Kazakhstan’s 5 Million AI Training Target: The Five Programmes Doing the Work Right Now

Kazakhstan says it will train more than 5 million people in digital and AI skills over the next five years. The figure was presented on 24 September 2026 at the AI Skills Coalition meeting in New York, alongside real enrolment numbers from five named programmes already running: Tech Orda, AI-Sana, TUMO Astana, Tomorrow School, and the newly opened Qazaq AI Research University.

This is not a slide-deck ambition. Vice Minister Doszhan Musaliyev of Kazakhstan’s Ministry of Labour and Social Protection cited specific completion and employment figures for programmes that predate the announcement, which is more evidence than most national digital-skills pledges bring to the table. It is also, on the numbers published so far, a target that the current delivery base does not yet come close to matching.

This piece walks through what each of the five programmes actually delivers, checks Kazakhstan’s own arithmetic against the enrolment figures on record, and looks at what a national target of this size implies for anyone building or buying training-delivery capacity in Central Asia, whether that is a government running its own AI-Sana equivalent or a vendor evaluating the region for LMS and content partnerships.

What is Kazakhstan’s 5 million AI training target, and where was it announced?

Kazakhstan’s Ministry of Labour and Social Protection presented a target to train more than 5 million people in digital and AI skills over the next five years, announced on 24 September 2026 at the AI Skills Coalition meeting held on the margins of the 81st UN General Assembly high-level week in New York.

The meeting was organised by the International Telecommunication Union (ITU) and also included delegations from Trinidad and Tobago, Pakistan and the Dominican Republic, alongside UNDP, UNESCO, the World Bank and private tech companies. Kazakhstan used the platform to present evidence for its target rather than just the number itself: completion and employment data from Tech Orda, and enrolment figures from 79 Kazakh universities running IT and AI programmes, currently teaching more than 76,000 students.

What has Tech Orda actually delivered so far?

Tech Orda is Kazakhstan’s flagship national AI-training programme, and its cited numbers are more than 12,000 people trained with an employment rate above 90 percent among graduates. Those two figures did the heavy lifting in New York, since a 90 percent placement rate on a program this size is a genuinely strong data point if it holds up to scrutiny.

The catch is scale. Twelve thousand graduates is a rounding error against a target of 5 million, and the announcement gave no breakdown of what “trained” means inside Tech Orda: course length, assessment method, or whether the 90 percent employment figure covers full-time roles, contract work, or self-employment. Kazakhstan also runs Tech Orda alongside an AI Olympiad and specialised tracks for civil servants and senior managers, none of which had enrolment figures attached in the New York presentation.

What is AI-Sana, and why do Coursera, Huawei and NVIDIA appear in it?

AI-Sana is Kazakhstan’s university-level AI skilling programme, delivered through partnerships with Coursera, Huawei ICT Academy, NVIDIA’s Deep Learning Institute and Astana Hub. More than 100,000 students have completed advanced training through it, generating over 703,000 certificates, since students typically complete multiple courses across the partner platforms.

The vendor mix matters more than the certificate count. AI-Sana pairs a MOOC platform (Coursera) for breadth, a hardware vendor’s academy (NVIDIA) for applied technical depth, and a telecoms vendor’s academy (Huawei ICT Academy) for infrastructure and networking skills, then routes the strongest output into a startup pipeline: of more than 22,000 student-proposed AI projects, 515 are being tested or implemented, with a target of 1,500 student teams developed into registered startups by 2027.

What the partner mix signals to other governments

A government training programme naming Coursera, Huawei and NVIDIA as delivery partners in the same announcement signals that Kazakhstan is comfortable outsourcing content and certification to established vendors rather than building a national curriculum from scratch. That is a faster path to volume, but it also means Kazakhstan’s “trained” figures partly reflect vendor completion metrics, which are typically looser than a government-run assessment.

How do TUMO Astana and Tomorrow School differ as delivery models?

TUMO Astana and Tomorrow School both target people with no prior technical background, but they sit at opposite ends of the age and delivery spectrum. TUMO Astana is a free, drop-in style learning centre for 12 to 18 year olds, licensed from the Armenian TUMO model established in 2011, covering programming, robotics, generative AI, technology and design; it has 2,376 active students and is aiming for 5,000 by the end of 2026.

Tomorrow School is an adult-facing, 18-month structured course in programming and AI that explicitly requires no prior coding experience, now running in 14 regions of Kazakhstan outside the capital. It is smaller in headcount, 493 students, but its employment claim is specific and checkable: 57 participants already placed in IT-sector jobs, a completion-to-employment ratio training providers evaluating similar bootcamp models will want to watch as the cohort matures.

What does the new Qazaq AI Research University add?

Qazaq AI Research University (QAIRU), which opened in September 2026, adds a research-university tier that the other four programmes do not cover: bachelor’s degrees in AI and machine learning and in “Physical AI” (robotics-adjacent AI), plus graduate programmes combining AI with finance, energy, medicine, industry and public administration. It enrolled 525 students in its first intake: 388 undergraduate, 32 master’s, and 105 in an AI+X Executive MBA track aimed at working professionals rather than school leavers.

QAIRU is the piece of the puzzle aimed at producing the instructors, researchers and senior specialists the other four volume-focused programmes will eventually need, rather than adding meaningfully to the raw headcount toward 5 million.

Why did Kazakhstan’s target jump from 1 million to more than 5 million in one year?

Kazakhstan’s Minister of AI and Digital Development, Zhaslan Madiyev, set a public target of training 1 million people in AI skills by 2030 on 7 January 2026, broken down as 500,000 school students, 300,000 university students, 90,000 government officials, 80,000 entrepreneurs and corporate professionals, and 30,000 people from other sectors.

Eight months later, the number presented in New York was more than 5 million over five years, a fivefold jump with no published reconciliation of the two figures. The most plausible explanation is scope: Kazakhstan also announced in mid-2026 that AI literacy content would enter the national school curriculum from 2026 onward, which would sweep in millions of students by default rather than through opt-in enrolment, unlike the named programmes in the table above. Until Kazakhstan publishes an updated breakdown by category the way it did in January, the 5 million figure should be read as a five-year policy ceiling, not a sum of currently enrolled people.

Do the enrolment numbers add up to 5 million people in five years?

Not yet, and not by a wide margin. Adding up the distinct people reported across the five named programmes (using AI-Sana’s 100,000 completed-training figure rather than its larger certificate count, since one student can hold several certificates) gives roughly 115,000 to 200,000 people trained as of September 2026, depending on whether the 76,000 university AI/IT students are counted separately from AI-Sana or treated as an overlapping population.

The named-programme headcount so far

Even the generous count, treating every university AI/IT student and every AI-Sana completer as a distinct person, lands under 200,000. That is under 4 percent of the 5 million target, roughly a year into the programmes that have published numbers.

Where the other 4.8 million would have to come from

For the target to be reachable, the bulk of the 5 million has to come from population-wide mechanisms that were not detailed in the New York presentation: the national school AI curriculum reaching the full student population by default, mass-market MOOC completions on AI-Sana’s partner platforms that a headline number does not yet capture, or looser definitions of “trained” for large-scale public-awareness modules aimed at government workers and the general public. None of that is disqualifying, national digital-literacy pushes elsewhere have used exactly this mechanism, but it does mean the 5 million figure describes a different kind of “training” than a Tomorrow School graduate receives.

Can Kazakhstan’s 90%+ employment-rate claim be verified?

Not independently, at least not from what has been published. The Tech Orda employment figure, like the other completion numbers cited in New York, is self-reported by the ministry running the programme, with no third-party audit, sample methodology, or definition of “employed” disclosed alongside it.

That does not make the number false. It does mean anyone using it, a training vendor citing Kazakhstan as a regional proof point, or another government benchmarking its own programme, should treat it as a headline claim rather than a verified statistic until Kazakhstan (or an independent body such as the World Bank or ITU, both present at the New York meeting) publishes the underlying data.

What could other Central Asian governments realistically replicate?

Three parts of Kazakhstan’s model travel well to neighbouring governments without requiring Kazakhstan’s specific budget or vendor relationships: the age-segmented delivery structure (youth hub, adult bootcamp, university tier, research tier, rather than one program for everyone), the practice of naming vendor partners publicly to borrow their credibility, and publishing category-level breakdowns of a training target the way Kazakhstan did in January, even if that breakdown later changes. A government starting from scratch would do well to run a training needs analysis before picking which of the five delivery models to copy first, since a youth hub and an adult bootcamp solve different labour-market gaps.

• Run at least one free, no-prerequisite adult bootcamp modelled on Tomorrow School’s structure before attempting a national target.
• Pair a MOOC-platform partnership with a hardware-vendor academy, since certification breadth and applied technical depth solve different parts of the skills gap.
• Publish the enrolment denominator alongside any completion or employment percentage, not after it.

The harder parts to replicate are the ones that depend on national conditions Kazakhstan already had in place: existing digital infrastructure reaching 14 regions, an established startup ecosystem through Astana Hub to absorb AI-Sana’s project pipeline, and enough IT-sector employer demand to plausibly place Tomorrow School and Tech Orda graduates at a 90 percent rate in the first place.

What does this mean for training-delivery infrastructure and corporate L&D teams in Central Asia?

For corporate L&D and training operations teams working in or evaluating Central Asia, Kazakhstan’s programme mix is a preview of the delivery infrastructure question every regional employer will face: how much AI-skills training gets absorbed from a national pipeline (Tech Orda, AI-Sana graduates) versus built in-house, and which vendor certifications (Coursera, Huawei ICT Academy, NVIDIA) a hiring pool will already carry.

A learning management system built to absorb this kind of external credentialing needs to ingest and verify certificates from multiple outside vendors rather than assuming all training happens on one internal platform, since a growing share of the Kazakh workforce will arrive holding a mix of Tech Orda, AI-Sana and Coursera credentials rather than a single employer-issued one. Teams weighing whether to build that pipeline in-house or buy it can start from the same groundwork covered in how to build a corporate training programme, and the credentialing layer itself is exactly what AI in LMS platforms are increasingly built to automate, mapping an external certificate to an internal skills taxonomy instead of retraining skills a candidate already has.

Conclusion

Kazakhstan’s 5 million AI training target is unusual for how much evidence accompanies it, five named programmes, real enrolment figures, and named vendor partners, which is more transparency than most national skilling announcements offer. It is also, on the numbers available today, a target the current delivery base reaches only a fraction of the way toward, with the gap most plausibly closed by population-wide mechanisms like the incoming school AI curriculum rather than by scaling the five named programmes alone.

If you run training operations for an organisation hiring into Kazakhstan or the wider Central Asian market, the immediate task is not to wait for the 5 million figure to resolve. Build a credential-verification step into your hiring and onboarding process now, so a Tech Orda, AI-Sana or Tomorrow School credential on a candidate’s résumé maps to something your own LMS for government-adjacent training programmes can actually validate, rather than retraining skills a candidate already has.