Decoding the hidden environmental cost of AI

At 7 AM, Divya's smart home wakes her with a personalised morning briefing. Her AI assistant has already planned her most efficient commute and summarised her work emails. By 8 AM, she is video conferencing with colleagues across three continents, her translation app converting languages in real-time. Her medical app has cross-referenced her recent health data with global research, flagging a potential treatment.

This is our world now, one with a hyper-connected, artificially intelligent ecosystem that would have seemed like science fiction just a decade ago. Technology has changed every aspect of human existence, from how we work, communicate, learn, and live. AI writes our emails, composes our music, and predicts our next move.

But beneath this technological utopia lies a dark truth. The very systems creating this world are silently consuming our planet's future, one kilowatt at a time. The truth is brutally simple. While our tech giants parade their climate pledges, they are simultaneously stoking an AI fire that threatens to consume our planet's future.

In 1908, scientists first observed that if atmospheric CO2 concentrations were to double, Earth's surface temperature would rise by 4°C. More than a century later, we stand on the precipice of that prediction. North American data centres saw their power requirements explode from 2688 megawatts at the end of 2022 to 5341 megawatts by the end of 2023, nearly doubling in just twelve months. 

Researcher Sasha Luccioni noted that generative AI consumes 30 times more energy than a traditional search engine. In simple terms, a single average data centre now consumes enough electricity to heat 50,000 homes yearly. The electronic waste generated globally amounts to 57 million tons annually, equivalent to the weight of the Great Wall of China.

The broken promise of tech giants

It’s ironic. The very companies that talk about climate goals are simultaneously racing towards an AI-powered future that threatens those same environmental commitments.

Microsoft

  • Climate pledge: Carbon negative by 2030.

  • AI goal: Democratise AI through Copilot, integrating AI into every Microsoft product, and has invested $13 billion in OpenAI.

  • Energy consumption: Over 10 terawatt-hours annually in data centres, and greenhouse gas emissions rose 30% in fiscal year 2023.

Google 

  • Climate pledge: 24/7 carbon-free energy by 2030.

  • AI goal: AI-first approach, integrating generative AI into search, productivity tools, and cloud services. Its key AI products include Google Bard, AI-enhanced Search, DeepMind research.

  • Energy consumption: 18.3 terawatt-hours in 2022, with AI rapidly increasing this figure, with a 48% increase in carbon emissions since 2019.

Amazon

  • Climate pledge: Net-zero carbon by 2040.

  • AI goal: Generative AI through AWS, Bedrock, and Alexa improvements. It already has AI products like AWS AI services, generative AI for logistics and cloud computing.

  • Energy Consumption: 29 terawatt-hours in 2022.

Meta 

  • Climate pledge: 100% renewable electricity operations.

  • AI goal: Open-source AI models, AI agents across platforms

  • Energy Consumption: Approximately 9 terawatt-hours annually with multiple large-scale models (LLaMA 3) increasing energy demands.

Take Microsoft. In 2020, it unveiled an ambitious "carbon moonshot", a pledge to combat climate change. Fast forward to 2023, and its greenhouse gas emissions have risen 30% since that pledge. Microsoft has invested over $13 billion in OpenAI, effectively turning everyone into a "prompt engineer" while simultaneously watching their carbon footprint expand.

Google tells a similar story. Its 2024 Environmental Report reveals a 48% increase in carbon emissions since 2019, with a 13% rise in the last year alone. 

By 2026, data centres are projected to consume 1,050 terawatts of electricity: a consumption level that would place them fifth on the global electricity consumption list, nestled between Japan and Russia. Currently, data centres already account for 1% to 1.5% of global electricity use and 0.6% of global carbon emissions.

If generative AI adoption continues at its current pace, AI-related electricity use could consume 1–3% of global energy demand by 2030.

The environmental impact extends beyond electricity. Sir Keir Starmer's AI ambitions for the UK have already raised concerns about water supplies. Giant data centres require massive quantities of water to prevent overheating. This is an environmental cost that goes beyond carbon emissions.

The paradox

The most painful irony lies in the potential of the technology itself. AI has the capability to solve complex environmental challenges, yet its current implementation is exacerbating the very problems it could help resolve.

According to ChatGPT, the energy requirements of major AI models are:

 

AI Model

Estimated Energy Used

Equivalent CO₂ Emissions

GPT-3 (OpenAI, 2020)

~1,287 MWh

~550 tons CO₂e

GPT-4 (Estimate)

3x–5x GPT-3

~2,000–3,000 tons CO₂e

Google’s PaLM

~2,500 MWh+

~1,000 tons CO₂e

Meta’s LLaMA 2

~500–1,000 MWh

~200–400 tons CO₂e

 

To put this in perspective, this is equivalent to the annual energy consumption of 128 average households for a single model's training.

What the future holds

From my perspective, the future only has two possible scenarios. 

Scenario 1: Business as Usual

If these tech companies continue their current trajectory, we are looking at a future where global energy consumption by data centres rises to 1,050 terawatts by 2026, AI-related electricity use consumes 1-3% of global energy demand by 2030, carbon emissions continue to skyrocket, and water scarcity becomes a critical global issue.

Scenario 2: Transformation

An alternate future demands actual action wherein tech giants transparently report their AI's full carbon footprint, track environmental impact for every AI model, invest in renewable energy infrastructure, and keep strict energy consumption caps for AI model training.

The next few years will determine whether technology becomes the source of our destruction. Consider the Industrial Revolution. Initially, it was an environmental nightmare with smokestacks, unregulated pollution, and complete disregard for ecological consequences. But humanity adapted. We implemented regulations, developed cleaner technologies, and found more sustainable ways of progress.

With AI in the picture, we need to adapt. We cannot afford the same decades of environmental devastation.

Not all is lost. Some companies are showing more responsible approaches. For example, Apple focuses on on-device AI to reduce server energy consumption. Similarly, Salesforce integrates AI directly into climate tracking tools, while some tech giants are investing in renewable energy and more efficient cooling systems.

The IPCC report notes that we have until 2040 to reduce emissions. Without intervention, we risk a future where global temperatures could rise 4°C by 2100, potentially leading to ecosystems collapsing, regions becoming uninhabitable, and extreme exacerbation of global inequality.

While the world celebrated World Technology Day on May 11, we must ask ourselves a question: Are we celebrating innovation, or are we applauding our own potential extinction?

The future of technology is not about how smart we can make our machines, but how wisely we can use them to preserve our homes. These innovations need to match up to the climate goals.



News