AI, climate, and the future of power: how DeepSeek changed everything

Let her sleep, for when she wakes, she will shake the world. — Napoleon Bonaparte, referring to China

Some stories are too compelling to ignore, especially when they sit at the crossroads of two of the biggest challenges facing humanity today. AI and climate change are subjects I’ve written about extensively, and both are reshaping economies, geopolitics, and the way we live.

DeepSeek’s latest breakthrough has sent shockwaves through Silicon Valley and rattled investors in America’s biggest tech firms. You’ll find plenty of discussions on investment trends and geopolitical implications elsewhere, but I want to focus on something at least as significant, DeepSeek’s potential impact on climate. This isn’t just another step forward in AI development, it could be a sustainability breakthrough.

At first glance, AI and climate might seem like separate concerns, one driven by the relentless march of technological progress, the other by the urgent need to decarbonize our economies. But AI’s rise comes with a staggering energy cost. The massive compute power required to train and operate these systems is driving up global electricity consumption, extending the lifespan of fossil fuel infrastructure, and placing unprecedented strain on water resources.

Enter DeepSeek, a Chinese AI startup that has achieved something few thought possible, a state-of-the-art AI model built at a fraction of the cost and energy of its Western competitors. While tech giants like Meta have spent $65 billion, Microsoft $80 billion, and OpenAI is planning a $500 billion AI infrastructure push, DeepSeek reportedly trained its model for just $5.5 million. More strikingly, they did it without access to the latest AI chips, until now considered essential for building powerful models.

For the first time, we have proof that AI doesn’t have to be an environmental catastrophe. The question is no longer whether we can build AI more efficiently, but whether the industry will take the 'right' lessons from DeepSeek or continue down an energy-hungry, emissions-intensive path.

AI’s energy problem: a climate crisis in the making

AI and climate change have an uneasy relationship. On one hand, AI holds immense potential to drive sustainability, optimizing energy grids, predicting extreme weather, and improving efficiency across industries. On the other, the very systems powering these breakthroughs are among the most resource-hungry technologies ever created.

We opened Pandora’s box on AI at a time when we need to be cutting emissions dramatically. Training a single large AI model generates as much CO₂ as flying from New York to London, return, 125 times. As a result, the global AI boom is extending the lifespan of coal and gas plants, worsening emissions at a time when rapid decarbonization is essential. Less often mentioned is AI data centers need vast amounts of water. Communities often share drinking water with tech infrastructure, plotting this out in a warming and drier world gets ugly.

These inconvenient truths are often ignored or downplayed. AI firms emphasize breakthroughs but rarely disclose their models’ energy consumption. Governments push AI development but fail to regulate its environmental impact. DeepSeek’s breakthrough challenges the assumption that more AI always means more emissions.

But how?

There is a fair amount of scepticism around DeepSeek’s claims, for example $5.5m is not the full cost of the model as it doesn’t include all the research and the engineers' salaries so the real cost is likely significantly higher. They may also have more powerful chips that they mention however in general they do seem to have just built a really good model.

As Plato observed more than two thousand years ago: 

Necessity is the mother of invention.

Faced with geopolitical and technological constraints, DeepSeek had to innovate. Without getting into technical details on Mixture of Experts Architecture, they used some smart engineering to squeeze the maximum out of the chips they did have. Constraints forced them to do more with less. 

The result? A model that can compete with OpenAI’s GPT at a fraction of the energy cost, suggesting that AI’s insatiable hunger for power may not be an unavoidable truth after all. Napoleon may not have foreseen AI or machine learning, but his prediction about China waking up to disrupt the global order feels very relevant here.

Karl Marx & the Jevons paradox: will AI’s energy demand ever shrink?

Great, problem solved, so we can have the benefits of AI without the environmental costs? Not so fast… Big Tech’s response has been to defend its position, making the argument that we still have to build the supercomputers to stay at the forefront of AI but think of what we can achieve using the DeepSeek efficiencies in our billion dollar systems. 

In 1865, economist William Stanley Jevons observed that as steam engines became more efficient, coal consumption didn’t decline, it skyrocketed. Cheaper, more accessible energy led to a massive increase in steam engine adoption, fuelling the rapid expansion of the Industrial Revolution.

This became known as Jevons’ Paradox, efficiency doesn’t necessarily reduce resource consumption, it can amplify it. If the same pattern emerges in AI. More efficient models like DeepSeek won’t reduce infrastructure spending, they’ll accelerate AI’s growth. From Victorian coal mines to modern data centres, Big Tech is betting that Jevons was right, efficiency gains don’t reduce infrastructure needs, they explode them.

And unless AI’s economic incentives change, AI’s energy use will explode as well. Unless AI companies, regulators, and investors redirect efficiency gains toward sustainable AI, Jevons’ Paradox could turn AI into one of the biggest climate threats of the 21st century. In the immortal words of Karl Marx, ‘history repeats itself, first as a tragedy, second as a farce’ 

Where will AI’s value be captured, and will it be sustainable?

For anyone who was worried about the concentration of power in the hands of a small number of wealthy individuals this is good news. A start up from China has democratised AI. The conventional wisdom has been that AI will follow the pattern of past tech markets, concentrating power in a few dominant players, just as Google dominated search, Amazon dominated online shopping and Facebook dominated social media. 

DeepSeek has not only thrown the assumption that you need to be huge to compete in AI into question. It has also released its software as open-source software so anyone anywhere can download it into their own computers or tweak it so serve their own purposes. The barriers to entry into AI have crashed as developers or startups wishing to develop their own AI tools have an open-source off the shelf piece of software available to them. Bottom line is as long as you have an internet connection anyone, anywhere can now access a very powerful model, for free.

Open-source AI may play out more like the commercialization of electricity. If true, the real value won’t be captured by a single monopoly AI firm, but by the businesses and industries that build on top of AI infrastructure. AI could become a fundamental utility, embedded in climate modelling, industrial automation, and clean energy innovation, rather than a walled garden controlled by a handful of tech giants.

But this future is far from certain. AI may continue down the Jevons Paradox path, where efficiency only fuels more emissions, and infrastructure spending locks in high-carbon AI growth. But if regulators, investors, AI developers and society in general learns from DeepSeek, we could see an alternative path emerge. Lowering the cost of entry will encourage more scrappy startups with their innovative spirit to enter the space. Let’s task one of those startups with finding a model for tracking computational and energy consumption in AI systems and mandate their disclosure. Let’s have another that buildings AI-powered climate solutions, where efficiency gains are redirected toward solving global environmental challenges.

So, amidst all the geopolitical noise the real question isn’t just who will capture AI’s value, it’s whether AI will create value for the planet, or simply extract it.

To quote a Chinese proverb:

When the winds of change blow, some people build walls, others build windmills.

AI’s future is not just about economic value. AI is at a crossroads. One path leads to hyper-concentrated power, rising emissions, and a runaway expansion of AI infrastructure. The other leads to AI as a tool for sustainability, one that prioritizes efficiency in service of planetary limits. Which path we take will define AI’s legacy in the 21st century.

illuminem Voices is a democratic space presenting the thoughts and opinions of leading Sustainability & Energy writers, their opinions do not necessarily represent those of illuminem.