SECTOR FOCUS: INFRASTRUCTURE MAINTENANCE SOFTWARE

image credits: Unsplash

A well-functioning, modern infrastructure is central to economic development and to quality of life. From the roads and railways needed to transport people and goods, to the power plants and communications networks that underpin economic and household activity, to the basic human need for clean water and sanitation, infrastructure matters. The annual requirement for infrastructure investment is evaluated at €3.4tn[1] until 2040 i.e. 3.5% of world GDP.

Maintenance of infrastructure assets is paramount to ensure their continued functionality, safety, reliability, longevity, and low environmental impact. For instance, the global road maintenance market is evaluated at c. €14bn[2] in 2022, to grow to c. €23bn in 2030 (5.1% CAGR 2022-2030).

Thanks to the recent investment increase in satellite data[3], drone data collection[4] and data computing[5], infrastructure maintenance use cases powered by Artificial Intelligence are getting more affordable. Consequently, 27% of the global infrastructure companies are planning to invest in AI for operations optimizations and 45% of them have already started to and plan to expand it[6]. The technological framework is usually the following: various data sources (technical documents, visual, radar or thermal images taken by humans, drones or satellites, etc.) are centralized to build and update a digital twin[7] of the asset.

This digital representation of the physical world is regularly analyzed by Artificial Intelligence tools to address an operational need. Here are some use cases:

• Detect equipment evolution that requires preventive or curative maintenance action. For instance, eSmartSystems, unusuals, Hitachi and Alteia help utilities to analyze the grid. They start by screening thousands of images with computer vision algorithms[8], before realizing a human check on the ones where an anomaly is detected.
• Prevent power outages or road accidents caused by surrounding vegetation. For example, Spotlight utilizes satellite images to assist Norscut (a Meridiam portfolio company) and Egis in anticipating trimming needs around the road. This is particularly important for utilities managing grid crossing forests, such as Florida Power & Light[9]. Some technological solutions can even provide automated operational advice, such as the number of woodcutters required to cut the estimated volume of timber. Accenture reports that for most utilities, vegetation management is their first item on annual budget ; and that US utilities spend $[6-8]bn annually on clearing vegetation from overhead lines[10].
• Anticipate, detect and fight wildfire through thermal and visual satellite data. Indeed, wildfires are getting more frequent and more difficult to prevent because of climate change[11]. Greece is addressing this growing challenge at scale with OroraTech[12] for instance. It is also one of Breezometer’s numerous use cases, an Israelian startup bought by Google in 2022[13].
• Follow a construction site and potentially adapt its design, comparing the plan with the actual work done. VINCI Construction also uses AI to examine a broader number of solutions to address a technical issue[14].

The value chain is typically structured as such:

Most of the data collection processes have been commoditized, be it by numerous players in drone or a few ones in satellite. However, there are still some gaps to fill, for instance in satellite thermal data collection[15].

The development of AI agnostic models is mostly trusted by deep pocket players since it is highly capital intensive both in terms of data computing and payroll. As an illustration, Mistral AI, a French startup founded in 2023, has raised over €1bn cumulatively[16].

These agnostic models then must be adapted to specific use cases, in order for infrastructure managers to benefit from them. Since there is plenty of data and AI models on the market, the software developers’ challenge is to find the most efficient combination. As an illustration, some players combine open-data satellite sources and expensive drone images to maximize the ROI of their solution.

Finally, most of the software developers are working with go to market and integration partners, which sometimes take high margins.

In conclusion, there are interesting SMEs to watch in the field of adapting AI models to specific infrastructure maintenance use cases. In this highly competitive environment, the winners will probably be the ones which are able to find the most efficient combinations of data sources and AI models.

[1] Oxford Economics, Global Infrastructure Outlook, 2015

[2] Global Market Insights, 2022

[3] Source: European Space Agency, May 2021

[4] Source: Morgan Stanley, Counterpoint Global Insights, Drones, December 2021

[5] Source: Goldman Sachs, August 2023

[6] Source: study by Forrester which interviewed over 160 decision makers in IT and OT roles at global infrastructure companies.

[7] According to IBM, a digital twin is a virtual representation of an object or system designed to reflect a physical object accurately. It spans the object’s lifecycle and is updated from real-time data.

[8] According to IBM, computer vision is a field of artificial intelligence (AI) that uses machine learning and neural networks to teach computers and systems to derive meaningful information from visual data.

[9] Source: Florida Power & Light website

[10] Accenture, Vegetation Management key levers for cost savings

[11] Source: World Economic Forum, The Next Frontier in Fighting Wildfires: FireAId Pilot and Scaling, January 2023

[12] Source: Elendil Space, July 2024

[13] Source: CTECH

[14] Source: Leonard, Yearbook 2024

[15] Source: ESA, 2022

[16] Source: Pitchbook