The Carbon Footprint of AIin Fashion: A Hidden Cost?
AI is being sold to fashion as a sustainability solution. The reality is more complicated — and more important — than the industry is willing to admit.
The Green Promise and the Energy Bill Nobody Mentions
Fashion has a sustainability problem it has spent decades trying to solve. Overproduction, textile waste, carbon-intensive supply chains, water consumption at industrial scale — the industry’s environmental record is one of the most scrutinised in the global economy.
Artificial intelligence arrived carrying a compelling promise. Smarter demand forecasting would eliminate overproduction. Algorithmic inventory management would end the cycle of surplus and markdown. For an industry responsible for approximately 10% of global carbon emissions — roughly 1.2 billion tonnes of CO2 equivalent annually, surpassing the combined emissions of international flights and maritime shipping — AI looked like the missing mechanism.
That promise is not false. In several measurable areas, AI is delivering genuine sustainability gains. But the conversation has largely ignored a parallel reality: AI systems themselves carry a significant and growing carbon footprint. Training large machine learning models, running inference at scale, and maintaining the data infrastructure that powers fashion intelligence are all energy-intensive activities.
The carbon footprint of AI in fashion is only hidden because the industry chooses not to look.
Research published in Nature found that training a single generative AI model can emit as much carbon dioxide as five cars over their entire lifetime. For fashion brands developing proprietary models — Kering’s eco-design AI agent or Mango’s 15 internal machine learning platforms — these training costs accumulate rapidly.
The IEA reported that data centres accounted for 1.5% of global energy consumption in 2024, projected to rise to 3% by 2030 — and AI workloads will represent half of all data centre capacity by that point. A 2025 Nature paper projected that AI servers across the US alone will generate 24–44 million metric tons of CO2 annually, equivalent to adding 5–10 million cars to American roads. The annual water footprint for cooling: 731–1,125 million cubic meters.
Major AI providers including OpenAI and Anthropic have not publicly disclosed any emissions data, creating a fundamental accounting problem. When Cotopaxi’s sustainability team uses AI to calculate their carbon footprint, they cannot verify the footprint of the tool itself. As Annie Agle, VP of Impact at Cotopaxi, notes: “arriving at an exact figure is extremely difficult.”
Against this energy cost, the sustainability gains from AI in fashion are real and documented. According to Stanford University research, AI-optimised supply chains demonstrate a 45% reduction in overproduction and a 30% decrease in logistics-related carbon emissions. Every kilogram of unsold garment that gets incinerated carries with it the embodied carbon of its entire production process.
AI-powered computer vision is revolutionising textile recycling. Refiberd uses AI and spectroscopy to accurately classify fibre composition in used garments. Circ applies AI in chemical recycling to recover polyester and cotton from blended textiles — enabling true fibre-to-fibre circularity. AI design tools by Unspun reduce fabric waste by 40% through made-to-order production. AI-powered sorting improves recycling facility efficiency by 60% compared to manual separation.
Carbon accounting automation addresses a structural problem. Traditional carbon accounting in fashion is fragmented and labour-intensive. AI platforms like Greenstitch integrate with PLM, ERP, and SCM systems to provide real-time emissions tracking and automated Scope 1, 2, and 3 calculations — enabling brands to identify emissions hotspots with unprecedented granularity.
AI is neither a sustainability silver bullet nor a hidden environmental catastrophe. It reflects the intentions of those who deploy it.
The Core Truth
H&M has achieved a 23% reduction in Scope 3 emissions — one of the industry’s genuine success stories. Its AI systems optimise inventory placement, reduce overproduction, and automate sustainability reporting. However, H&M simultaneously operates extensive AI marketing platforms and chatbots handling millions of queries. The brand does not disclose the energy cost of these non-sustainability AI applications, creating an incomplete emissions picture.
Shein’s proprietary algorithms analyse real-time consumer behaviour to generate 10,000 new SKUs daily. While this is “efficient” in minimising inventory risk, it has driven emissions up 170% over two years — 6.3 million tons CO2 yearly, equivalent to 1.2 million cars. Here, AI optimises for speed and volume rather than sustainability, demonstrating that the same technology can accelerate environmental destruction when governance frameworks are absent.
Kering (Gucci, Balenciaga) developed an eco-design AI agent at its Material Innovation Lab to guide design teams on sustainability. However, its March 2026 appointment of a Chief Digital, AI and IT Officer signals massive AI expansion across all operations — including generative AI for marketing and customer analytics — raising legitimate questions about net impact at scale.
Logistics optimisation: 30% emission decrease
Fabric waste reduction: 40% decrease
Returns reduction: 8.25 kg CO2/item saved
Carbon accounting efficiency: 60% time reduction
US AI server emissions 2030: 44M tons CO2/yr
Data centre water cooling: 1,125M m3/year
AI workload energy share: 50% of data centres
Brand AI Scope 3 opacity: Largely unmeasured
A Framework for Fashion Executives
Not all AI applications have the same carbon profile. Ask: for each AI application, what is the approximate energy cost, and what is the approximate sustainability benefit? The ratio should inform prioritisation.
Most brands rely on third-party platforms whose energy consumption data they never request. Make energy efficiency and renewable energy sourcing part of vendor evaluation criteria — this creates market incentives for suppliers to improve their efficiency.
Most corporate carbon reporting frameworks do not yet require detailed accounting of AI system energy consumption. That will change under CSRD and the EU AI Act. Brands that develop this capability now will be better positioned when disclosure becomes mandatory.
Rather than each brand training proprietary models, the industry must develop shared AI infrastructure — open-source fashion AI models trained once and fine-tuned by multiple brands. Redundant training of similar models by competing brands is an enormous and unnecessary carbon cost.
Claiming sustainability benefits from AI without acknowledging the energy costs is a form of incomplete reporting that will increasingly attract regulatory and investor scrutiny. Present the full picture: gains, costs, net assessment, and improvement plan.
For most high-impact applications — demand forecasting, logistics optimisation, inventory management — sustainability benefits likely outweigh energy costs, often significantly. The carbon saved by reducing overproduction likely dwarfs the carbon cost of the AI systems generating the forecasts.
But that conclusion is not universal, not permanent, and not an excuse for sloppy accounting. The technology is too powerful to abandon and too energy-intensive to deploy uncritically. The next decade will determine whether AI becomes fashion’s greatest decarbonisation tool — or its most sophisticated greenwashing mechanism.
We cannot manage what we do not measure.
As long as AI providers refuse emissions transparency and brands exclude computational infrastructure from Scope 3 accounting, fashion’s sustainability transformation rests on incomplete data.
