Generative AI is gradually moving beyond office tasks and starting to transform industry and manufacturing. Companies are proceeding cautiously here, but the first results of implementing artificial intelligence in industry are already evident: from accelerating design to optimising logistics.
Recent research states that successful pilots will transition to industrial operation within the next 1-2 years. We’ve put together a list of the most interesting findings and use cases.
The implementation of AI in industry in 2025 is happening very rapidly. Here are the most persistent trends:
Multi-agent systems
While companies previously worked with a single model, “teams” of AI agents are now becoming fashionable. They can distribute tasks: one searches for data, another verifies it, and a third makes predictions. This approach is convenient for complex production processes where coordination is crucial.
RAG (Retrieval-Augmented Generation)
The model “pulls” fresh data from a database or documentation into its response. This is especially important for neural networks in industry, where an error in instructions can cost millions. Experiments are underway with RAG for video analytics, such as assessing equipment behaviour via cameras.
New approaches to model training
Training AI on vast datasets is becoming too expensive. Therefore, methods requiring less data or where the model “teaches itself” are being developed. This paves the way for small language models (SLMs), for example, for analysing machine operation or a specific production process.
Data as a product
If companies previously guarded their data jealously, a market for industry-specific AI marketplaces is now forming. Unique datasets are becoming a commodity, just like machinery or raw materials.
AI agents and copilots
According to Gartner forecasts, by 2025, 114 million laptops with built-in AI copilots will be sold worldwide. In industry, these are “copilots” for engineers and operators, helping them find errors faster or check equipment parameters.
AI governance and AI control
Platforms for AI management are actively developing in the US and Europe: compliance with laws, ethics, and data protection. For the industry, this is a safety issue, for example, when using AI in the chemical or energy sectors.
Energy-efficient computing
Training AI is very energy-intensive. Therefore, interest in new types of computing is growing: quantum, photonic, and neuromorphic. Gartner predicts that by 2028, 30% of AI solutions will be “green.”
Physical AI
A new wave of AI trends in industry involves models that understand physics and spatial relationships. This is important for robotics or building design: AI considers the laws of the real world, not just text or numbers.
AI use cases in industry: from software development to predictive maintenance
Research and development
Rolls-Royce Aerospace (UK) — new alloys
Engineers connected the Alchemite™ platform, which can find hidden patterns even in incomplete and noisy data. AI suggested which alloy formulations were worth trying and which experiments were most important.
Result: A new alloy was developed and tested with 90% fewer experiments than usual, resulting in savings exceeding £10 million.
Insilico Medicine (Hong Kong) — next-generation drugs
The biotech company used the Chemistry42 platform, where over 40 AI models generated and verified molecules. This led to finding a new PHD inhibitor that helps the body produce red blood cells, a potential drug for anaemia in chronic kidney disease.
Result: The drug development process accelerated and became cheaper. The drug is already considered a breakthrough in pharma.
Botswana Diamonds (UK) — diamond search
The company implemented the Xplore platform, which used AI to analyse hundreds of thousands of geological exploration data points: from soil samples to aerial surveys. The algorithms not only processed the vast amount of information but also generated new hypotheses about deposits.
Result: Seven new kimberlite targets and promising zones were found, including gold, copper, silver, and platinum. The company has already applied for new licenses.
GE Aviation (USA) — more economical and lighter jet engines
GE, together with Autodesk Research, used generative design to improve jet engine construction. AI helped optimise reinforcements and cooling lines so the engine could withstand extreme loads while being more economical.
Result: Part weight reduced by approximately 30-35%, fluid pressure decreased by 91%, and thermal energy savings amounted to about 16 GJ, directly impacting fuel consumption and CO₂ emissions.
LEAP 71 (UAE) — rocket engine in 3 weeks
The Dubai company LEAP 71 developed its own AI system, Noyron, which fully designed an Aerospike rocket engine. From idea to prototype, it took just three weeks instead of months or years of traditional development. The engine’s key feature is that it automatically adapts to atmospheric conditions and remains efficient at different altitudes.
Result: Creating a working prototype took only three weeks, and the new engine architecture made it simpler to produce and more reliable in operation.
ICON (USA) — AI home designer
The company, known for 3D-printed building projects, introduced the Vitruvius platform. It can design homes for clients: from architectural style to layout, considering building codes. The user inputs parameters like “three-bedroom house in Mediterranean style,” and AI offers ready-made options
Result: Ability to quickly and inexpensively create quality housing designs that rival the work of professional architects.
Obayashi Corporation (Japan) — faсades with a click
Together with SRI International, the company developed the AiCorb® system, which turns photos and drawings into ready-made building faсade options. One algorithm generates the design in 2D, and a second turns it into a 3D model for customer approval.
Result: Architects get dozens of options in minutes and reach a consensus with clients on the final project faster.
Bosch (Germany) — quality control without manual inspection
Bosch is implementing generative AI in its German factories to replace manual inspection of fuel injection system components. Algorithms create synthetic images, learn from them, and recognise defects, passing only “doubtful cases” to humans. Such AI in industry cases show how synthetic data helps accelerate model training.
Result: Component testing time reduced from 3.5 to 3 minutes, and launching new AI applications now takes weeks instead of months.
Siemens (Germany) — AI assistant for maintenance
The company is developing Industrial Copilot for repair, prevention, and predictive maintenance of equipment. The solution based on Microsoft Azure combines instructions for engineers with the ability to predict failures. This is an example of how AI in manufacturing becomes part of daily work.
Result: Time for reactive maintenance reduced by an average of 25%.
Schneider Electric (France) — Copilot for engineers and operators
Together with Microsoft, the company introduced an AI assistant in the EcoStruxure Automation Expert platform. It integrates with various systems and helps engineers work with real-time data: from diagnostics to predictive maintenance. Such solutions have been one of the main AI trends in the industry in recent years.
Result: Simplification of engineering processes and reduced employee workload due to fast access to data and recommendations.
OreFox (Australia) — mineral exploration with an AI model
A joint project between OreFox AI and Australian universities created the PorphyryGPT model for geologists. It is trained on research and data about porphyry systems and helps find patterns in geochemistry and mineralogy.
Result: Geologists spend less time analysing literature and making decisions faster.
Predictive equipment maintenance
Westinghouse (USA) — AI for nuclear power
In 2024, the company introduced the Hive system: the first generative AI platform for the nuclear industry. Its core is the Bertha language model, trained on over a century’s worth of engineering data, operational reports, and regulatory documents. Hive automates the creation of technical documentation, generates equipment maintenance recommendations, and even creates scenarios for personnel training and accident simulation.
Result: Design and licensing timelines for nuclear reactors are shortened, equipment condition prediction becomes more accurate, and work with regulations is automated.
US Steel (USA) — assistance with repair and maintenance
In 2023, US Steel, together with Google Cloud, launched the MineMind system for its mines in Minnesota. Engineers can now ask AI questions about hydraulic repair, sensor calibration, or conveyor belt replacement. MineMind finds the necessary data in technical manuals, generates step-by-step instructions with diagrams, and even shows the confidence level of the answer.
Result: Time to complete work orders reduced by approximately 20%, increasing productivity and reducing maintenance costs.
IBM + AWS (USA) — automation of the entire development cycle
IBM, together with AWS, created a solution for SDLC, which is available on the AWS Marketplace. It uses large language models via Amazon Bedrock (including Claude from Anthropic) and automates key tasks: monitoring architectural standards, security checks, working with APIs, documentation, and testing. For developers, this means less routine and more focus on code.
Result: Development time was reduced by up to 30%, test generation became 25% faster, code quality improved by about a quarter, and analysis took 60% less time. Additionally, collaboration between teams improved.
Cognizant + Google Cloud (USA) — training and integration of Gemini
In 2024, Cognizant expanded its partnership with Google Cloud, betting on the Gemini model. Over 70,000 company employees are being trained to use AI for writing and testing code. Furthermore, Gemini’s capabilities are being integrated into new platforms, including Cognizant Flowsource, which helps automate developers’ processes.
Result: The company expects accelerated code development, automated testing, and improved incident analysis, enhancing the efficiency of internal processes and client projects.
Suzano (Brazil) — fast access to corporate data
Suzano, the world’s largest pulp producer, implemented the VagaLúmen AI agent based on Gemini Pro. Employees can now write queries in natural language, and the agent converts them into SQL code and retrieves the necessary data from BigQuery and SAP.
Result: The time for requesting raw materials and supplies was reduced by 95%: from two minutes to eight seconds. In the pilot, 473 employees used the system, and soon it will be available to six thousand users.
Honeywell (USA) — workplace trainers and assistants
The company integrated Google’s large language models into its Honeywell Forge platform, which collects industrial data from around the world. AI agents make this mass of information simple and understandable: they help engineers and technicians troubleshoot, suggest design improvements, and find ideas for preventive maintenance.
Result: Employees spend significantly less time searching for information in the company’s knowledge base and find solutions to problems faster.
South32 (Australia) — more manganese, less waste
The mining company South32 implemented generative AI that analyses the chemical composition of ore and provides recommendations for processing. The system advises operators in real-time on how to adjust production parameters to maximise manganese extraction efficiency.
Result: In the first year of operation, commercial output increased by 19 thousand tons, and the manganese content in waste decreased by 12-15%.
Ola Electric (India) — digital twins for factories
The electric scooter manufacturer launched the Ola Digital Twin platform based on NVIDIA Omniverse. It creates digital twins of production lines and allows testing changes in a virtual environment before implementation in the factory. AI also generates synthetic data for training robots and quality control systems.
Result: Time-to-market reduced by 20%, and the world’s largest scooter factory was designed in just 8 months.
Uber Freight (USA) — smart logistics
The company introduced the Insights AI platform to analyse transportation data. It helps identify bottlenecks, predict delivery times, and optimise routes. Users can ask questions in plain language and receive detailed analytical reports and forecasts.
Result: Uber Freight clients gained fast access to relevant analytics, improving supply chain management efficiency and reducing costs.
Aramco (UAE) — own giant model
Saudi Aramco launched the generative AI Metabrain, which was trained on 90 years of company data. It helps analyse drilling plans, geological data, and market indicators, providing recommendations for optimising drilling and downstream processes.
Result: Reduced drilling costs, increased efficiency of production operations, and more accurate forecasts of oil product prices are expected. By the end of the year, the company plans to increase the model size to one trillion parameters.
Industry is not adopting AI as rapidly as fintech or media. Companies here act cautiously: first, they conduct pilots, test them thoroughly, and then move to implementation. But it’s already clear that the effect of the first projects is significant.
The experience of global companies and AI cases in industry show that using AI allows for:
- Speeding up design several times over and automating documentation and standards compliance checks.
- Reducing equipment downtime by almost half thanks to integration with predictive analytics.
- Decreasing excess raw material inventories by tens of percent by synchronising demand and supply.
- Optimising logistics so that tasks taking hours are solved in minutes.
Experts agree: within the next year or two, the first successful pilots will transition to industrial operation, and AI in industry will become part of the daily processes at enterprises in 2025.
Learn how AI can drive innovation in your industry