Neural Networks Move Onto the Factory Floor
MMC Norilsk Nickel has become one of the first industrial companies in Russia to deploy generative AI within a secure enterprise environment, combining its own computing infrastructure and the domain-specific MetalGPT model with Yandex's cloud platform. The hybrid approach reduces the cost of supporting AI solutions while allowing sensitive corporate data to remain protected.
The solution is built on a hybrid architecture developed by Yandex B2B Tech, the company's business unit focused on enterprise technology solutions. Access to models hosted in Yandex AI Studio is provided through a dedicated physical connection, with no customer data stored on the provider's infrastructure. That allows AI agents to operate entirely inside Norilsk Nickel's corporate network without requiring internet access, fully meeting the company's information security requirements. The platform already runs eight artificial intelligence agents built on Yandex AI Studio together with Norilsk Nickel's own MetalGPT model.
The metallurgical company is developing AI agents for engineering design, project management, inventory management, and finance using its cloud infrastructure. Those digital assistants are powered by MetalGPT and integrated directly into business processes. In some use cases, tasks that previously required 30 days can now be completed within a matter of hours.
The next stage of Norilsk Nickel's collaboration with Yandex will expand employee access to AI tools, including through a Yandex AI Studio module that enables staff to build simple AI agents.
The First Domain-Specific LLM for Metallurgy
Norilsk Nickel uses artificial intelligence throughout its operations, from mining and metal production to corporate support functions. In December 2025, the company introduced MetalGPT, the world's first open-source domain-specific large language model for metallurgy and mining. The 32-billion-parameter model was trained on 10 GB of specialized data, including more than one million internal documents – such as operating procedures, technical regulations, patents, and engineering documentation – together with 500,000 question-and-answer pairs derived from real manufacturing tasks.
AI Agents for Everyone
Yandex AI Studio is a cloud platform for developing and deploying AI agents within enterprise workflows. Its defining feature is a set of low-code tools that allow economists, logistics specialists, process engineers, and other professionals to build digital assistants without programming experience.
The platform also offers ready-made solutions, including an AI legal assistant for legal guidance and Yandex Speech Sense for analyzing customer communications. Other capabilities include voice agents for contact centers, multi-agent systems, and intelligent document search. AI agents integrate with external services through the MCP protocol.
From Cloud Computing to Generative AI
The collaboration between the two companies began back in 2023, when Norilsk Nickel migrated development of its mining geology modeling and planning applications to Yandex Cloud. That move accelerated several key processes: provisioning computing resources fell from 10 days to a single day, while development of machine learning models dropped from one month to 15–16 days.
The latest initiative builds naturally on that partnership, extending it into generative AI and autonomous software agents.
From Pilot Projects to Enterprise Platforms
During 2025–2026, Russian manufacturers began moving rapidly from isolated generative AI experiments to secure enterprise-wide AI platforms.
For example, in September 2025, Severstal launched Da Vinchi (Da Vinci), its own generative AI platform featuring a no-code environment for creating personalized AI assistants inside a secure corporate infrastructure. Users have access to multiple large language models capable of processing both text and images. At launch, the platform supported 500 users working on 200 AI solutions.
NLMK is deploying large language models to process engineering documentation. EVRAZ is implementing 15 AI initiatives ranging from predictive equipment analytics to manufacturing process optimization. MMK uses artificial intelligence to improve production efficiency and reduce operating costs.
Those investments are driven by clear economic incentives. According to market estimates, Russian companies spend more than 3 trillion rubles annually (approximately $39 billion) on technical support, legal services, accounting, and human resources. AI agents can assume part of those responsibilities, reducing employees' routine workloads by as much as 30%.
The number of AI agent deployments across industrial enterprises is expected to continue growing in the near future. Against that backdrop of rising demand, the joint project between Norilsk Nickel and Yandex Cloud could become a reference model for deploying generative AI across large industrial enterprises.
