AI Is Changing the Geography of Opportunity

In the previous publication in the series, we showed that the new energy system requires not only generation, but also grids, storage, critical minerals, equipment and industrial coordination.

Artificial intelligence relies on a similarly multi-layered system. Training and deploying models require semiconductors, computing capacity, data centers, electricity, cloud infrastructure and data. Turning these inputs into economic gains also requires skilled people, enterprises ready to change and the capacity to redesign production processes.

These elements are unevenly distributed across countries and territories.

Some cities concentrate computing power, universities, technology companies, capital, and skilled workers. Other territories use ready-made digital products but play only a limited role in their creation. Large enterprises are able to invest in data, infrastructure, and restructuring processes, whereas SMEs face a shortage of money, specialists, and suitable solutions.

The central question, therefore, is no longer simply how widely AI spreads. It is which companies, workers and regions will be able to turn AI into productivity and new opportunities — and which will mainly gain automation of individual functions while becoming more dependent on external platforms.

This is the fourth publication in TALAP’s series on the new growth model. The final article will focus on the state’s ability to connect investment, energy, technology, human capital, and regional development into a single implementation system.

The outer layer of AI is becoming increasingly accessible.

According to the OECD, the number of developers of language models for cognitive tasks, including analysis and programming, increased from 9 in January 2024 to 47 in April 2026. The number of active text-based models rose from 22 to 453 over the same period. Their quality is improving, while the cost of using comparable capabilities is declining.

This dynamic creates the impression of rapid market convergence. An enterprise in almost any country can gain access to a language model, an image generator, a software assistant, or a data analysis tool.

However, the model market represents only one layer of the system.

At the lower levels, there remains a high concentration of advanced semiconductors, computing capacity, cloud infrastructure, large data sets, capital, and highly specialized talent.

The OECD warns that the concentration of computing, data and specialists, first-mover advantages, and vertical integration can strengthen the positions of a limited number of companies. Of particular importance are the links between chip manufacturers, cloud platforms, model developers and providers of application services.

A dual structure emerges.

At the user interface level, competition is increasing, and entry is becoming cheaper. At the infrastructure level, high capital and technological barriers remain.

This means that access to the model and control over the terms of its use remain different kinds of opportunities.

A user may choose between several services, while still depending on a limited number of providers of computing, cloud services, and foundation models. A national company can build an application, but its cost, functionality, and resilience will be determined by infrastructure located outside the country.

Therefore, technological sovereignty is gradually shifting from the question of whether a country has its own model to the more complex question of which elements of the entire chain the country is able to control, develop, or reliably source from multiple suppliers.

The data center becomes the physical hub of the new technological system.

It connects computing equipment, power grids, cooling, communications, and software infrastructure. Large computing complexes attract investment and make it possible to develop cloud services, model training, and the processing of large data sets.

However, a data center by itself does not guarantee broad economic gains for the surrounding region.

Its impact depends on the surrounding ecosystem, including research centers, university programs, software developers, corporate clients, technology suppliers, service and engineering companies, mechanisms for business access to computing resources, applied projects in industry and public administration.

In the absence of such an environment, the data center may remain a large infrastructure facility that consumes energy and serves external clients. It will increase the volume of investment, but its links with local enterprises, workforce, and the labor market will be limited.

In a more advanced model, the computing complex becomes the core of the ecosystem. Universities use it for research, companies — for product development, industrial enterprises — for process optimization, and local specialists gain access to complex tasks.

Competition among regions for data centers is therefore shifting from the provision of land and electricity toward the quality of the surrounding ecosystem.

The spread of generative services has sharply increased the number of people and companies that formally use AI. However, occasional use of a chatbot by an individual employee differs from integrating the technology into the production process.

The OECD defines systematic adoption as the deep integration of AI into production and business processes, going beyond the occasional use of user interfaces by individual workers.

Systematic adoption requires an enterprise to move through several stages:

1. identify the process where the technology can deliver a measurable effect;
2. prepare and structure the data;
3. choose or develop a solution;
4. integrate it into existing information systems;
5. change the distribution of functions between workers and algorithms;
6. establish accountability for the outcome;
7. ensure security and quality control;
8. assess the economic impact.

The main difficulty often arises outside the model itself.

An enterprise may be hindered by fragmented data, outdated equipment, the absence of digital record-keeping, incompatible information systems, weak cybersecurity, or management’s unwillingness to change processes.

Therefore, acquiring a software product yields limited results without organizational investment.

AI raises productivity when an enterprise changes both its technology and its way of working at the same time. Employees take on different tasks, managers receive new data for decision-making, and the production process follows a different order of control and interaction.

The adoption of AI among companies is uneven.

In OECD countries, the share of enterprises using AI increased from around 7% in 2021 to 20% in 2025. Large companies, startups, and more innovative enterprises are leading the way. SMEs more often face high implementation costs, a lack of infrastructure, and a shortage of specialists.

A large enterprise has several advantages: a larger volume of proprietary data, specialized units, the ability to fund pilots, access to external consultants, cybersecurity resources, and the ability to spread costs across a large volume of operations.

A small enterprise can quickly start using an off-the-shelf service, but it is more difficult for it to restructure the entire process. The cost of integration, data preparation, and employee training can be high relative to the scale of the business.

This creates a new productivity gap.

Companies that already have capital, data, and strong management gain benefits from AI more quickly. Higher productivity gives them additional resources for the next stage of implementation. They strengthen their market position and obtain even more data.

Enterprises with a weaker starting base use individual tools, but retain their existing work organization. Their relative lag increases even with formally broad access to the technology.

Therefore, the widespread adoption of AI does not necessarily lead to convergence among companies. Without specific mechanisms, it can accelerate existing disparities.

Discussion of the labor market often revolves around the question of which occupations will disappear.

The OECD report Skills in the AI Age offers a more precise framework. AI affects employment simultaneously through the automation of existing tasks, the creation of new tasks and occupations, and productivity gains. The overall effect depends on the balance among these processes.

High exposure to AI does not automatically mean that a job will disappear.

Many skilled occupations overlap most strongly with AI: managers, analysts, engineers, and other information workers. However, a significant share of their work involves complex decisions, responsibility, interaction with people, and work in non-standard situations. These functions are harder to fully automate.

The composition of work changes within a single occupation.

For example, a specialist may spend less time on searching for and processing information, preparing standard documents, performing routine calculations, drafting initial versions, classifying requests, and handling routine communications.

At the same time, the importance of task setting, result verification, professional judgment, contextual understanding, communication, accountability for decisions, and handling non-standard cases increases.

Therefore, AI can simultaneously reduce some operations and raise the requirements for the remaining functions.

A worker retains their position when they are able to move from carrying out a standardized procedure to monitoring, interpreting, and solving more complex tasks. If this is not possible, reducing the routine part of the job may lead to a lower need for staff.

The labor market needs specialists capable of building models and complex digital infrastructure. However, this group remains small.

According to OECD estimates, about 1% of the workforce has advanced competencies in machine learning, data analysis, and related fields. For most workers, a different combination of skills matters.

• The first level consists of basic abilities: literacy, mathematics, an understanding of scientific logic, and the ability to work with information.
• The second level is associated with digital skills: using software tools, data, automated systems, and communication tools.
• The third level consists of complementary competencies: critical thinking, asking questions, creativity, teamwork, communication, and the ability to keep learning.

OECD emphasizes that such competencies make it possible to interact effectively with AI and adapt to changing tasks.

Industry-specific knowledge must be added to this set.

In industry, value is created by a specialist who understands the production process and can use AI to diagnose equipment. In agriculture, it is a person capable of linking satellite, meteorological, and production data. In medicine, it is a doctor who can assess an algorithmic recommendation within the broader clinical picture.

Domain knowledge makes it possible to distinguish a substantive result from a convincingly presented error.

Therefore, preparation for the AI economy is not limited to mass training in programming or working with chatbots. It is about combining digital, professional, and managerial competencies.

Although generative AI is too new to provide a long historical record, the OECD Employment Outlook offers several findings that are relevant at the regional level.

According to the report, in more than half of OECD countries, the gap in employment rates between individual regions exceeds 20 percentage points. Population characteristics explain no more than half of this gap. The same worker with the same qualifications faces different opportunities depending on where they live.

The reason lies in the structure of the local economy.

Territories differ in their industrial mix, the number of employers, the size of enterprises, the quality of infrastructure, access to education, worker mobility, the depth of the services market, and links to major markets.

Technological shocks pass through this local economic structure.

A region with strong universities, large companies, and a developed services sector creates demand for new skills more quickly. Enterprises adopt AI, attract specialists, and create additional jobs in consulting, development, and services.

An industrial area may see productivity growth through automation, but a smaller increase in employment. When displaced workers do not find new positions in the same locality, the technological effect is accompanied by migration and a decline in local demand.

A rural or peripheral region may use digital services while remaining primarily a consumer of solutions developed in major centers. The economic effect may appear as lower costs without translating strongly into a local high-productivity services sector.

Structural adjustment is often assessed through the overall balance: how many jobs disappeared and how many were created.

OECD shows the limitation of this approach. Workers who have lost their jobs in industry rarely move into new service-sector jobs. These vacancies are often filled by young people entering the labor market for the first time.

At the national level, a decline in one sector may be offset by growth in another. At the level of an individual worker and city, this compensation does not occur automatically.

A new job may be in another region, require different qualifications, offer a different level of pay, imply relocation, and emerge several years after the old position has been eliminated.

Therefore, the technological transition creates several temporal and territorial gaps.

The economy adds new activities faster than a worker can retrain for a new occupation. A vacancy opens up in a large city, while layoffs occur in an industrial center. An educational program graduates specialists after the company has already filled the position by hiring externally.

As a result, overall productivity growth can coexist with a local deterioration in employment and incomes.

Two regional cascades can emerge

AI can trigger a positive cascade of opportunity concentration:

computing and educational infrastructure → the arrival of technology companies and investment → growing demand for specialists → higher incomes and an expanding services market → an influx of new workers and businesses → further strengthening of the territory.

A reverse cascade can emerge in another region:

weak AI adoption and a limited employer market → lagging productivity → declining attractiveness for investment → outflow of skilled workers → shrinking local demand → further weakening of the economic base.

These processes are not inevitable. Their direction depends on the quality of regional and sectoral policy.

OECD recommends combining local job creation with support for mobility. For lagging regions, diversification of industries, vocational training, and support for businesses are important. For workers, affordable housing, transport, childcare facilities, recognition of qualifications, and targeted relocation assistance are important.

Employment policy therefore extends beyond retraining courses. Jobs, skills, housing, services and regional development become parts of the same transition.

In July 2026, Kazakhstan approved the Digital Qazaqstan strategy until 2029. It views AI as a multi-layered system encompassing energy and infrastructure, computing capacity, data and models, digital platforms, and applied services. This approach corresponds to the structure identified by international analysis: economic outcomes are shaped by the entire chain, not by a standalone software product.

The strategy envisages a deep digital transformation of industry, subsoil use, logistics, energy, the agro-industrial complex, construction, the financial sector, and SMEs. The flagship infrastructure project is to be the AI Hub in the form of Data Center Valley, with a projected computing capacity of up to 1 GW. It also plans to at least double the export of Kazakhstan’s IT services and bring no fewer than three domestic companies valued at over $1 billion to the global market.

In the area of human capital, the strategy envisages large-scale training of specialists through the academic environment and regional IT schools, as well as raising the population’s literacy in generative AI to at least 20%. The AI-Sana program is aimed at involving up to 100,000 students in developing applied solutions for priority sectors and in technology entrepreneurship.

These directions create the foundation for the technological transition. The next question concerns their territorial and production architecture.

Large computing centers require robust power supply, reliable networks, communications, and specialized maintenance. Therefore, they are located in a limited number of suitable areas.

Kazakhstan’s Data Center Valley creates an opportunity to establish a large-scale computing center in Ekibastuz. The project provides for an initial available capacity of about 300 MW and subsequent expansion to 1 GW.

Its long-term impact will depend on what functions emerge alongside the computing infrastructure.

The maximum outcome assumes access for Kazakh companies and researchers to computing capacity, the development of engineering and software services, links with universities, localization of part of the equipment and maintenance, the emergence of applied projects for the energy and industrial sectors, and the creation of sustained demand from domestic business.

Under such a scenario, Ekibastuz could gain an additional technological specialization alongside its energy specialization.

Under a narrower scenario, the territory provides land and electricity, while development, data management, and service creation remain in other centers or countries. The investment project will go ahead, but its local economic impact will be significantly smaller.

Early adoption of AI is usually concentrated where corporate headquarters, financial institutions, government agencies, universities, IT companies, and skilled professionals are already located.

For Kazakhstan, this means a high likelihood of accelerated AI development in the largest cities. Here, it is easier to build teams, find clients, attract capital, and organize interaction between business, government, and educational institutions.

Such concentration gives the country fast-growing hubs. At the same time, it can intensify internal migration and widen the gap between major centers and other territories.

Astana is already maintaining a high migration inflow: in January–May 2026, the city’s internal migration balance amounted to about 26.5 thousand people. The average nominal wage reached 609.7 thousand tenge in the first quarter.

AI can further enhance the attractiveness of territories where high-paying jobs and modern services are concentrated. Therefore, the development of regional competencies becomes part of the policy for managing agglomerations, housing, and infrastructure.

For industrial territories, the main potential lies in applying the technology to the existing economic base.

These may include predictive maintenance of equipment, industrial safety management, quality control, energy consumption optimization, geological analysis, transport planning, warehouse automation, and production flow management.

Such solutions can improve the productivity of existing enterprises and create demand for local engineers, integrators, and service companies.

However, improved efficiency at an individual large enterprise does not yet guarantee the development of the entire territory.

A systemic effect emerges when training programs, local technology suppliers, testing sites, access to industry data, joint projects between enterprises and universities, and the ability to scale the solution across companies are formed around the project.

In this case, AI becomes a tool for diversifying an industrial region. The enterprise gains a technology; the region gains a new capability.

Large corporations and government entities are able to finance complex implementation first. However, broad productivity gains depend on the ability of SMEs to use the technology.

For Kazakhstan, this is especially important in trade, transport, agriculture, construction, manufacturing, and services.

Small businesses more often need not their own models, but affordable applied solutions: inventory management, demand forecasting, automation of accounting and administrative operations, customer analytics, quality control, logistics planning, document preparation, and support in entering foreign markets.

The main barrier is the ability to choose the right process, prepare the data, and integrate the solution into operations.

Therefore, a policy for the diffusion of AI among SMEs may include technology assessment, off-the-shelf sector-specific solutions, shared platforms, training for managers, and support for pilot projects.

The key indicator will not be the number of companies that have used a generative service at least once, but the share of enterprises that have improved productivity, revenue, or management quality with its help.

Mass AI literacy creates a general level of readiness. Practical economic impact requires the next layer — sector-specific training.

To do this, the educational program must answer a specific question: how will the work of an engineer, agronomist, doctor, teacher, civil servant, logistics specialist, or finance professional change.

The most effective training emerges in conjunction with an employer and a real task.

The student or employee gains access to sector-specific data, studies the existing process, develops a solution, and tests it directly at the enterprise. The company participates in shaping the program and gains the opportunity to assess the result.

OECD identifies on-the-job training, employer-led programs, flexible modular pathways, and continuous upskilling as key elements of preparation for the AI economy.

For Kazakhstan, this approach will make it possible to connect AI-Sana, universities, and regional enterprises. Then the educational program will simultaneously produce a specialist, an applied product, and a new channel of interaction between the university and the economy.

One occupation includes tasks with varying probabilities of automation. Therefore, the risk map should be based on an analysis of specific tasks within industries and enterprises.

Such an approach will make it possible to see where AI will complement labor, where it will reduce routine workload, where it will change qualification requirements, where it may reduce the number of jobs, and what new functions may emerge in the same territory.

When planning large data centers, it is important to identify in advance the circle of national and regional users, educational partners, and suppliers.

The infrastructure project must have its own ecosystem development program; otherwise, the technological value may remain outside the host territory.

Industrial regions need expertise in industrial AI, agricultural regions need expertise in geospatial data and precision farming, transport hubs need expertise in logistics, and large cities need competencies in digital services and the management of complex systems.

A common baseline can be combined with regional applied tracks.

Startups create new products, while existing small and medium-sized enterprises determine how widely the technology spreads.

Support for SMEs can become a channel for transmitting the effects of digital infrastructure into traditional industries and regions.

Specialists choose not only a job, but also a location. Housing, schools, transport, healthcare, and the cultural environment affect a region’s ability to attract and retain people.

Therefore, regional competition for talent cannot be resolved by a single educational program or a higher salary alone.

For Kazakhstan, two models of the spread of AI’s economic effects can be identified.

The first is built around concentration.

Computing capacity is created in several locations. Developers, universities, and headquarters are concentrated in the largest cities. Large enterprises and government bodies adopt new systems more quickly. Other territories receive ready-made services but play only a limited role in creating technologies and competencies.

Such a model can rapidly create national technology hubs and improve the efficiency of large organizations. At the same time, it widens disparities between companies, workers, and regions.

The second model combines infrastructure concentration with distributed application.

Large computing centers remain concentrated in a limited number of locations, but regional universities, enterprises, and developers gain access to them. Applied programs are developed around regional specializations. SMEs receive tools for implementation. Workforce training is linked to existing production facilities and public services.

In this model, the country does not try to distribute all technological infrastructure evenly. It creates channels through which its economic benefits spread more widely.

The key divide is between the geography of infrastructure and the geography of opportunity.

• The first will almost inevitably be concentrated. Advanced computing capacity, major universities, and technological capital cannot be distributed evenly across all territories.
• The second depends on policy. Opportunities can spread through access to computing, sector-specific solutions, training, digital platforms, business support, and links between regions and national centers.

AI does not by itself equalize or divide regions. It amplifies the structure into which it is introduced.

In an economy with strong regional linkages, technology expands enterprises’ and workers’ access to new opportunities. In a fragmented system, it more quickly increases the advantages of already developed centers.

Kazakhstan’s task, therefore, is to connect the technological leap with the productivity of existing industries, the development of regional competencies, and the creation of new pathways for workers.

It is impossible to solve such a task within a single digital program.

AI depends on energy, education, industry, science, competition, the labor market, regional development, and public administration. These areas have different institutional boundaries, budgets, indicators, and implementation timelines.

Therefore, the final publication in the series will be devoted to the main constraint of the entire new growth model — the state’s ability to coordinate complex priorities and bring them to measurable results.

The next publication in the series is “Deliverability Is Becoming a Key Resource for Development”.

The article is based on three OECD reports that describe the structure of AI markets, changing skill requirements, and the geographic distribution of jobs and incomes.

1. OECD — Artificial Intelligence Markets: Recent Developments and Competition Issues

https://www.oecd.org/en/publications/artificial-intelligence-markets_d531d73f-en.html

2. OECD — Skills in the AI Age

https://www.oecd.org/en/publications/skills-in-the-ai-age_972bd15e-en.html

3. OECD — OECD Employment Outlook 2026: Geographic Disparities in Jobs and Incomes

https://www.oecd.org/en/publications/oecd-employment-outlook-2026_7e710f54-en.html

4. Ministry of Artificial Intelligence and Digital Development of the Republic of Kazakhstan — Kazakhstan adopts the nationwide Digital Qazaqstan strategy through 2029

https://www.gov.kz/memleket/entities/maidd/press/news/details/1238140?lang=en

5. Ministry of Artificial Intelligence and Digital Development of the Republic of Kazakhstan — Memorandum signed to advance the Data Center Valley project

https://www.gov.kz/memleket/entities/maidd/press/news/details/1211917?lang=ru

6. Ministry of Artificial Intelligence and Digital Development of the Republic of Kazakhstan — Data Center Valley: package of agreements with Firebird and NVIDIA

https://www.gov.kz/memleket/entities/maidd/press/news/details/1240802?lang=en

7. Bureau of National Statistics of the Republic of Kazakhstan — Astana city statistics

https://stat.gov.kz/ru/region/astana/