KSA Vision 2030 AI 5-year review (2021-2026): what got built, what didn’t, what’s next

Where the starting line actually was

Vision 2030 was published in April 2016[^1]. For the first three years, AI was a sub-theme rather than a pillar. The real institutional foundations landed in a tight 2019-2020 window:

• 2019 — SDAIA (Saudi Data and AI Authority) established by Royal Order, given mandate over the national data strategy and AI policy[^2]
• 2020 — Global AI Summit hosted in Riyadh, signalling international ambition[^3]
• 2020 — NSDAI (National Strategy for Data and AI) drafted, eventually published with 2025 milestones[^4]

The question for the halfway review is not whether the starting infrastructure existed but whether the 2021-2026 execution lived up to the framework.

2021-2024: the build-out phase

This was the period of policy, platforms, and the foreign-vendor framework getting written. The concrete deliverables in this window:

• NSDAI ethical AI principles published, giving SDAIA a regulatory mandate beyond infrastructure
• National AI platforms built under SDAIA — registration, identity, cross-government data interoperability
• ALLaM v1 published in May 2024 by SDAIA on IBM watsonx (13B parameters, trained on 3T tokens, developed by NCAI within SDAIA) — the first major public release of an Arabic foundation model from the Kingdom at a credible parameter scale[^5]
• NDMO (National Data Management Office) data classification standard published, giving in-Kingdom workloads a clear taxonomy with four impact tiers (high/medium/low/none) applying to all public entities and partners[^6]

This phase was deliberately about institutional plumbing — the kind of work that is invisible to international press but determines whether the next phase can run.

2024-2026: the acceleration phase

This is the phase we are still inside as of mid-2026. The defining event was the formation of HUMAIN in May 2025 as the PIF-owned national AI champion, chaired by Crown Prince Mohammed bin Salman, which consolidated the strategic AI build into a single operating company[^7]. The notable deliverables since:

• HUMAIN partnerships announced with Nvidia (chips of record), AMD and Qualcomm (diversification compute), Groq (inference), Cisco (network fabric), AWS (cloud), with multi-year procurement commitments — the AMD/Cisco JV is structured around 1GW of AI infrastructure expected to begin operations in 2026[^8]
• Microsoft KSA datacenter region — three Azure availability zones in Eastern Province, construction complete, scheduled for Q4 2026 general availability, giving a hyperscaler-grade landing zone for regulated workloads[^9]

Read HUMAIN 2026 procurement — practical read and Sovereignty: NEOM buys AI locally for deeper reads on these.

What actually got built — be specific

If a buyer at a European AI lab asked me “what does KSA’s AI stack actually run on today, not in slides,” here is the honest list:

That is a serious list. A skeptic could argue that most of these would have been built anyway as part of digital government modernization. The Vision 2030 question is whether they were built faster, more interoperably, and with a more coherent sovereignty stance than they would have been on default trajectory. My read: yes, they were — particularly Nafath and the SDAIA platform layer. The infrastructure is real.

Items still in progress at the halfway point — honest assessment

Founder voice means I have to be honest about the misses too. As of mid-2026, these gaps are visible:

• Arabic foundation model at full frontier scale. ALLaM v1 is real and useful. But an ALLaM that competes head-to-head with GPT-class or Claude-class models in Arabic — and is exportable as a sovereign FM to other Arabic-speaking governments — is not yet shipped. HUMAIN’s 2026-2027 build is the path to that, but the deliverable is still ahead of us.
• A vibrant indigenous Saudi AI startup scene. SDAIA’s Garage and various accelerators exist. But the count of Saudi-founded AI companies at Series A+ scale, with international customers, is still small relative to the Kingdom’s AI ambition. The talent is increasingly there; the company-formation density is the lagging indicator.
• AI export and regional dominance. Vision 2030 framed KSA as a regional AI leader, but the export — Saudi-built AI products winning contracts in Egypt, the Levant, North Africa, Pakistan, Indonesia — is still nascent. The Kingdom is consuming AI faster than it is exporting it.
• Talent retention against global compensation. Saudi AI researchers are still being recruited by US and European labs at compensation levels that strain even PIF-backed counter-offers.

These are not failures. They are mid-course deliverables that need to land between 2026 and 2030. A halfway review that pretends every box was ticked would be marketing, not analysis.

What’s next: the 2026-2030 delivery window

If I were sitting in a HUMAIN or SDAIA strategy room today, the four-year delivery list would look something like this:

1. HUMAIN training compute at announced steady-state — the Nvidia + AMD + Qualcomm + Groq stack actually online, with throughput proven on next-generation training runs
2. Next-generation Arabic FM at frontier-competitive capability, with safety eval and RLHF programs scaled to international standards
3. FIFA 2034 readiness as a forcing function — crowd, mobility, hospitality, broadcast, and security AI all hardened against a global event window (FIFA formally confirmed Saudi Arabia as host on 11 December 2024)[^12]
4. Post-oil revenue diversification dependency met through non-oil sectors that AI materially enables — tourism, logistics, financial services, healthcare, mining
5. Indigenous startup density — at least 50-100 Saudi-founded AI companies at Series A+ with international customers by 2030
6. Talent layer — the Saudi AI workforce (residents + nationals + naturalised) at a scale where HUMAIN, SDAIA, and the giga-projects can hire without burning every offer through global comp ceilings

That list is achievable. None of those items is on a 10-year timeline elsewhere in the world — they are 4-year items if the institutional momentum stays where it is. The question is execution discipline, not strategic ambition.

What this means if you are an AI vendor or buyer

If you are a foreign AI vendor reading this from outside the Kingdom: the window from now to FIFA 2034 is the most concentrated procurement and partnership window the region will see in a generation. Get MISA-licensed, understand in-Kingdom vs sovereign data residency, get NCA ECC compliance, and read the procurement reality check before you assume your standard playbook works.

If you are a Saudi government or PIF-backed buyer: the gap between strategy and execution narrows when vendors with real Arabic-first capability are picked over vendors with pretty slides. Read our government and sovereign AI solutions and the foundation models solutions page. For the operating model on training data programs, see the KSA sovereign FM lab SFT/RLHF blueprint.

If you are a buyer in a neighbouring market — Kuwait, UAE, Bahrain, Oman, Qatar — the KSA blueprint is the most comprehensive sovereign AI playbook in the region. Read the Kuwait sovereign AI lead persona and the KSA smart city AV mobility persona for adjacent applications.

The honest bottom line

Five years in, Vision 2030’s AI track is in better shape than most skeptics expected and more behind schedule than most boosters admit. The institutional foundations are real. The deployment layer is real and at scale. The frontier foundation model and the indigenous startup density are the two areas where the 2030 destination still requires hard execution between now and then.

A founder voice on this: I will take this trajectory over almost any other regional alternative. The combination of capital, policy clarity, and forcing-function events (FIFA 2034, NEOM operational milestones, the post-oil narrative) creates a delivery window that does not exist elsewhere in MENA at this scale. The work between now and 2030 is the work of actually shipping — not of reframing.