The energy system is complex and evolving. It is becoming increasingly electrified, digitalised, connected and decentralised, with mounting cost pressures. These drivers have encouraged energy companies to deploy applications that utilise artificial intelligence to optimise systems, improve production, reduce costs, raise efficiency, improve uptime, cut emissions and enhance safety.

Many of the desired goals of AI’s application in the energy sector – such as cost reductions, enhanced reliability and improved resilience – are challenging to quantify at a broader sectoral level, beyond the confines of individual case studies. It is also challenging to predict the nature, adoption and impact of AI applications that might emerge in the future. Given these limitations, we explore the impact of impact that known AI applications at the sectoral level by 2035 through a Widespread Adoption Case. This case considers only existing AI-led interventions informed by real-world case studies that can be scaled to the sectoral level. It assumes that many of the existing barriers to the sector-wide adoption of these AI-led interventions are overcome. But it stops short of considering the full theoretical potential of AI-led interventions, as it factors in certain insurmountable structural issues that would block their complete adoption. For example, we consider variations in adoption by region by factoring in the availability of enabling digital infrastructure. 

Oil and gas companies have been among the earliest adopters of new technologies to boost exploration and production. In 2000, 11 supercomputers operated by oil and gas companies ranked among the world’s 500 fastest. By 2024, this number had increased to 24, and total computing capacity has grown at almost 70% annually, outpacing the broader supercomputing industry. AI has various applications in the sector, including for subsurface data processing, reservoir simulation, remote operations, predictive maintenance, regulatory compliance, leak detection and repaid automation.

AI has various applications in electricity systems owing to the complexity of supply, transmission and demand profiles. In the Widespread Adoption Case, the application of AI in power plant operations and maintenance yields potential cost savings of up to USD 110 billion annually by 2035 from avoided fuels and lower costs. AI also enables greater integration of renewable electricity into the grid. Our analysis finds that up to 175 GW of additional transmission capacity could be unlocked in existing lines with the use of AI. 

The applications of AI in end-use sectors are varied but have significant potential. In industry, AI is being used in various ways to optimise production processes. In the Widespread Adoption Case, energy savings of 8% could be achieved by 2035 in light industry, such as the manufacturing of electronics or machinery. AI in transport can enhance vehicle operation and management, which could reduce energy consumption; it also has applications in reducing contrails and improving electric vehicle ranges. In buildings, the potential is limited by the rate of digitalisation, but there are compelling illustrations of impact, such as on efficiency and demand response.

Additionally, advanced economies currently have a competitive advantage in many of the technologies required for digitalisation and automation. Across three core segments – industrial automation, industrial software and robotics – the vast majority of leading global companies are headquartered in advanced economies. Europe is leading on automation and North America on industrial software, with each having more than half the market share in their respective segments. Asia is clearly leading on robots, hosting around two-thirds of the top 40 companies by market share.

Additionally, accurate weather forecasts and analysis of changing weather patterns in a warming world are essential to optimise the operation, planning and resilience of energy systems. AI has been improving the accuracy of weather forecasts and also reducing computational demand.

The adoption of such AI applications at a sector-wide level, however, is not a given. Various barriers are limiting the extent to which existing AI applications can be implemented, hindering the pace of change. These include unfavourable regulation, lack of access to data, inaccessibility, interoperability concerns, critical gaps in skills, the paucity of digital infrastructure and, in some cases, a general resistance to change.

Innovation is essential to achieving secure, affordable and sustainable energy. The energy sector continues to innovate: from 2010 to 2024, driven by technology growth and lower costs, unconventional oil and gas went from 10% to 25% of global oil and gas supply; solar photovoltaic (PV) went from 30 terawatt hours (TWh) of annual generation to around 2 000 TWh; and electric cars went from 0.01% to over 20% of global sales.

Innovation takes time. For energy technologies ranging from internal combustion engines and air conditioning to lithium-ion batteries and solar PV, time from invention to first commercialisation averaged over 30 years, and mass market uptake 20 further years. The core technology of today’s artificial intelligence, the artificial neural network, took 35 years to progress from prototype to first commercialisation.

AI, which can help speed up this process, is increasingly central to innovation pipelines. In medicine, AI led to a 45 000-fold acceleration in the scientific rate of discovery of the three-dimensional structures of proteins, the functional building blocks of human cells.

Patent and start-up data suggest that AI-first approaches to innovation are under-represented in the energy sector. Around 1% of energy-related patents reference the use of AI as part of the patented innovation; this share is similar across fossil fuels and clean energy. Only 2.3% of energy start-ups have an AI-related value proposition, lower than the 7% for life sciences and 4.3% for agriculture.

However, many areas of energy innovation are characterised by the kinds of problems AI is good at solving: highly complex design spaces, the need to balance performance trade-offs for an optimal outcome and rich datasets. For example, the discovery of a perovskite that is stable and easy to manufacture could accelerate cheaper and less space-intensive solar PV, and yet less than 0.01% of possible perovskite materials have been experimentally produced. AI could dramatically accelerate this process.

A core challenge of energy innovation is integrating new innovations into complex products and new products into industrial-scale supply chains. AI can help here, too. A battery gigafactory can produce up to 10 billion data points per day. Analysing these with AI models can help to detect faults, predict performance and diagnose problems, reducing the risks, costs and timelines for innovative chemistries.