A new analysis from technology services provider NTT Data warns that the rapid, large-scale deployment of artificial intelligence is creating an environmental crisis that could derail global climate commitments. The firm projects that the immense and growing appetite of AI systems for electricity, water, and strategic minerals poses a direct threat to international net-zero carbon emission goals unless fundamental changes are made to how the technology is developed and operated.
In a whitepaper titled “Sustainable AI for a Greener Tomorrow,” the company details the surging resource requirements behind the AI boom. Researchers predict that by 2028, AI-related workloads could drive more than half of all power consumption in data centers. This escalating demand extends beyond the power grid to freshwater reserves needed for cooling and finite rare-earth minerals required for advanced hardware. The report argues for an urgent, industry-wide shift from prioritizing raw performance to embedding sustainability and efficiency as core design principles in AI systems from their inception.
Surging Energy and Water Demands
The primary environmental challenge posed by AI is its staggering electricity consumption. The complex calculations required to train large language models and maintain always-on AI services are intensely power-hungry. NTT Data projects that as AI workloads become ubiquitous, they will account for over 50% of data center power use by 2028. This surge is expected to help more than double global data center electricity demand by 2030, consuming an amount comparable to the entire nation of Japan.
This escalating energy use has a direct impact on carbon emissions, with data center carbon footprints expected to climb to approximately 860 million tons of carbon dioxide equivalent by 2030. Beyond electricity, AI’s thirst for water represents a significant and often overlooked strain on resources. Data centers use vast quantities of fresh water in cooling tower systems to dissipate the immense heat generated by servers. Training a single large AI model can require hundreds of thousands of liters of water, and merely running between 10 and 50 queries can consume half a liter. The global AI demand in 2027 is projected to require between 4.2 and 6.6 billion cubic meters of water withdrawal, a volume greater than the annual water withdrawal of Denmark.
The Hidden Costs in Hardware
Abiotic Resource Depletion
AI’s environmental footprint extends deep into the global supply chain through the constant demand for new hardware. The report highlights the concept of abiotic resource depletion—the exhaustion of non-renewable materials like essential minerals and metals. The production of graphics processing units (GPUs), servers, and user devices requires significant quantities of copper, aluminum, and rare-earth elements such as cobalt and palladium.
Currently, digital user devices alone are responsible for 9.4% of global cobalt production and 8.9% of palladium output. These minerals are becoming increasingly scarce, yet the industry standard involves replacing servers every few years to keep pace with performance demands, compounding the problem of resource extraction and leading to massive volumes of electronic waste.
The E-Waste Challenge
This rapid hardware replacement cycle is a primary contributor to the growing global e-waste crisis. The report criticizes the prevailing tech lifecycle, where components are discarded rather than designed for modularity and upgrades. NTT Data advocates for a circular economy model, which emphasizes extending the lifespan of hardware through refurbishment, reuse, and responsible recycling to minimize the environmental impact of discarded electronics.
A Necessary Shift from Performance to Efficiency
Historically, the AI industry has focused almost exclusively on performance metrics like accuracy, speed, and latency. This singular focus has led to an arms race for computational power, with some modern AI models consuming over 300,000 times more energy than their predecessors. The result is what the report describes as an “increasingly exclusive domain,” accessible only to organizations with the vast financial resources needed to sustain such high energy demands.
NTT Data argues that this paradigm is unsustainable and calls for a fundamental reorientation. The firm’s experts urge the adoption of holistic sustainability goals where resource efficiency is treated as a core design principle, not a secondary consideration or a trade-off for performance. This involves building sustainability into AI systems from the very beginning of the development process.
Pathways to Sustainable AI
Addressing AI’s environmental impact requires coordinated action across the entire technology ecosystem, from hardware manufacturers and data center operators to software developers and policymakers. The report outlines several key interventions and strategies to forge a more sustainable path forward. Among the chief recommendations is the application of green software engineering—coding practices specifically designed to reduce energy and resource consumption.
Another proposed solution involves intelligently scheduling AI workloads to run in locations and at times that align with the availability of renewable energy sources. The company has put these ideas into practice with its own initiatives, including remote GPU services that shift processing tasks to energy-optimized locations. NTT Data has also developed its own large language model, named tsuzumi, which it claims requires 250 to 300 times less energy for training compared to conventional models.
Overcoming Systemic Barriers
Despite the availability of potential solutions, significant barriers to sustainable AI remain. The report identifies a lack of standardized, verifiable metrics as a major challenge. Without consistent frameworks comparable to those in traditional industries, it is difficult to quantify and compare the environmental footprint of different AI systems accurately.
The analysis also notes that many organizations have a narrow focus on energy consumption or carbon emissions, while failing to consider the comprehensive impact of water usage, material depletion, and e-waste. NTT Data calls for the industry-wide adoption of lifecycle thinking and circular economy principles as essential prerequisites for sustainable AI development. This holistic approach requires shared accountability across the entire value chain, including hardware makers, cloud providers, and end-users, to ensure that technological progress aligns with global sustainability goals.
The Dual Role of AI
While AI’s rapid expansion poses daunting resource challenges, the report emphasizes that the technology itself can be a powerful tool for solving the very environmental problems it creates. “AI’s amazing capabilities can help manage energy grids more efficiently, reduce overall emissions, model environmental risks and improve water conservation,” said David Costa, Chief Sustainability Business Officer at NTT Data. The key, according to the report, is to recognize the challenge immediately and consciously build sustainability into AI systems from the start.
