Digital Twins: Why Kazakhstan Is Implementing Artificial Intelligence in Industry

Kazakhstan is launching a “comprehensive implementation” of artificial intelligence in industry. The government, in collaboration with Samruk-Kazyna (the sovereign wealth fund that is a co-owner of most state-owned companies, including KazMunayGas, “Kazatomprom,” “Air Astana,” etc.), will select pilot enterprises where production processes will be restructured using AI. In a column for Oninvest, Vadim Novikov, an advisor to the president of Almaty Management University (AlmaU), discusses how the operational efficiency of such companies is expected to improve, what risks exist, and what investors should keep an eye on.

Industrial AI analyzes production data and helps manage processes. Computer vision detects defects, predictive models identify signs of future breakdowns, and optimization algorithms determine the optimal processing parameters for raw materials. The system can advise the operator or control the equipment on its own.

In a Rockwell Automation survey , 1,560 executives from manufacturing companies with revenue of $100 million or more across 17 countries estimated that 34% of their operations are supported by AI; by 2030, they expect that figure to exceed half.

At the level of a specific operation, the application works as follows, for example: the BMW GenAI4Q system, developed in collaboration with the startup Datagon AI, creates customized quality inspection plans for vehicles based on data regarding their configuration and assembly.

The American mining company Freeport-McMoRan illustrates the financial scale of this approach. After implementing AI-driven recommendations for ore processing, copper output at the Bagdad mine increased by 5%. The potential for scaling this approach across assets in North and South America was estimated at $350–500 million in EBITDA per year (assuming a copper price of $4 per pound and unit costs below $2). This is a scaling scenario, not net returns after all expenses.

When manufacturers use AI, they can reap benefits even without an increase in capacity: fewer defects, unplanned downtime, and unnecessary operations. However, before these savings materialize, the company must pay for the system and integrate it into its production process. During the initial implementation phase, performance may even decline.

This occurred at Newmont, the largest gold mining company, at its Boddington mine in Australia. In 2021, the company invested $150 million in the transition to autonomous trucks. During the rollout, heavy rains, problems with excavators, and issues with site preparation complicated ore transportation. The company lowered its annual production forecast by 140,000 ounces of gold compared to the initial plan. The project continued, and Newmont later reported improvements in safety and productivity. This example shows that automating machines is not enough—the entire process in which they operate must be prepared.

A government initiative should add its own value to such implementations—for example, by accelerating them or reducing costs. An assessment by the research firm Ipsos showed that participants in Made Smarter Adoption, a British government program designed to accelerate the digital transformation of small and medium-sized businesses, increased their revenue faster than similar businesses not participating in the program. However, when compared to other applicants who did not receive significant assistance, no statistically significant effect on revenue was found. One explanation offered by the authors is self-selection: businesses that would have grown faster even without support may have been more likely to seek it.

In Kazakhstan, major companies began moving in this direction even before the directive was issued. ERG launched an AI project for the smelting furnace at the Aksu Ferroalloy Plant: the model is designed to predict electrode failures and reduce downtime. The company completed its data analysis and, if the results were positive, planned to roll out the solution to other furnaces. In a PwC survey, KazPetrol CEO Yerzhan Tazabekov discussed plans to apply AI in geology to locate oil deposits more accurately.

Another sector is the oil and gas industry. In October, the government instructed the Ministry of Energy, Samruk-Kazyna, and KazMunayGas to roll out AI and digital twins—computer models of facilities used to test operating modes—across production assets. At 12 major fields, the digital twins are expected to be fully operational by the end of 2028. For KMG shareholders, this is already part of the strategy: at the Vostochny Moldabek field, the company has set up remote data collection from all stationary automated group metering units. 99% of the production wells at this field are equipped with smart control stations.

The Samruk-Kazyna Fund itself set the benchmark for this impact by establishinga target for its portfolio companies: a 5% increase in EBITDA driven by AI.

How does the system for implementing AI in the real sector in Kazakhstan work? The Industrial AI Acceleration program brings large enterprises together with developers. However, at the end of 2025, Astana Hub reported that three-quarters of the tasks submitted related to support functions: human resources, finance, and procurement. There were fewer requests related to production.

Now, initiatives must come from the bottom up—from engineers, process engineers, foremen, and operators who see losses and malfunctions with their own eyes. Teams are promised an open selection process, access to data and testing facilities, and the opportunity for successful solutions to be rolled out. President Kassym-Jomart Tokayev cited repeated approvals and complex procurement processes as obstacles, which make it difficult to transfer a ready-made solution to a neighboring plant. The terms for selection, funding, and scaling up have not yet been published, and it is precisely these factors that will determine whether the directive translates into actual orders. Qazaq AI Research University will be responsible for training specialists and conducting applied research.

A pilot program for the priority procurement of domestic IT solutions for Samruk-Kazyna’s subsidiaries has been in effect since September 21: products from the special Astana Hub registry are procured through a simplified request for proposals.

Such a mechanism can help a developer secure their first contract more quickly and then offer the product to other companies within the group. However, a competitive advantage in procurement and benefits for production are two different things. For shareholders, it is not only the speed of contract conclusion and the product’s origin that matter, but also its cost, reliability, and proven effectiveness.

For the developer, the next question is whether the first contract will lead to repeat business. If every order requires gathering data from scratch, modifying the program, and going through the entire approval process, the technology will remain expensive to deploy.

A customer’s savings mean lost revenue for someone else. If repairs are billed by the hour and AI reduces the number of hours worked, the service company’s revenue drops at the same rate. If equipment breaks down less often, some emergency service calls disappear entirely.

McKinsey examines this conflict using the example of the oil and gas services industry: payments for days, hours, and operations are shrinking along with the demand for them. A decline in revenue does not necessarily mean an equal decline in profit—a contractor can cut its own costs or secure more orders if the client ramps up production. But under existing contracts, the contractor has little incentive to help the client scale back operations.

The solution is to change the payment model: pay for the reliability of the equipment or a share of the verified savings. That way, the contractor has a vested interest in minimizing repairs.

Another approach is to sell the technology along with the services. In January, the oilfield services company Baker Hughes signed a multi-year contract with Expand Energy, North America’s largest natural gas producer, for the Leucipa AI-based production optimization system. The service company now generates revenue not only from operations but also from software.

Getting from the office to the shop floor doesn’t necessarily mean reaching a point of profitability. For a shareholder, reducing the cost of implementation makes sense only if it improves the company’s cash flow. EBITDA does not account for capital expenditures. A project can boost this metric without recouping the costs of equipment, development, and integration. To assess payback, you need to consider the additional cash flows after all project costs have been accounted for; the impact of raw material prices must be separated from the effect of AI.

Investors should look at three things: whether companies publish production metrics (downtime, defects, yield) rather than just the number of projects; whether they disclose the full cost of the systems; and how contracts with contractors and developers are changing.

And one last thing: cost savings don’t always stay with the party that achieved them. If all market players reduce their costs, competition for customers will pass on part of the benefit to buyers through lower prices. Companies with a strong market position or exporters who sell at global prices are more likely to retain these benefits. Reducing maintenance costs is only the first step; costs, contracts, and the market determine who will earn more.

This article was AI-translated and verified by a human editor