Artificial intelligence is on course to become a major driver of global resource consumption, with its environmental footprint extending far beyond carbon emissions, according to a new United Nations University (UNU) report.
By 2030, the global data centres underpinning AI systems could consume 945 terawatt-hours (TWh) of electricity annually – almost 3% of projected global demand. That is nearly triple the combined electricity consumption of Pakistan, Bangladesh and Nigeria. At the same time, AI’s water footprint could reach 9.3 trillion litres – equivalent to the domestic needs of 1.3 billion people in sub-Saharan Africa.
A growing footprint beyond carbon
Crucially, the report argues that AI’s environmental impact is being systematically underestimated because it is typically measured through a carbon-only lens. In reality, every unit of electricity used by AI carries additional water and land costs, linked to cooling, power generation and infrastructure footprint.
These impacts do not move in tandem. Efforts to reduce carbon – such as switching to certain renewable energy sources – can significantly increase water and land use. The result, the report warns, is that focusing narrowly on emissions risks shifting environmental burdens onto water- or land-stressed regions.
From training to everyday use
The study also challenges assumptions about where AI’s energy demand originates. While training large models has attracted most scrutiny, it is the day-to-day running of AI systems – known as inference – that dominates.
Inference accounts for an estimated 80-90% of total AI energy use. With products like ChatGPT handling billions of daily prompts, the cumulative energy demand quickly scales to industrial levels. Moreover, energy consumption varies dramatically by application: generating AI images or video can be orders of magnitude more resource-intensive than basic text processing.
Efficiency gains alone are unlikely to contain this growth. The report highlights the “rebound effect”, whereby cheaper and more efficient systems drive higher usage, ultimately increasing total consumption.
Infrastructure at country scale
AI infrastructure is already reaching country-scale resource demand. Data centres consumed an estimated 448 TWh in 2025, putting them among the world’s largest electricity users. By 2030, their land footprint could exceed 14,500 square kilometres – roughly twice the size of the Jakarta metropolitan area.
Local impacts are also becoming visible. In Ireland, data centres accounted for 21% of electricity demand in 2023, prompting limits on new grid connections. Meanwhile, water stress linked to data centre expansion has raised concerns in parts of Mexico and Uruguay.
The report also highlights global inequalities: more than 90% of AI compute capacity is concentrated in the US and China, while the environmental costs of mineral extraction and e-waste – projected to reach 2.5 million tonnes annually by 2030 – are often borne elsewhere.
“The global system building artificial intelligence must also govern it sustainably and fairly,” said Professor Tshilidzi Marwala, Rector of the United Nations University and Under-Secretary-General of the United Nations. “The concentrated development of AI infrastructure in the privileged areas of the world is creating a large digital divide that poses profound challenges in the equitable development of AI. AI can certainly advance prosperity and human well-being. Whether it does so equitably is now a governance question, not a technical one.”
Agribusiness remains bullish
Despite these projections, appetite for AI in agriculture shows little sign of slowing. Global agrifoodtech investment surpassed $16bn in 2025, according to AgFunder, with roughly one-third – around $5bn – flowing into deeptech areas including AI.
For major agribusinesses, the central calculation is not the footprint of AI infrastructure itself, but the balance between its costs and downstream benefits.
Syngenta, for example, is explicit in its position. “We operate on the premise that the environmental ‘cost’ of AI compute will always be far outweighed by the environmental ‘savings’ it generates in the field,” a company spokesperson told AgNavigator.
The company argues that, in the context of 4.8 billion hectares of global agricultural land, the physical footprint of data centres remains relatively small. More importantly, it sees AI as a critical enabler of resource efficiency.
‘Savings’ in the field
According to Syngenta, AI-driven tools can help reduce agriculture’s environmental footprint in multiple ways.
Applications include precision irrigation, crop mapping and predictive analytics, all aimed at optimising input use and reducing waste. AI-led seed breeding and crop protection research are also designed to increase yields on existing farmland, potentially limiting the need to convert forests and natural habitats.
Digital tools such as satellite-based imagery platforms and AI-driven irrigation models are already being deployed to help farmers cut water use and improve productivity.
At the same time, the company says it is factoring sustainability into its digital infrastructure choices, including selecting cloud providers with commitments to renewable energy and “water positive” operations.
A question of balance
The tension highlighted by the UNU report reflects a broader shift as AI moves from hype to hard economics in agribusiness.
On one side, researchers warn that unchecked growth in AI infrastructure could place significant new strain on global resources. On the other, industry players argue that AI is indispensable to improving the efficiency and sustainability of food production.
For Syngenta, the key lies in responsible deployment. “AI is a powerful tool, but not always the best solution,” the spokesperson noted, adding that the company prioritises efficiency and uses AI “only where it is needed”.
The UNU report ultimately reaches a similar conclusion: keeping AI within planetary limits is achievable – but only with better measurement, transparency and coordinated action.
What remains unresolved is whether the environmental savings promised in the field will be enough to offset the rapidly expanding footprint of the systems behind it.
