The recent launch of DeepSeek, a Chinese artificial intelligence (AI) company reported to potentially reduce computing power per query by 90%, has sent shockwaves through financial markets. Investors reacted swiftly, causing energy stocks to tumble and calling into question long-held assumptions about AI’s skyrocketing energy demands.

While this market tumult has put a spotlight on AI’s soaring energy consumption, the frenzy surrounding DeepSeek’s high-performance, low-cost offering has also exposed a deeper, more troubling reality: There is no reliable way to forecast AI-driven energy demand. This uncertainty underscores the need for a risk-hedging strategy that ensures AI’s growth does not undermine global climate goals.

The stakes are high: Last year was officially the hottest on record, according to the World Meteorological Organization, with extreme weather exacting a heavy toll worldwide. In the US alone, 27 individual weather and climate disasters during 2024 caused $182.7 billion in damages, underscoring the financial consequences of a warming planet.

AI’s rapid expansion must be balanced with holistic mitigation strategies that integrate clean energy, infrastructure improvements and carbon removal to prevent further escalation of climate risk.

AI deployment versus clean energy growth

DeepSeek may have proven that the training of AI models can become more energy efficient, but the real question its launch raises is whether the technology’s breakneck deployment will outpace clean energy.

Enterprise AI spending is projected to reach $407 billion by 2027, growing at a compound annual rate of 36.2%. While renewables are also scaling up "at an unprecedented rate", AI’s accelerating deployment could still surpass clean energy availability. The International Energy Agency (IEA) estimates that, by 2030, renewables could provide nearly half of the world’s electricity. This would add 5,500 gigawatts (GW) of new capacity globally – roughly equivalent to the current power capacity of China, the EU, India and the US combined.

Based on my analysis of available figures, if AI-related electricity consumption reaches 134 terrawatt hours (TWh) annually by 2027, and only 38% of that demand is met with renewables, that would still leave 83.08 TWh powered by fossil fuels. Using 2023’s global average carbon intensity of 481 grams of carbon dioxide per kilowatt-hour (gCO2/kWh), this would result in approximately 39.96 million metric tons of CO2 per year. This highlights a critical issue: AI’s energy demand is set to grow faster than the clean energy expansion, locking in continued fossil fuel dependence.

Historical trends suggest that, even as AI efficiency improves, energy demand may still rise. Jevons Paradox, first observed during the Industrial Revolution, showed that, as coal-burning technology became more efficient, overall coal consumption rose due to greater economic productivity and accessibility.

If this holds for AI, as the technology becomes cheaper to operate, adoption will likely accelerate, driving up total energy use even as efficiency improves.

The AI infrastructure gap

This is troubling because, as the IEA has also reported, transmission bottlenecks, permitting delays and grid congestion are preventing the rapid deployment of clean energy. In the US, new energy projects face waits of up to five years for grid connection, which is currently delaying 2,600 GW of clean energy capacity. In both the US and Europe, data centre projects – some of which can require as much electricity as a small city – can also face multi-year delays in connecting to the grid.

With demand for energy outpacing reliable supply, AI companies cannot rely on clean energy – that is, emission reductions – alone. The shortfall means unavoidable emissions, which makes carbon removal not just a corporate responsibility, but a fundamental requirement for AI’s continued growth.

Unlike emissions reductions, which prevent new emissions from being released, carbon removal eliminates carbon dioxide (CO2) that is already in the atmosphere, as well as residual emissions that cannot be eliminated. Without it, AI companies risk entrenching high emissions for decades to come and face growing scrutiny from regulators, investors and customers demanding credible net-zero strategies.

While there are no mandatory compliance requirements today, regulations are coming. The EU has already formally adopted the world's first government certification framework for permanent carbon removals, paving the way for its integration into the region’s well-established emissions trading system.

So, alongside urgent emission reduction efforts, carbon removal provides a backstop against emissions that cannot be mitigated through clean energy solutions and efficiency measures. Deploying all of these strategies simultaneously will help meet AI’s growing energy demands while maintaining progress toward decarbonization.

Investing in carbon removal now

To manage the risk of rising emissions and carbon removal supply shortages, AI companies should secure long-term agreements with carbon removal providers now. By 2030, annual demand for durable carbon removal could range from around 40 to 200 million metric tons of CO2, according to a 2023 Boston Consulting Group report. This study also points out, however, that projects announced to date only account for about 33 million metric tons.

This significant supply gap requires substantial investment and early action. Scaling durable carbon removal solutions like biochar carbon removal (BCR) and direct air capture (DAC) will require years of infrastructure development, permitting and financing – these projects can take up to 10 years to become operational. Without immediate investment in infrastructure, there will not be enough carbon removal supply to meet future demand.

Microsoft, a member of the World Economic Forum’s First Movers Coalition, has emerged as the largest corporate buyer of carbon removal so far, accounting for 70% of all-time contracted volume through the end of the third quarter of 2024. Alongside emission reductions, it has invested in both nature-based carbon removal solutions such as reforestation, and engineered removal technologies such as BCR, DAC and bioenergy with carbon capture and storage (BECCS). These investments serve a dual purpose: balancing Microsoft's own emissions while also ensuring there is enough carbon removal supply to meet future demand.

Mitigating AI's climate risk before it’s too late

The recent financial market reaction to DeepSeek shows just how little visibility exists into AI’s long-term energy consumption. As the costs of climate inaction continue to mount – imposing a growing economic and societal toll – AI’s rising energy demand requires immediate action.

Efficiency in AI operations and clean energy will help balance AI’s growing energy demands, but this won’t eliminate all of its emissions. That’s where carbon removal comes in – not as a corporate afterthought, but as a necessary safeguard. Carbon removal can play a critical role in addressing residual emissions and ensuring AI continues to scale without undermining climate progress.

This article is also published on the World Economic Forum. 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.