact and sustainability of these next-generation AI models.

The environmental impact of AI

The operation of generative AI models, particularly LLMs, requires substantial computational resources.

The training and deployment of these systems require vast amounts of electricity and water, which in turn generate significant carbon emissions.

This scenario is particularly pronounced in models like ChatGPT-4, which boasts an enormous 1.8 trillion parameters, significantly more than its predecessors.

It is narrative arc that is very familiar: as the industry and the product become more sophisticated, the environmental impacts inevitably increases.

In a hypothetical business case where an organisation employs ChatGPT-4 to respond to one million emails each month, Greenly found that AI could generate 7,138 tCO₂e annually – the equivalent of 4,300 round-trip flights between Paris and New York. 

A single text-based request consumes as much energy as charging a smartphone to 16% according to research from Carnegie Mellon University and Hugging Face.

The cumulative impact is notable even with minimal usage of generative AI. A routine annual usage of AI under similar conditions would lead to 514 tCO₂e emissions.

The study further revealed that more energy-intensive applications, such as text-to-image tools like DALL-E, result in emission levels 60 times greater than those for text generation.

The potential of DeepSeek’s efficient design

In the search for more sustainable AI solutions, Chinese platform DeepSeek may have an answer.

This generative AI model uses a Mixture-of-Experts (MoE) architecture, activating only the relevant sub-models per task – significantly reducing the computing power needed.

DeepSeek’s model was trained using just 2,000 NVIDIA H800 chips, significantly less than the 25,000 used by ChatGPT-4 and the 16,000 employed by Meta Llama 3.1.

These chips are designed to be less energy-intensive than their more common counterparts.

DeepSeek’s operation consumes only a tenth of the GPU hours compared to Meta’s model, effectively lowering its carbon footprint as well as the resource demands on servers and water cooling systems.

However, Greenly highlights a potential issue, noting that while these advances tightly focus on efficiency, the rising global use of AI may offset gains due to increased volume.

“DeepSeek’s emergence has put energy efficiency at the heart of the battle between AI models,” says Alexis Normand, CEO and Co-founder of Greenly.

“But it remains to be seen if other players will follow this path, or continue to prioritise raw processing power at the expense of the environment.”

Efficiency standards through AI regulation

With the expanding integration of AI into various facets of society, regulatory efforts to establish ethical and sustainable boundaries have been underway. Among these is the European Union’s pioneering AI Act.

“AI has the potential to change the way we work and live and promises enormous benefits for citizens, our society and the European economy,” says Margrethe Vestager, Executive Vice President for a Europe Fit for the Digital Age.

“The European approach to technology puts people first and ensures that everyone’s rights are preserved. 

“With the AI Act, the EU has taken an important step to ensure that AI technology uptake respects EU rules in Europe.”

Despite the challenges posed by AI emissions, there is movement towards positive impact as AI technology gets leveraged to hasten decarbonisation, enhance energy efficiency, and achieve sustainable development objectives.

If harnessed judiciously, AI applications could contribute to a reduction in global emissions by 1.5-4% by 2030.

Promising strategies to shrink the tech sector's footprint include renewable-powered data centres, the proliferation of edge computing solutions and the reutilisation of open-source model frameworks.

“This act marks a major milestone in Europe's leadership in trustworthy AI,” explains Thierry Breton, Commissioner for Internal Market.

“With the entry into force of the AI Act, European democracy has delivered an effective, proportionate and world-first framework for AI, tackling risks and serving as a launchpad for European AI start-ups.”

Ultimately, while models like DeepSeek showcase promising strides in sustainable design, the wider technology sector must synchronise innovation with ecological accountability.

Explore the latest edition of Technology Magazine and be part of the conversation at our global conference series, Tech & AI LIVE.

Discover all our upcoming events and secure your tickets today.

Technology Magazine is a BizClik brand