Russian universities and alternative training channels supply less than half of the country's estimated AI personnel needs—covering only 43.9%—with large leakages from admission to employment. A small subset of institutions disproportionately place graduates into AI roles and secure higher wages, implying policy gains from targeting high-performing programs and strengthening university–industry links.

Main Finding

Russian universities substantially increased output of graduates from specialized higher-education AI programs (20.8k graduates in 2024, up from 14.3k in 2023), and a majority of those finishing full‑time programs enter the labour market; however, when compared with estimated sectoral demand (including alternative supply from self‑education and retraining), formal supply covers only 43.9% of required AI personnel. The study documents where graduates go (employment channels, employers, positions, wages) and reveals important losses along the pipeline from admission to employment.

Key Points

• Sample and scope: monitoring of 191 Russian universities (annual monitoring 2022–2024) that reported specialized main professional educational programs (OPOP VO) in AI. The sample includes most grant winners and many Alliance-ranked AI universities.
• Graduation trends: 20.8k graduates from specialized AI programs in 2024 (18.5k full‑time). Planned 2025 output ~28.6k — clear increasing trend.
• Employment outcomes (2024, full‑time cohort from 136 universities; n = 16,013):
  • 57.0% of graduates were reported as employed (≈9.1k people).
  • Of employed graduates, 53.5% (≈5.0k) obtained positions in the AI sector.
  • 28.9% (≈4.6k) continued study at the next education level (mostly bachelor→master).
• Sector / degree observations:
  • Degree level (bachelor, master, specialist) did not substantially change the likelihood of working in the AI sector vs other sectors.
  • Graduates from mathematical and ICT-related groups were comparatively more ready to enter the AI labour market (lower rates of continuing study) than some other specializations.
• Labour‑supply coverage: combining university graduates with alternative channels (self‑education and professional retraining) meets an estimated 43.9% of Russia’s AI personnel need (i.e., a substantial shortfall persists).
• Additional outputs: authors identify “key” universities (by employment effectiveness and reported wages) and the most common job titles filled by graduates; they also quantify average first‑year expected wages as reported by universities.
• Policy conclusion: universities have largely met government directives to expand AI specialist training in quantitative terms, but supply still insufficient relative to demand and losses occur along the training → employment pipeline.
• Limitations emphasized by authors: universities self‑identified which programs count as specialized AI programs (possible misclassification); monitoring covers specialized programs only (excludes AI module insertions); wage and employer data are limited to top‑10 employer lists per university and are self‑reported expected first‑year wages.

Data & Methods

• Data source: annual university monitoring (Federal Project “Artificial Intelligence”) for 2022–2024; 191 participating institutions (includes 16/17 grant winners, most regional partners, and many Alliance-ranked universities).
• Unit of observation: specialized higher‑education programs in the AI field (OPOP VO) as reported by universities.
• Collected variables (self‑reported by universities): number of graduates, education level (bachelor/master/specialist), channel of post‑graduation placement (employment, continued study, military conscription, parental leave), top‑10 employers with counts, job titles, and expected average first‑year salary for 2024 graduates.
• Analytical approach: descriptive analysis of distribution across channels, by aggregated specialty groups (UGSN), by university status (Alliance ranking / grant participation), identification of high‑performing institutions, and comparison of produced supply (graduates + alternative training) against estimated sectoral demand to compute coverage (result 43.9%).
• Noted methodological constraints: reliance on university self‑classification of what constitutes a specialized AI program; inability to track graduates longitudinally beyond the reported snapshot; RosTrud and official statistics (Form No. 1‑ИИ) not fully available/compatible for program‑level employment matching.

Implications for AI Economics

• Supply–demand mismatch: Despite rapid expansion of formal AI program output, the measured supply meets less than half of estimated AI sector demand (43.9%), implying persistent upward pressure on AI wages, competition for qualified talent, and bottlenecks to AI adoption across industries.
• Mixed quality signals: A >50% employment rate is encouraging, but program self‑reporting and lack of program‑level administrative linkage to employment records complicate assessing true alignment of skills and job tasks — a critical consideration when modelling human‑capital effects on AI diffusion and productivity.
• Importance of alternative training: Self‑education and professional retraining are material contributors to the AI workforce; policy and economic models should explicitly account for non‑degree supply channels when forecasting labour-market capacity and returns to AI investments.
• Role of universities and rankings: The study suggests heterogeneity across universities in employment effectiveness and wages. For policymakers and economists, this implies spatial and institutional concentration of AI human capital — regional imbalances and prestige effects that affect labour mobility, wage dispersion, and the regional economics of AI clusters.
• Pipeline losses and policy levers: Documented attrition from admission to employment indicates multiple intervention points (admissions, curriculum alignment, career services, industry partnerships, incentives for retention) where policy can improve effective supply — these interventions change short‑ and long‑run labour elasticity and adoption rates in economic models.
• Data infrastructure need: The limits of current monitoring (self‑reports, lack of program‑level linkage to administrative employment data) weaken evidence for policy evaluation. Improved data (program‑level administrative employment matching, longitudinal tracking) would sharpen estimates of returns to AI education and better inform labour‑market forecasting and wage modeling.
• Research recommendations: incorporate program‑level supply constraints and alternative training flows into structural models of AI adoption; study wage premiums by program quality and employer type; quantify regional diffusion effects from concentrations of high‑performing universities.

Notes: figures and percentages are taken from the authors’ monitoring-based analysis for 2024 as reported in the article. The paper focuses on specialized higher‑education AI programs as self‑identified by universities; results should be interpreted with the authors’ stated limitations in mind.