29 May 2026

Forecasts of energy supply and demand from seconds to years ahead drive the decisions that ultimately determine where the energy we consume comes from, and at what cost. In the long-term, policy is made, infrastructure is built, maintenance is carried out. Energy is bought and sold from years ahead to just a few minutes before delivery. From seconds to days ahead, plans are made and re-made to ensure the reliable physical delivery of energy to end-users. Everything is put in place so that when we flick a switch, the lights turn on.

This blog series has previously highlighted the role of AI in forecasting, and indeed current best practice involves using Machine Learning and AI algorithms to learn patterns to predict energy production/consumption from weather and contextual data, be that power production from a wind farm or electricity consumption for a neighbourhood. However, the transition to net zero is already disrupting the practice of predicting demand and scheduling supply to match it. Weather-dependent renewables with no fuel costs introduce uncertainty on the supply side, and demand is increasingly flexible and less predictable.

Here I use “AI” broadly to include statistics, machine learning, and optimisation, as well as generative AI.

Potential roles for AI

Opportunities for AI in this area fall into two categories: improving forecast performance, i.e. accuracy, and improving use of forecast information leading to better outcomes.

The first category is the easiest to comprehend. Getting better at recognising patterns in ever larger datasets should result in more accurate forecasts and more reliable quantification of uncertainty. However, in recent history, the most significant improvements in energy forecast performance have come from finding new sources of predictability and pairing them with suitable AI/ML algorithms, not from advances in AI/ML per se. I include weather forecasts as a source of predictability here. Improvements in weather forecasting, and the extraction of information from weather forecasts, have both contributed to improvement in energy forecasts. AI weather forecasts may be one of the next new sources of predictability for energy forecasters to exploit. We should continue to search for other new sources of predictability too, and perhaps AI can help with that.

The second category, converting current forecasts into better outcomes, may be the greater opportunity for AI to accelerate decarbonisation but is harder to achieve. Forecast improvement only requires changes to forecasting systems. Changing the way forecasts are used often requires multiple business processes, policies and governance to evolve. Energy market participants and system operators have access to masses of forecast information; this is not the problem. Converting that information into decisions that minimise costs and carbon emissions (or maximise profit) while maintaining reliable supply is the real challenge. We shouldn’t overlook the potential role of AI tooling from chatbots to agents to accelerate deployment and maintenance of state-of-the-art forecasting and decision-support systems. The data and digital skills gap in the energy sector is a major constraint. Increasing the productivity of teams and individuals with the right know-how could have a significant impact. Small tech-focused companies may be well placed to exploit this opportunity, but regulated monopolies and large incumbent utilities are burdened with legacy IT and the necessary security measures that protect critical national infrastructure but hamper innovation.

From forecast improvement to forecast value

The variability and limited predictability of a wind and solar power mean that as these resources expand, short-term forecast uncertainty increases. For system operators and market participants (not just renewable energy producers) who buy/sell their anticipated consumption/production ahead of time, quantifying and managing forecast uncertainty is of increasing importance. Around the world, the energy sector has responded by allowing energy to be traded closer to delivery, in more granular time intervals, and through more diverse market mechanisms. Digitalisation of energy markets has given rise to automatic and algorithmic trading where computers process (forecast) information and execute trades faster than rival human traders, a well-trodden path in financial markets. Control room engineers are also aided by data-driven support systems and varying degrees of automation.

In both cases, managing risk is central and quantifying forecast uncertainty is vital. Uncertainty in energy forecasts is highly structured with implications for users. Imagine, for instance, a weather front crossing the UK that will bring a change in temperature, cloud cover and wind speed: supply, demand and the flow of energy around networks will all be affected. The impact of the front arriving earlier or later than anticipated affects the whole system in a complex but structured manner. Forecasts can describe this structure, but human users struggle to make sense of such vast and complex information. Creating efficient operational plans that include recourse to handle all credible future outcomes has always been the challenge facing energy system operators. In low-carbon energy systems, weather-dependence expands the range and complexity of possible futures, and uncertainty persists closer to real-time. Where automation is not possible, safe or otherwise acceptable, AI has potential to support human operators in myriad ways, including providing advice and insight, even identifying novel risks, and executing decisions that may require many individual actions to be taken.

Finally, we must recognise that the success or failure of these methods in supporting decarbonisation will depend as much if not more on development of the environment in which they operate as it will on advancements in AI itself. Market design, regulation, communication and control systems both constrain and enable AI’s role helping to manage energy systems. For example, demand-side flexibility and storage should alleviate the challenges brought by variable and uncertain production from renewables, but are widely viewed as a forecasting challenge. Mechanisms for matching supply and demand need to be reimagined for the low-carbon future we aspire to create.

Summary

Predictive analytics are already and will continue to improve the efficiency of energy markets and control rooms. AI in all its flavours will play a role in supporting decarbonisation, but realising its benefits will require structural change as part of the wider energy transition as well as adoption by individual actors.

12 March 2026

The role of our oceans and the potential of AI

This year, Sir David Attenborough said, “After almost 100 years on the planet, I now understand the most important place on Earth is not on land, but at sea.”  He followed that up with “If we save the sea, we save our world,”¹. For the sake of a sustainable planet, one that is dominated by our oceans, we need to use every tool we have to advance decarbonisation. This can be very direct methods of reducing the amount of emissions from shipping, to understanding/monitoring the health of seagrass that are critical for capturing carbon.  

Given how expansive our oceans are geographically, but also how expansive and complex the industries using our oceans are, one key tool in our toolbox is Artificial Intelligence (AI). With the vast amount of data we can collect and the need to study it quickly, it is critical that we are able to use AI to understand and address the challenges around decarbonisation of our oceans in a timely manner.  

That said, every solution often has a price. The cost of AI is still being studied, and a large part of that is the integrity and accuracy of its findings. This includes challenges like bias, hallucinations, but also intentional cyber-attacks. The other challenge is the carbon cost of AI itself.

A balanced approach to sustainable, secure marine AI

The synergy between the University of Plymouth and the data-centric engineering approach of the Alan Turing Institute’s Sustainability Mission (funded by Lloyd’s Register Foundation) is to conceive of AI solutions to save our oceans as a double-edged sword, looking at how it can both help and hurt. We are walking that fine line between driving development of AI for net-zero and identifying and mitigating the costs of those solutions more widely through raising awareness and research. 

In the last two years, I have looked at several areas where we can contribute to net-zero under the marine and maritime theme. These include but are not exclusive to offshore renewable energy and marine autonomy.

Early in the first year of my secondment, we asked what the risks of cyber-attacks are on remote offshore renewable wind farms. This resulted in a policy paper written with other Turing collaborators². In 2025, a workshop with key stakeholders was held at the opening of the University of Plymouth’s new wind network testbed that can generate valuable data to understand how AI can be used to support this type of renewable energy, as well as a testbed to try AI solutions to protect the cyber security of these turbines³.

With AI being increasingly used for marine autonomy, many businesses are focusing on developing the hardware and are often using off the shelf AI tools for functions like object recognition. Working with the National Cyber Security Centre, we were concerned about the lack of cyber awareness in the lifecycle of AI for autonomy, which are considered key solutions for reducing shipping emissions. From this, a few research projects spanning 2023-2025 looked at the cyber-security of AI for marine autonomy, and how we can improve cyber security at all stages of development^4,5.  

Looking ahead: inclusive futures for marine, maritime and AI

In addition to continuing with technical and policy research outlined above (plus other wonderful projects), in the following year for the marine and maritime theme, we hope to address some social challenges such as workforce diversity in marine/maritime/AI. The hope is to ensure everyone has equal voice in net-zero initiatives, and equal opportunity to heed Sir Attenborough’s call to save our oceans.

¹ At 99, David Attenborough shares strongest message for the ocean - oceanographic
² Enhancing the cyber resilience of offshore wind - The Alan Turing Institute
³ World-first facility enhances the UK’s energy and cyber resilience - University of Plymouth
⁴ SeXTANt (Secure X Trustworthy AI Navigation) - The Alan Turing Institute
⁵ SAIMAS – Secure AI within Marine Autonomy Systems - University of Plymouth

12 February 2026

Data that works for farmers

Agriculture accounts for around 10% of the UK’s total greenhouse gas emissions, driven mainly by methane from livestock and nitrous oxide from fertilisers. As regulators and retailers push to cut emissions, farmers face growing pressure to monitor and reduce their footprint, while staying profitable under increasingly challenging climate conditions. The challenge then is twofold: how do we measure emissions accurately, and how do we ensure farmers benefit from the process?

These two issues are naturally linked. Accurately monitoring emissions within complex living systems like soil and livestock is difficult and requires active farmer participation. Yet without a clear return on investment, why would farmers commit time and resources to data collection?

This is where AI can play a transformative role. When implemented effectively, AI tools can turn fragmented agricultural data into actionable insights, helping farmers improve efficiency and cut emissions. At the same time, these platforms can provide regulators and organisations with robust datasets to shape decarbonisation strategies and develop new incentive schemes to reward low-emission farming practices.

The challenge of measuring agricultural emissions

Accurate emissions data in agriculture is notoriously difficult to achieve. Farm emissions come from living systems where biological variability plays a huge role. Emissions are not only influenced by management practices, but by weather, soil microbiology, and other factors that can cause emissions to fluctuate daily.

Traditional sampling methods are expensive and time-consuming, and only provide snapshots of the whole picture. Meanwhile, other valuable data sources such as machinery logs, weather stations, soil sensors, and third-party services are often fragmented across multiple platforms and subscriptions. These sources generate vast amounts of valuable information, but they’re often siloed, hard to access, and even harder to integrate. Add on top of this concerns over who owns and benefits from the data gathered from their farm, and it’s not difficult to see why farmers hesitate to invest in systems that don’t deliver clear, tangible value to their business.

Turning data into action

AI however has the potential to integrate various data streams and uncover patterns, predict outcomes and provide suggestions that can help farmers achieve their productivity and sustainability goals.

We are already seeing examples of projects and agri-tech business piloting solutions to tackle agricultural emissions and improve efficiency. Advanced machine learning models are being used to predict yields, track carbon sequestration, model emissions, and simulate how changes in feed or management practices could influence both emissions and yield.

Meanwhile AI-enabled nutrient management tools are being developed to help growers maximise fertiliser use efficiency by integrating sensor data with weather forecasts and crop models to provide optimal fertiliser timing and rates.

By turning complex datasets into practical suggestions, these services enable farmers to improve efficiency, cut emissions, and maintain or even improve profitability. Crucially, they transform emissions data collection from a compliance burden into a strategy to drive productivity and resilience.

AI and carbon markets

While AI is starting to become available to help farmers improve efficiency and reduce emissions, it has the potential to go further and underpin carbon reward schemes. Initiatives, like the UK Farm Soil Carbon Code that is still under development, aim to lay the groundwork by developing clear protocols for tracking changes in soil carbon and greenhouse gas emissions, whilst funding mechanisms such as the Exchange Market Fund and retailer-led carbon insetting projects are exploring how to directly reward farmers for reducing emissions or sequestering carbon. Likewise, the proposed relaunch of the Sustainable Farming Incentive could again open a pathway for financial support to farmers for adopting low-emission practices.

AI has the potential to make these schemes more accessible and reliable by providing farm-specific reports and recommendations, not only alleviating the administrative burden but helping farmers make informed decisions to best access these incentives.

What AI platforms must deliver

Trust is difficult to gain and easy to lose, especially when people are tying their livelihoods to new technologies. To achieve widespread trust and adoption platforms must be designed with several key features that prioritise transparency and reliability:

• Validated methodologies and transparent algorithms, with traceable data sources aligned to recognised standards
• Detailed outputs formatted to comply with national and international carbon accounting standards, ensuring compatibility with third-party monitoring, reporting, and verification requirements
• Full auditability, with every data point is linked to its original source and every recommendation linked to its relevant data points
• Safeguards against manipulation or false reporting
• Exportable and interoperable data compatible with systems used by verifiers, regulators, and supply chain partners
• Continuous or high-frequency data collection to capture variability over time
• High-resolution data and recommendations that reflects real management practices and enables site-specific interventions
• Farmer control over how their data is used and which recommendations to follow
• Affordability and accessibility for farms of all sizes

From data to decarbonisation

Emerging AI platforms can enable farmers to reduce emissions while maintaining or improving profitability. But with thoughtful development and integration, AI platforms could become the foundation for how the agricultural sector monitors, reports, and acts on its environmental impact. By connecting fragmented data sources, automating complex measurements, and providing practical insights, these systems could support both farm-level decision-making and supply chain-wide decarbonisation. To realise this potential, stakeholders across industry must invest and collaborate to develop and demonstrate trustworthy, scalable AI solutions that empower farmers and drive meaningful climate action.

Sean Lockie, Associate Director at Arup

Steel, cement and chemicals form the backbone of modern economies, but they also account for roughly 25% of global carbon emissions. These sectors are notoriously hard to abate as their processes rely on extreme heat and carbon intensive reactions. As climate targets tighten and carbon pricing escalates, the race to decarbonise these industries is no longer optional - it’s essential. Enter Artificial Intelligence (AI): a transformative force accelerating the shift to low carbon production.

Why these sectors are so challenging

Steel production alone contributes about 7% of global CO₂ emissions, primarily from blast furnaces and coking operations. Cement adds another 8%, driven by the calcination of limestone and fossil fuel combustion in kilns. Chemicals, including ammonia and plastics, emit vast amounts of CO₂ and nitrous oxide during synthesis processes. Traditional mitigation strategies such as alternative fuels, and electrification remain vital but cannot deliver the scale of change required without systemic efficiency gains. This is where AI and circular economy principles can accelerate progress.

AI as an optimisation engine

AI is already helping the industry make smarter decisions. In steelmaking, advanced algorithms analyse sensor data from blast furnaces, adjusting temperature, pressure and raw material mix to minimise energy use and emissions. Predictive maintenance reduces downtime and prevents energy waste, while computer vision monitors particle sizes for optimal combustion.

Similar gains are seen in cement, where AI-driven kiln optimisation and digital twins - a virtual representation of the kilns - fine tune fuel mix and heat distribution, cutting fuel-derived emissions by up to 5% per plant annually. Machine learning can also be used to optimise concrete mix designs, delivering lower carbon outcomes.

While AI excels at optimisation, designing fundamentally new processes, materials and business models still requires human creativity and interdisciplinary expertise. That’s where circular thinking becomes critical.

Circular economy: Beyond reduce, reuse, recycle

Decarbonisation isn’t just about cleaner production - it’s about rethinking the entire lifecycle. The “9 R’s” framework reminds us that redesign is as important as reuse. For steel and cement, this means designing for disassembly, enabling components to be recovered and repurposed rather than demolished. Today, dismantling is blunt and labour intensive, but tomorrow, robotics and machine learning could automate sorting and recovery. Materials passports embedded in building information models (BIM) and digital twins will allow precise tracking of what’s in our buildings, which is essential for urban mining and reuse.

Arup is already exploring circular approaches through urban mining pilots in Europe, mapping building technologies to identify concrete suitable for recycling. The consultancy is also driving standardisation of building elements like glass partitions and ceiling tiles, which can further unlock reuse potential. As part of one of its many research investment programmes, Arup is currently developing DeepEnergy, an AI-enabled framework which will help clients plan decarbonisation across entire portfolios, even when operational data is incomplete.

Additionally, industrial symbiosis - sharing waste heat between facilities - is another opportunity, already being trialled in eco-industrial parks such as Kalundbord in Denmark.

Mindset matters

AI is a powerful enabler, but it works best as part of the broader strategy. It can optimise processes, predict outcomes, and accelerate progress, but lasting impact depends on collaboration across industry, academia and policy. Success also requires openness to new materials, new design approaches and new data models, especially in emerging markets where emissions data is scarce. A prime example is the Global Whole Life Carbon Tracking (GWLCT) project, funded by the UK government Department for Energy Security and Net-Zero (DESNZ). The initiative is developing global carbon tracking methods and tools, providing data and insights to accelerate decarbonisation efforts, especially in developing economies, and highlighted opportunities for lower carbon construction through analysis across the world. Part of the method involves using earth observation and machine learning to track the changes in built environment areas.

Bottom line

Decarbonising steel, cement and chemicals demands both AI and circular thinking. AI helps us move faster and make better decisions, helping these carbon-rich industries cut emissions at scale. But real transformation comes though combining these digital capabilities with redesign, reuse and systemic change. Together, these approaches can turn ambition into action and shape a built environment that supports a net zero future.

The need for global Artificial Intelligence (AI) guardrails was established in May 2023, when G7 Leaders identified priority topics in the ‘Hiroshima Artificial Intelligence Process’. As part of the stocktaking of opportunities and challenges related to generative AI, the subsequent OECD report highlighted a range of common policy priorities. Among those priorities considered urgent and important the ‘responsible use’ of generative AI technologies was widely viewed as the most “urgent” and most “important” for global policy. Responsible use of AI may take multiple forms. In the United Kingdom (UK), for instance, risk mitigation is at the heart of the International Scientific Report on the Safety of Advanced AI. Here, risk from general purpose AI is categorised into three areas: Malicious use risks, risks from malfunctions, and systemic risks. Alongside these critical guardrails and areas for risk mitigation is the essential need to recognise and address the environmental implications of AI by placing ‘People & Planet’ at the heart of a responsible AI lifecycle management framework.

AI Compute and the Environment

What is ‘compute’? AI systems are dependent on computing resources that are commonly referred to as compute resources. The OCED.AI policy observatory define AI compute as comprising “one or more stacks of hardware and software used to support specialised AI workloads and applications in an efficient manner”; and as AINOW outline can “include chips; software to enable the use of specialized chips like GPUs; domain-specific languages that can be optimized for machine learning; data management software; and infrastructure in data centers.”

Integrating sustainability considerations into AI guardrails is crucial given the huge amount of compute resources AI systems demand. To unpack this, we zoom in on the compute needs of large-scale machine learning (ML) based AI systems. Drawing insights from the OECD digital economy paper—A Blueprint For Building National Compute Capacity For Artificial Intelligence—there are two compute-intensive steps in the development and use of AI systems: (1) training, meaning the creation or selection of models/algorithms and their calibration, and (2) inferencing, the process that a trained ML model uses to draw conclusions from brand-new data.

There is a huge need and demand for compute resources for ML based AI system training and inference. Subsequently compute availability is considered both scarce and a key bottleneck for AI development and deployment, according to a recent article in AINOW. Indeed, the UK’s National AI Strategy recognises that the compute capacity for AI must be increased to meet future needs, as outlined by The Alan Turing Institute.

The demand for AI compute resources is increasingly rapidly as the sophistication of AI systems has evolved, particularly for deep learning applications.

AI Compute Awareness and Understanding

In their February 2023 Digital Economy Paper, the OECD report how a large cloud compute provider “estimates that its enterprise customers spend 7-10% of their total compute infrastructure expenditure on supporting AI and ML applications, broken down to 3-4.5% for training and 4-4.5% for inference. This includes about 60% spent on compute platforms featuring hardware accelerators like GPUs and about 40% spent on CPU-based compute platforms”. This illustration provides one indication as to the compute resource commitments across training and inferencing in practice.

However, these insights are largely anecdotal, highlighting a significant challenge in assessing the energy consumption and environmental impact of AI and, by extension, its sustainability. A key difficulty is distinguishing when computational resources are specifically dedicated to AI systems.

Compute resources can be general-purpose, serving both AI and non-AI workloads, or specifically tailored for AI, yet this distinction is often not clearly made. This issue is further complicated by the lack of standardized and validated data on AI-related compute, as emphasized by the OECD.AI expert group on AI Compute and Climate. The expert group highlight a range of related challenges when it comes to capturing compute resource needs and commitments; the group’s Insights from preliminary survey results revealed the following:

• 31% reported that they do not measure how much AI compute they have.
• 20% reported that they did not know whether they measure AI compute.
• 52% of respondents reported challenges accessing sufficient AI compute.

When asked about the percentage of their organisation’s total annual costs spent on AI compute:

• 37% reported that they did not know.
• 5% reported no annual costs spent on AI compute.
• 26% reported 10-40% of costs.
• 3% reported that AI compute costs were 50% or more of annual costs.

Though the findings serve only as an illustration given the limited sample size, it is clear that issues of transparency, understanding, and knowledge pertaining to AI compute resources are real. This presents significant challenges in attempts to capture the real energy and fiscal costs of AI compute its subsequent impact on the environment. This measurability issue is further complicated by the lack of an agreed standard on how to measure AI-related compute. Consequently, there is an inability to forecast the environmental impact of AI, as well as data more broadly, leading to significant debate as to how we can obtain an approximate average figure for one unit of data.

Does the Type of AI Model Matter Impact Electricity Consumption?

Luccioni et al. (2024) highlight the energy requirements of different AI models in their recent paper in the Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency. They report on the energy consumption per 1000 inferences of three model types: Classification tasks including images and text range between 0.002 to 0.007 kWh, Generative tasks including text generation and summarization require around 0.05 kWh; and, Multimodal tasks such as image captioning and image generation) range between 0.06 kWh to 2.9 kWh. Within task-specific models, there is significant variation between image-based tasks and text-based tasks in energy consumption intensity; as reported by the authors, image-based tasks are far more carbon intensive at 100g of 𝐶𝑂2𝑒 per 1,000 inferences compared with 0.6g per 1,000 inferences for text-based tasks. So, how AI models are deployed does matter for energy consumption and the environmental impact of AI.

Takeaways for Responsible AI Management

Undoubtedly, AI has great potential for fighting climate change but the impact of AI on GHG emissions can also be negative. Projections suggest that electricity demand from components of the AI infrastructure are set to increase – electricity demand from data centres are expected to more than double by 2030 – and so the responsible use of AI becomes critical. We draw out three aspects of responsible AI management below:

(1) The environmental impact of training AI models is often not reported and yet as Luccioni and colleagues explain, can be “orders of magnitude more energy- and carbon intensive than inference”. One recent observation suggests compute has increased by approximately 4x per year since 2010 for training general purpose AI systems and 2.5x in training dataset size. As outlined in the digital decarbonisation article—Three steps for businesses to make AI data and compute more sustainable—embracing responsible data management practices can play a pivotal role in minimising unnecessary data storage and reducing the environmental impact of AI. AI models often demand extensive datasets for training, and inefficient data storage practices can result in heightened energy consumption and environmental strain. Though the number of inferences made can subsequently result in total costs of greater than the total cost of training, the environmental impact of training must be acknowledged and addressed. By adopting data management strategies that prioritise data minimisation, efficient storage, and responsible data disposal, we can significantly decrease the ecological footprint of AI.

(2) How we engage with AI is crucial to ensuring its responsible use, especially given that, as Luccioni and colleagues highlight, using multi-purpose models for specific tasks is more energy-intensive than employing task-specific models. Therefore, as emphasized by the digital decarbonization design group at Loughborough University, a key step before deploying any AI system is to critically assess the level of AI truly required. While it may be tempting to adopt the latest AI technology, it is essential to evaluate whether these advancements genuinely enhance our decision-making processes or if simpler, less resource-intensive models might suffice. Sustainable AI involves striking a balance between leveraging cutting-edge technologies and opting for more environmentally sustainable models (e.g. symbolic AI). If adopting multi-purpose AI technologies, considerations ought to be given to how such models are, or have been, trained. For instance, open-source AI training datasets may help to reduce duplication and limit the demand for new data acquisition and annotation, both of which carry environmental and economic costs. However, this relationship is still under-explored and warrants further investigation. Though currently, open-source training language models are predominantly in English (57%), potentially creating language divides in AI datasets that must be addressed.

(3) Other practical steps to engage more responsibly with Generative AI and reduce its energy footprint are outlined in the recent UNESCO report—Smarter, Smaller, Stronger: Resource-Efficient Generative AI & the Future of Digital Transformation. Suggested actions to reduce the environmental impact from the structure, size, and flow of digital interactions that occur from Generative AI use include:

• For tasks that are specialized and repetitive (e.g. translation or summarization) use smaller, task-specific models rather than large general-purpose models – doing so can result in energy reductions of up to 90%, while maintaining strong performance (pp. 4-5).
• By shortening AI-generated responses from 300 to 150 words, users can reportedly reduce energy consumption by over 50% per query (p. 18)
• Reducing the size of models (via compression techniques like quantization) and computational complexity may result in energy savings of up to 44% (p. 7).
• Combining the actions of quantization and reduced response lengths, energy expenditure can be reduced by up to 75%, “equating to approximately 30,000 U.K. households per day” (p. 20).

Electricity networks and the changing landscape

Electrical power systems worldwide are going through unprecedented changes motivated to a large extent by decarbonisation targets but also for various other technical, economic and social reasons. These changes affect most aspects of power system operation and planning: starting from generating electricity from primary energy sources, transmitting it over long distances, and finally distributing to consumers to be used.

The generation mix is changing drastically with more renewable generators being connected (e.g. onshore and offshore wind and solar) and conventional thermal units (e.g. coal) being decommissioned. The way we use energy is also changing in an effort to reduce carbon emissions in other sectors, by moving away from fossil fuels and electrifying parts of transport through Electric Vehicles (EVs), and heating through heat pumps. At the same time, due to the complex nature of power systems, being large complex nonlinear dynamical systems, electricity generation and demand needs to be continuously balanced, for the system to be stable. And also, we all expect very high reliability to meet our energy needs on demand.

In the UK, there are ambitious targets set to achieve net zero and some of the Future Energy scenarios of the National Energy System Operator (NESO) involve 50GW of offshore wind by 2030 and see up to 600,000 electric heat pumps/year by 2028 as well as the ban of petrol/diesel cars by 2035 with EVs being an alternative.

In a nutshell, the above changes cause increasing uncertainty and complexity in power system operation and planning. Increasing uncertainty largely comes due to the intermittent nature of renewable generation but also due to social dimensions related to how people use energy. For example, how will people (expect to) use and charge their EVs? In addition, market structures can also drive such behaviours. On the other hand, increasing complexity comes mainly due to the very different behaviour and technologies that these new devices use (interfaced through power electronics), governed to a large extent by complex control.

We also have more available measurement devices in modern power systems that can monitor the system and the demand in timescales from milliseconds to minutes/hours, providing valuable data. However, measurements can still be scarce, and new algorithms are needed to extract meaningful information from data, due to the complexity of the system. This is where AI can be helpful.

Importance of maintaining security

In this new paradigm, it is becoming increasingly important to ensure power system security, especially when considering the changing system dynamics which can introduce unknown modes of failure. Not only because power systems are part of critical infrastructure with provision of several other services relying on them (e.g. communications, water and gas supply, transport, etc.), but also due to increasing reliance on electricity as we are decarbonising energy (e.g. transportation and heating). Failing to do that can lead to power disruptions or in the worst-case blackouts, which can lead to huge economic costs, significant disruption to society and even loss of human life.

What it takes to run a power system

During this energy transition it is crucial to keep the pace of integrating low carbon technologies at the scale required but also to ensure we utilise them in a stable manner in operational timescales.

Technically speaking, planning (from days to years) and operation (from milliseconds to hours) of power systems requires solving a number of challenging problems related to optimisation, forecasting, and security assessment (i.e. checking against a number of external disturbances to ensure the system will remain stable), among others.

We need to forecast wind and solar generation as well as the demand in advance, to ensure adequate generation is dispatched in real time in a cost-efficient/optimal manner. And all this while ensuring the stability/security of the system.

How AI and data-driven methods can help

To this end, advances in Artificial Intelligence (AI) and more specifically Machine Learning (ML) domain are opening up avenues and possibilities for improving the above-mentioned aspects, including optimisation and security assessment. AI can also be particularly useful in utilising data and offer fast computation capabilities when dealing with large complex systems.

Notably AI is considered a state-of-the-art tool for forecasting of renewable generation which is crucial in fully utilising low carbon technologies and ensure the system remains balanced. It is also being considered as helpful for other applications such as predictive maintenance, visual inspection, reporting assistance, alarm management, and so on.

Finally, AI can be helpful in utilising data to improve situational awareness, provide decision support to system operators, and also enable automation, especially for cases when the timescales or complexity does not allow human interventions, e.g. in the millisecond timescales or when there is need to control millions of devices.

All these aspects are crucial in enabling the implementation of low/zero carbon technologies at the pace needed, as well as their efficient and secure utilisation in operational times.

Moving forward with AI for electricity networks

Due to the nature of the application to critical infrastructure, it is important to build trust in AI technologies and ensure implementation with proper failsafe mechanisms when considering implementations to electrical power systems. Data availability, quality and considerations around the need for exchange are also important to be kept in mind. Having said that, there is a very diverse range of potential applications around power systems which might have very different requirements in the above aspects, some stricter, some less so.

The Generation Gap in the Built Environment

Much like the generational divide in human societies, the built environment is split between old, inefficient buildings and the fast-changing AI-driven solutions that promise to decarbonize them. In the UK, nearly 80% of buildings were constructed before 1980. These buildings—many still reliant on gas boilers, outdated insulation, and inefficient heating systems—are now being asked to adapt to a new world of net-zero targets and renewable energy.

But change is never easy, especially for those set in their ways. For AI solutions to succeed, it must act as more than just a technical fix—it must embody the qualities that make intergenerational relationships work: communication, empathy, trust, responsibility, and perhaps even some playful interactions!

Communication: Speak to Aging Buildings

AI solutions face fundamental language barriers with older buildings. Think GenX lingo versus the Boomers! Where most AI solutions are being built atop the expectation of easy access to IoT sensors, automation, and cloud connectivity, pre-1980 structures operate on analogue systems, manual controls, and deeply entrenched operational habits.

To bridge this gap, AI solutions needs to act as translators between the old and new. We need non-invasive and accessible techniques to map energy performance and identify inefficiencies that are specific for a given building. We need techniques to ‘listen’ to older Building Management Systems (BMS) — decoding their thermal behaviour, energy demand cycles, and maintenance needs — before proposing radical changes. At the same time, we urgently need a renewal of facilities management as a profession that aligns with current technical advances in AI and machine learning.

Frugality please!

We are at the cusp of change. Generative AI is transforming our technologies, our decision-models, and our energy systems. This transformation is backed by a vastly improved landscape of data. Yet, data also comes with a carbon cost. Thus, it is not just in the interest of aging buildings that we need AI solutions that are data frugal, but also for the wider benefit of decarbonisation.

Empathy: Value Legacy

Resistance to change is common, not because progress is rejected, but because of disruption, loss of control, or unfamiliarity. The same applies to aging buildings. Replacing gas boilers with heat pumps, integrating smart meters, changing windows, or shifting to AI-driven demand response systems might make technical sense, but unless the process respects the existing structure, these changes will face pushback from planners, owners, and occupants. Older buildings hold architectural and societal value. AI technologies need to be aligned to aesthetic values and at the same time challenge outdated and prescriptive planning constraints.

We have some incredibly successful examples of architectural retrofits where the old and the new co-exist harmoniously, eg. the British Museum, St. Pancras International to name just two. Surely, we can harness the same level of creativity in bringing energy efficient solutions to the aging building stock.

Trust: Building Confidence in AI’s Decisions

Intergenerational relationships thrive on trust, which is earned over time. AI, too, must gain the trust of building managers and occupants by demonstrating transparency, reliability, and security. AI based solutions must present its recommendations in clear, actionable terms. For example, instead of saying, “Optimising Heating Ventilation and Air Conditioning (HVAC) system loads based on dynamic energy pricing,” AI could communicate, “Reducing heating at night can lower energy bills without affecting comfort.” Furthermore, trust grows when results are tangible. AI-powered energy optimisations should be implemented in measurable phases, allowing occupants to see real-world benefits, such as lower energy bills and improved comfort.

Playful discussions: Making Sustainability Engaging

One of the most underrated aspects of learning is playfulness, or gamified experiences. AI, too, can introduce an element of play in the process of decarbonising buildings, through adaptive and responsive interactions. Giving AI-driven building systems a relatable interface can make interactions more engaging, fostering a sense of connection between the building and their operators.

The UK manufacturing sector is undergoing a major transformation, striving to balance economic growth with sustainability. At the forefront of this shift is artificial intelligence (AI), a powerful tool helping manufacturers reduce their carbon footprint while maintaining efficiency. By enabling smarter operations, reducing waste, and optimising energy use, AI is helping the industry lower its carbon footprint while maintaining productivity.

AI-powered energy efficiency

One of the primary ways AI contributes to decarbonisation is through optimising energy use. AI-driven systems can monitor energy consumption in real-time, identify inefficiencies, and suggest corrective actions. Machine learning algorithms, for instance, can predict energy demand and adjust machinery operations accordingly, reducing unnecessary energy use. This not only cuts emissions but also lowers costs—a win-win for manufacturers striving to meet sustainability goals.

Many UK smart factories already use AI to integrate renewable energy sources like solar and wind into their operations.. By predicting energy supply fluctuations and adjusting production schedules, these factories maximise the use of clean energy while minimising reliance on fossil fuels.

Cutting waste with AI precision

Material waste is another significant contributor to carbon emissions in manufacturing. AI helps address this issue by improving precision and reducing overproduction. Advanced predictive analytics and computer vision technologies enable manufacturers to optimise material use, ensuring that every resource is used efficiently.

For example, in additive manufacturing (3D printing), AI algorithms can simulate and optimise designs before production begins, minimising errors and material waste. Similarly, in industries like automotive and aerospace, AI-powered quality control systems can detect defects early, reducing the need for rework and scrap—processes that are both energy- and material-intensive.

Greener supply chains with AI

AI is not just transforming what happens within factories; it’s also revolutionising supply chains. Manufacturing supply chains are complex, often spanning multiple countries and involving significant transportation emissions. AI tools can optimise logistics by identifying the most efficient routes, consolidating shipments, and reducing idle times for vehicles. These measures collectively lower fuel consumption and emissions.

AI also improves supply chain transparency, tracking and analysing the carbon footprint of materials from sourcing to final production. This enables manufacturers to make more sustainable sourcing decisions and align with the UK’s broader net-zero targets.

Predictive maintenance and prolonging equipment life

Predictive maintenance is another area where AI is making a difference. Traditional maintenance schedules often lead to unnecessary part replacements and energy use. AI-driven predictive maintenance uses sensors and machine learning to identify when equipment is likely to fail, ensuring repairs are made only when necessary. This reduces downtime, extends equipment lifespan, and minimises the environmental impact of manufacturing new parts.

Supporting decarbonisation goals

The UK government has set ambitious targets to achieve net-zero greenhouse gas emissions by 2050, and manufacturing—which accounts for around 16% of the UK’s carbon emissions—is a critical area for action. AI is not just a technological advancement—it’s a key driver in the UK’s journey to net-zero. As industries adopt AI-driven solutions, they can accelerate sustainability efforts while maintaining competitiveness.

While challenges remain, such as the need for skilled workers to implement and manage AI systems, the potential benefits are undeniable. By embracing AI, UK manufacturers can lead the way in sustainable innovation, setting an example for industries worldwide.

AI is proving to be a powerful ally in the fight against climate change. From optimising energy use to reducing waste and enhancing supply chain efficiency, its applications are driving significant progress towards created economic, social and environmental value for the UK’s manufacturing sector. As adoption continues to grow, AI will undoubtedly play an even greater role in creating a greener, more sustainable future for the industry.

There is an increasing immediacy in our discussion around climate change. Many experts expect that keeping global temperatures below +1.5°C is impossible^1,2, with evidence showing that there is a 1 in 10 chance of going over +2°C if we get to +1.6°C³. Many countries are already starting to see changes in agriculture and severe weather conditions that are causing increased damage to urban areas⁴. Recent discussions in the past few weeks have increasingly focused on the cost of climate change⁵. As we move closer to +1.5°C, we will see more severe weather effects and the cost of protecting ourselves increases. This will eat into the budgets that governments have set aside to fight climate change, reducing our chances of limiting temperature increases.

Despite this immediacy, much of the focus is on technologies that are slow burners. We seem to focus on technology solutions that are net zero, but targeted at 2040, or solutions that will remove CO₂ from the atmosphere, risking overshoot⁶. It’s clear we need to start reducing emissions now, not just waiting for net zero solutions to become available.

The role of data

Data can play a part. Across transport, construction and agriculture solutions are available that can help to improve productivity and/or reduce emissions. They are often relatively cheap (or at least cheap compared to technologies that require greater transformation of a sector). Importantly they can be installed by individuals, rather than waiting for consensus amongst industries and time-consuming infrastructure and regulation changes. Many of these approaches can be applied immediately.

For example, many of our friends and family changed their driving habits when they had MPG calculations available in their vehicles, often shocked at how small changes can lead to relatively big savings, with more efficient driving showing the ability to reduce emissions by 15-20%⁷. The same sort of approach can be applied in shipping, showing a theoretical saving of 18% using a combination of data based green tech that is already on the market⁸. Just by understanding our systems better and operating more efficiently. With Very Low Sulphur Fuel Oil (VLSFO) bunkering in Singapore at $567.50 per metric ton⁹, there is an economic as well as social driver.

Immediate change is needed

However, we still see relatively limited uptake of these approaches. But we can’t continue to wait. Data approaches allow immediate reductions in emissions and need to be part of business operations. When applied effectively they save money in the short term, allowing a bigger pot for future investment while lowering the cost of implementing net zero products. Immediate change is needed to reduce climate shocks and give us the best chance to protect people. If you haven’t implemented these approaches, what’s holding you back?

¹ Can we still avoid 1.5 degrees C of global warming? » Yale Climate Connections ² 2024 first year to pass 1.5C global warming limit ³ Once we pass 1.5°C of global warming, there is no going back | New Scientist ⁴ What a year of freak weather tells us about climate change, and what comes next ⁵ A ‘doom loop’ of climate change and geopolitical instability is beginning ⁶ ‘Overshoot myth’ risks catastrophic global warming · GreenFutures ⁷ Climate explained does your driving speed make any difference to your cars emissions ⁸ Greater than the sum of their parts: merging green technologies ⁹ Singapore Bunker Prices

Sam Young, AI Practice Manager, Energy Systems Catapult

Organisations in every sector are embracing AI to improve efficiency and reduce operating costs. If we’re aiming for decarbonisation then that misses the point.

Operating costs are seldom the sticking point holding us back from accelerating to Net Zero. High operating costs aren’t why the UK is failing to meet its heat pump installation targets. High operating costs aren’t why it takes over a decade to connect a wind farm to the electricity network. High operating costs aren’t why it is costing us more and more to balance the electricity network. AI can be used to accelerate us to Net Zero, but we need to apply it to the things that are actually slowing us down.

There are three big sources of friction that AI can really help with: delays, information gaps, and uncertainty about the future.

Reducing delays with AI

Delays plague many of the processes that are core to decarbonising our energy system, from installing heat pumps to connecting wind farms to the network. Eliminating these delays would get us to Net Zero sooner.

Reducing or eliminating delays can have important effects on the whole system. It allows quicker feedback, so that things that work well are adopted faster and less time is wasted pursuing dead ends. It can change how willing people are to do things – people won’t wait for a heat pump in a few weeks if they can have a gas boiler today. And sometimes it can underpin completely new ways of doing things – imagine how hard remote working would be if we were still relying on snail mail.

We need to be asking: where can AI help make things that take weeks or days take only minutes?

How do we actually do that? Look at the processes involved in decarbonising our energy system. Talk to the people who live and breathe those processes. Examine all the delays and imagine what step changes could occur if some of those were eliminated. And then use AI’s delivery of efficiency to eliminate those delays.

Overcoming information gaps with AI

Decisions rely on information. We have many decisions to make in the transition to Net Zero, and often key information is missing. Which houses would benefit most from insulation? Where are people going to charge their electric vehicles (EVs)? When will an industrial site be connected to a hydrogen network? We can make better decisions about future investment if we know the answers to those questions.

Now AI can’t answer many of those questions directly. In fact, traditional economic modelling is often a better tool for long term planning. But AI can help in two ways: new sources of information and more accurate short-term predictions.

Often records about buildings and assets are incomplete and outdated, and updating those manually is incredibly time consuming and expensive. AI offers the ability to use less traditional data sources like satellite imagery, LiDAR or thermal imaging to gather data quickly and at scale. For example, if you don’t have a database of existing rooftop solar in an area you can use AI to pick them out of satellite images.

If it’s too slow and expensive to collect the data you need manually, ask “could AI help gather this data at scale?” The data won’t be perfect (no data ever is), but if we need it to make good decisions quickly, then AI can help.

Predicting the short-term future with AI

Whilst information to improve long-term decisions is important, accurately predicting the short-term is also key to decarbonising our energy system. In an increasingly decarbonised, decentralised energy system, balancing supply and demand on a minute-by-minute basis is getting more difficult – and more costly. The more accurately we can predict the state of the system over the next few hours and days, the easier and cheaper it is for the system operator to keep it stable.

But it’s not only the system operator who depends on good short-term forecasts. The business models of the flexibility providers that our system will increasingly rely on – from battery storage to demand response and smart EV charging – are fundamentally dependent on accurate short-term forecasts. The better at forecasting they are, the more easily they can make money providing services to the grid.

We need to look for opportunities to use AI to improve short-term predictions in ways that support a decarbonised energy system.

AI has the potential to be a powerful tool for accelerating decarbonisation, but only if we apply it to what is slowing us down. We should be focused on how it could eliminate important delays, accelerate access to information that is critical for investment decisions, and improve the short-term predictions the sector relies on.

So the next time someone pitches an exciting AI use case to you, stop and ask yourself: is this really going to accelerate us to Net Zero?

If you need more inspiration, I talk more on the topic here, or you can have a look at our AI for decarbonisation challenge cards and our webinars on AI for Flexibility or AI for Home Decarbonisation.