Artificial intelligence (AI) is already monitoring mining trucks, production lines, and workplace safety at industrial sites in Kazakhstan. Yet a gap remains between a successful experiment and technology that works every day at a factory or mine. Closing it was one of the main issues discussed at the third meeting of Kazakhstan’s Council for the Development of Artificial Intelligence.
A study by the British Chamber of Commerce in Kazakhstan, with contributions from Narxoz University and the Kazakh-British Technical University, found that 58.8% of AI projects among the industrial companies surveyed remained at the pilot stage, with none reaching full-scale deployment.
The sample was small, comprising just 17 executives and specialists from several industrial sectors, so its findings cannot be extrapolated to Kazakhstan’s entire industrial sector. The responses illustrate the problems companies encounter after their first experiments with AI.
Respondents most often cited a shortage of in-house specialists and expertise, followed by implementation costs. Yet 70.6% of completed pilots cost less than $50,000. The greater expense can come when the technology has to be integrated with existing equipment, data, and production processes.
The government’s 2025–26 roadmap covers 41 projects at 11 major industrial enterprises. They include Qarmet, which operates the country’s largest steelworks in Temirtau; copper producers Kazakhmys and KAZ Minerals; gold producer Solidcore Eurasia; and mining and metals group Eurasian Resources Group (ERG).
Industrial AI Acceleration has become one testing ground for deploying artificial intelligence directly at industrial sites. More than 12 companies submitted over 140 problems for AI teams to tackle, including equipment maintenance, safety, and procurement. Of 24 pilot projects initiated under the program, seven have reached industrial operation.
Some technologies have already made it onto factory floors and into mines. ERG uses its manufacturing execution system, ERG MES, to monitor production in real time. It collects data on output, technological processes, production losses, and equipment downtime. The Ministry of Industry and Construction estimates the economic effect at around 1.5 billion tenge, or approximately $3.39 million at the official exchange rate on October 2.
At KAZ Minerals’ Aktogay copper mine in eastern Kazakhstan, AI is used to manage the movement of mining trucks. Solidcore Eurasia uses the technology in medical examinations for employees working in hazardous production environments.
The Council also discussed the bureaucratic obstacles that can hold up successful projects. After a pilot, companies can face further approvals and testing, along with procurement and regulatory hurdles. Tokayev proposed avoiding the entire testing cycle again when a solution has already demonstrated results at a company operating under similar production conditions.
The government and Samruk-Kazyna are to select companies for wider AI deployment. Samruk-Kazyna is Kazakhstan’s sovereign wealth fund, which manages state assets across major industries.
The government also wants production workers to identify problems for developers. Engineers and operators see equipment downtime, product defects, and wasted materials firsthand. These problems are to form a list of practical challenges for technology companies and researchers.
The list is intended in part to guide Qazaq AI Research University, a specialized university in Astana focused on artificial intelligence and related technologies. Its research areas include machine learning, computer vision, and applied AI, while its laboratories also work on smart industrial automation.
AI-Sana is a state program for training university students in artificial intelligence and developing technology projects. It also connects student teams with problems from business and the public sector. Tokayev called for AI-Sana to be tied more closely to real industrial problems, and proposed testing more complex solutions at production sites together with company engineers.
The government is also expected to help reduce the early-stage risks faced by deep-tech projects based on advanced scientific and engineering research. Once a technology has been validated, further financing is expected to come predominantly from private capital. Under a proposed AI Challenge, companies would set specific problems for developers, whose solutions would then be tested with real data under production conditions.
Industrial AI also depends on usable data. Factories and mines have accumulated years of information on equipment, maintenance, and production. But it is scattered across different systems, stored in different formats, and is not always suitable for training AI models.
The authorities propose creating sector-specific industrial datasets with common standards for quality and security. Giving developers access remains difficult because some information is commercially sensitive or concerns critical infrastructure.
Tokayev proposed devoting the Council’s next meeting to transport and logistics. Kazakhstan sits on overland routes between China and Europe and is developing freight links across the Caspian Sea.
The meeting also addressed international AI rules. Tokayev proposed studying a voluntary White House agreement with leading technology companies and considering whether its principles could be applied internationally. He calls the idea the “Washington model” and plans to present proposals at the G20 summit in Miami in December.
The “Washington model” is Tokayev’s term rather than the name of a U.S. program. The White House agreement sets out voluntary commitments by technology companies on the safety of their most powerful AI models and contains no fines or other enforcement mechanism.
Kazakhstan has become the first Central Asian country to join the U.S.-led Pax Silica initiative, which focuses on cooperation across technology supply chains. It is also a founding member of the China-backed World Artificial Intelligence Cooperation Organization.
