# AI Sustainability: Practical Guide 2026 > Explore AI sustainability in depth: analyze the impact, from emissions to practical solutions for SMEs. Make informed choices for the Source: https://www.electe.net/post/sostenibilita-intelligenza-artificiale Site guide: https://www.electe.net/llms.txt Artificial intelligence can help a company consume less energy, reduce waste and make ESG reporting less manual. But those who build AI products also know the other half of the story: every model runs on energy-hungry infrastructure, every API call has a computational cost, and demand for compute isn't slowing down. This is why **AI sustainability** should be treated neither as a slogan nor as a sin to atone for. It should be managed as an operational problem. In Italy the debate has by now structured itself around two precise ideas: **sustainability of AI** and **AI for sustainability**. The first concerns AI's own footprint. The second concerns the use of AI to improve processes, consumption and environmental governance. In this same context, the ESG reporting component supported by AI has **"almost tripled in the last year"** according to [this analysis on AI sustainability and ESG governance](https://www.vegaformazione.it/PB/intelligenza-artificiale-sostenibilita-governance-ESG-p604.html). As the founder of an AI company, I find both automatic alarmism and lazy techno-optimism sterile. The point is not to decide whether AI is "good" or "bad" for the environment. The point is to understand **where it consumes, when it creates real value, and which concrete choices reduce the impact without destroying its usefulness**. ## Introduction: The Two Faces of Sustainable AI The conversation about **AI sustainability** has become more mature. Finally. Not because the problem has been solved, but because it has become impossible to reduce it to a catchphrase. On one hand, AI helps companies and facilities use energy better, limit waste and make ESG reporting more organized. On the other hand, these same systems require computing power, data centers, cooling, networks and hardware components that carry a real environmental cost. If you look at only one of the two sides, you make the wrong decision. ### The right language for talking about the problem The two categories I use most often are the ones now widespread in the Italian context as well: - **Sustainability of AI**. Reducing the energy and material impact of AI. - **AI for sustainability**. Using AI to improve environmental, energy and governance processes. The difference isn't theoretical. It forces you to ask two different questions. The first is: how much does my AI stack weigh? The second is: does that weight generate an environmental or operational benefit concrete enough to justify it? > The right question isn't "use AI or not". The right question is "does this task really need this amount of compute?". ### The most common mistake in companies The mistake I see most often isn't technological. It's a decision-making one. Many companies adopt AI as if all the available power should always be used. In practice, they choose the biggest model, the most complex workflow and the most extensive automation even when the problem was much simpler. In reality, a sustainable strategy starts from a less spectacular principle: **proportionality**. If a task can be performed well with less compute, less data transfer and less complexity, that isn't a sacrifice. It's a better choice. ## The Real Impact of AI on the Environment The most useful part of the discussion begins when you stop talking in the abstract. The environmental impact of AI isn't an opinion. It's an infrastructural issue with numbers already clear enough to demand caution. ### The numbers you shouldn't ignore An Italian source citing international data reports that a 2019 study from the University of Massachusetts estimated that training a single AI model can generate **over 284 tons of CO₂**, equivalent to the lifetime emissions of five cars. The same source reports that developing ChatGPT-3 would have required **around 1,287 MWh of electricity**, equal to the annual consumption of **around 120-130 average American homes**, and it also cites a Goldman Sachs Research projection according to which data center electricity demand could increase by **160%**, with a rise of **around 200 TWh per year between 2023 and 2030**, while by 2028 AI-related consumption could reach **19% of total data center energy demand**. All these figures are reported in [this Italian analysis on the environmental cost of AI](https://y7italy.com/il-costo-ambientale-dellia-e-le-priorita-della-sostenibilita/). The point isn't to use these numbers for scaremongering. The point is to understand the scale. If you're building or adopting AI, you're contributing to a rise in energy demand that affects entire infrastructures, not just your cloud budget. > **Practical rule:** when evaluating an AI project, look beyond the cost per token or per API call. The real problem often lies upstream, in the infrastructure that makes that call possible. ### Where the energy cost really concentrates Many people think the environmental cost of AI is almost entirely about training. That's a convenient simplification, but it's wrong. Consumption is spread across multiple levels: **Component****Why it matters**TrainingRequires large volumes of concentrated computationInferenceEvery user request generates repeated work over timeData centerPower supply, redundancy and ongoing infrastructure managementCoolingThe heat produced by the systems needs to be dissipated efficientlyNetworks and data transferMoving data between systems, regions and services has an energy cost This changes how **AI sustainability** needs to be approached. It's not enough to ask “how much does the model consume”. You also need to ask where it runs, how much traffic it generates, how many times it's queried, and whether the architecture was designed to limit waste. That's why it also makes sense to look at cases where AI is used to improve its own infrastructure. A useful example is the [energy optimization of data centers with AI](https://www.electe.net/post/sistema-di-raffreddamento-ai-di-google-deepmind-come-lintelligenza-artificiale-rivoluziona-lefficienza-energetica-dei-data-center), which clearly shows one simple thing: the problem isn't just the model, but the entire technical environment supporting it. ## AI as a Strategic Lever for Corporate Sustainability If you stop at the costs, you miss half the picture. AI can also be a serious lever for improving corporate sustainability. Not in a rhetorical sense, but in very concrete processes. ### Where AI creates concrete environmental value In the Italian context, when applied to energy management systems, AI can reduce operational consumption in industrial plants and SMEs by predicting demand peaks and regulating loads such as HVAC and lighting in real time. This has a direct impact on Scope 2 emissions and helps optimize the use of intermittent renewable sources, as described in [this in-depth look at AI-based energy management systems](https://ollum.it/blog/intelligenza-artificiale-sostenibilita-ia/). This is the kind of application I consider defensible even from an environmental standpoint. Not because it's free, but because it uses computation to eliminate continuous physical waste. In a plant with variable demand, adjusting ventilation, climate control or lighting in real time can be worth more than a thousand slides on the green transition. Three areas where AI tends to be genuinely useful are these: - **Operational energy**. Sensors, meters and algorithms help avoid unnecessary peaks. - **Supply chain**. Better demand forecasting can reduce unnecessary stock, handling and overproduction. - **ESG reporting**. Automating data collection, cleaning and structuring reduces manual work and fragmentation. ### The practical takeaway for an SME For an SME, the benefit often doesn't come from a spectacular model. It comes from a system that avoids duplicated work. Manual exports, multiple Excel spreadsheets, attachments sent over and over, reconciliations done by different people on the same data. All of this consumes time, machines, bandwidth and attention. > A process redone five times by different people is inefficient not just organizationally, but environmentally too. When AI centralizes data, automates repetitive steps and makes analyses reusable, the environmental value doesn't lie in the single inference. It lies in eliminating redundancy. For those working on reporting, an example close to this issue is [ELECTE for ESG reporting](https://www.electe.net/post/csrd-reporting-ai-automation), useful for understanding how automation can reduce manual steps and information dispersion without turning every activity into a complex IT project. ## Emerging Solutions for Greener AI The encouraging aspect is that the industry isn't standing still. AI is creating the problem, but it's also driving innovations that try to reduce its energy cost. It's worth being honest, though: there's no single solution. ### Hardware and infrastructure One promising direction focuses on interconnects and infrastructure efficiency. When computing systems wait for data, installed power is used poorly. This is one of the reasons why next-generation hardware should be read not only in terms of performance, but also of efficiency. In the technology debate, proposals are circulating that work on photonic connections, better data transmission and lower dissipation. The practical point, for those who buy cloud services rather than chips, is different: **choose stacks that do more useful work with less waiting, less traffic and less overhead**. This logic also applies outside AI in the strict sense. Those who want to think more broadly about less wasteful production models can find useful insights in these [sustainable economy practices](https://acasaloro.com/blog/esempi-di-economia-circolare), especially for connecting digital efficiency with the reduction of material waste in processes. ### Leaner software and architectures On the software side, the most interesting levers are often less glamorous and more immediate: - **Smaller models** when the task is well defined. - **Smart routing** between different models based on the type of request. - **Optimized inference** to reduce unnecessary calls, latency and data transfer. - **Traditional workflows** for tasks that don't really require an LLM. In my experience, this has been the choice with the greatest impact. The most effective change wasn't introducing some futuristic technology, but rather no longer using the most powerful model for everything. Classification, formatting, extraction of known patterns and many support operations don't always require a frontier model. They require architectural discipline. > If the result for the user remains equivalent, using less compute is not a compromise. It's better design. This is why I often talk about “the right model for the right task.” In practice, this means creating a routing layer that assigns each operation to the level of power it actually needs. It's a cost choice, but also a sustainability choice. Those who want to explore the connection between technical solutions and environmental impact can also read [ELECTE on AI and the environment](https://www.electe.net/post/lintelligenza-artificiale-per-lambiente-innovazioni-e-soluzioni-2025), which brings together several cases and development directions without presenting them as miracle cures. ## The Efficiency Paradox and the Responsibility of Choice The most important objection is also the most uncomfortable one: making AI more efficient doesn't automatically guarantee a reduction in overall consumption. In fact, sometimes the opposite happens. ### Why efficiency isn't enough When computing becomes cheaper, more accessible and faster, companies tend to use it more. More automations, more requests, more integrations, more processes generated “because it costs so little anyway.” This is the heart of the efficiency paradox. That's why **artificial intelligence sustainability** can't just be about better chips or more compact models. It has to become a priority criterion. Some questions should be asked before every implementation: - **Does this task really require generative AI?** - **Do you need a large model, or will a lightweight one do?** - **Does the automation eliminate a real cost, or does it just add elegant complexity?** - **Will the output actually be used, or will it just produce more digital noise?** One of the most obvious gaps in the Italian debate is exactly this: the lack of standardized metrics and a holistic approach to measuring AI's footprint across its life cycle, a point highlighted in [this systemic perspective on artificial intelligence sustainability](https://ratioiuris.it/la-sostenibilita-dellintelligenza-artificiale-una-prospettiva-sistemica/). The problem, then, isn't slowing down innovation. It's stopping treating computing as a free, inexhaustible resource. Every organization should treat it the way it treats budget or people's time: something to be allocated where it actually creates value. ## Practical Guide for SMEs: 4 Steps to Sustainable AI For an SME, talking about sustainability without an operational plan doesn't get you far. It's worth starting with four decisions that can be made right away, even without a perfect measurement of environmental footprint. An Italian source notes that many discussions on AI sustainability ignore the cost-benefit trade-off, while data centers already consume **about 3% of global energy**. The underrated question is: when does it make sense to use a smaller model or a traditional process to avoid an unsustainable rise in consumption and costs? The point is well summarized in [this article on AI's environmental trade-off](https://www.bo-om.it/news-ed-eventi/ai-sostenibilita-impatto-ambientale/). ### Four decisions that really matter 1. **Choose a model proportionate to the problem** This is the lever with the best ratio between simplicity and impact. Don't default to the most powerful model. Use it only where it makes a real difference. For many everyday tasks, lighter models or non-generative systems are enough. 2. **Evaluate cloud providers on their energy profile too** Don't just look at price, latency and compliance. Also look at transparency around data centers, regional energy mix and approach to operational efficiency. Geography matters. A workload doesn't have the same footprint in every region. 3. **Cut redundancy in workflows** If the same data is exported, recopied, reformatted and resent multiple times, you're wasting resources before you even start talking about AI. A platform like **ELECTE, an AI-powered data analytics platform for SMEs**, can be used to centralize data, automate reporting and reduce duplicate analysis. The advantage here isn't “doing more AI.” It's doing less unnecessary work. 4. **Measure, even imperfectly** Not every company has the tools to measure kWh or CO₂ per single inference. That's fine. Start with operational metrics that are already within reach. ### What to monitor even without perfect metrics In the absence of a truly established industry standard, I would start with a simple table: **Operational metric****What it tells you**API calls per completed taskWhether you're doing too much work to get an outputAverage model size per operationWhether you're overprovisioning tasksVolume of data transferred per sessionWhether your workflow generates avoidable trafficAnalyses produced but not usedWhether you're automating output nobody reads > Measuring poorly but consistently is better than not measuring at all. If after three months you see that the number of calls per task is dropping, that large models are being triggered less often and that data transfer is more contained, you still don't have perfect carbon accounting. But you do have something very useful: a discipline of efficiency that reduces computational waste, cost and, quite likely, environmental impact too. ## Conclusion: Lighting the Way to Responsible AI The sustainability of artificial intelligence isn't solved with a slogan. Nor with an ideological rejection of AI. It's built with better technical and managerial decisions. Anyone using AI today needs to hold two truths together. The first: the environmental cost exists, is growing, and shouldn't be minimized. The second: AI can produce concrete benefits when it reduces waste, improves energy use, and replaces repetitive, disorganized processes with smarter systems. The quality of the choice lies in the relationship between these two sides. For European companies there's also a second level of responsibility. Being compliant isn't enough. You need to understand how technology, infrastructure, governance and intellectual property intertwine. On this last point, for those who want to dig deeper into the legal framework accompanying AI adoption, I'd point to this [guide on artificial intelligence patents](https://www.studiolegalecoviello.com/intelligenza-artificiale-brevetti), useful for reading technological sustainability also as a long-term strategic choice. The right direction isn't to use less AI overall. It's to use **the right AI, in the right place, with the right level of power**. --- If you want to adopt AI more efficiently and more consciously, you can [see how ELECTE works](https://www.electe.net). The platform helps SMEs centralize data, automate reports and reduce redundant analytical work, with a pragmatic approach to model selection and operational efficiency.