The race to build Arabic AI is entering a new phase

A gap that global models haven't closed Despite years of rapid progress in large language models, Arabic-language performance has consistently lagged behind English across most general-purpose AI systems — a gap that shows up in everything from dialectal understanding to right-to-left text handling to cultural and religious nuance. That gap has real commercial consequences for any business trying to deploy AI for Arabic-speaking customers, and closing it has become one of the clearer strategic priorities inside Saudi Arabia's wider AI push, reflected in the Kingdom's own reporting on its national AI platform ( my.gov.sa ). The persistence of this gap is itself notable given how much overall progress large language models have made in recent years. Improvements in general reasoning and English-language fluency have not automatically translated into equivalent gains in Arabic, which suggests the underlying problem is less about model scale generally and more about the volume and quality of Arabic-language training data and evaluation available to model developers. ALLAM and the sovereign-model approach Saudi Arabia's approach to closing the Arabic AI gap has run on two parallel tracks. One is direct model development aimed specifically at Arabic — an effort that reflects the view that general-purpose global models, however capable in English, cannot simply be patched into strong Arabic performance without dedicated training and evaluation focused on the language itself. That track sits alongside the country's broader "sovereign AI" ambitions around data, compute and models, which TechScoop has broken down in What does "sovereign AI" actually mean in Saudi Arabia? . Direct model development of this kind is resource-intensive, requiring not just compute but large, carefully curated Arabic-language datasets and evaluation benchmarks that reflect the language's dialectal diversity across the wider Arabic-speaking world, rather than a single standardised form of the language. Investing in companies that already have Arabic capability The second track, and the one that has produced the most concrete news in 2026, is investment: rather than building every Arabic AI capability from scratch, HUMAIN has taken a strategic stake in Arabic.AI, a company built specifically around Arabic-language AI, according to Wamda 's reporting — a deal TechScoop covered in full in Arabic.AI secures strategic investment from HUMAIN . The same period saw HUMAIN also make a comparable strategic investment in MOZN, whose financial-crime and knowledge-intelligence products — used by more than 150 customers — depend on the kind of language and knowledge processing that Arabic-language capability underpins, covered in HUMAIN makes strategic investment in Saudi AI company MOZN . This investment-led approach effectively lets HUMAIN move faster than pure in-house research would allow, since Arabic.AI has presumably already accumulated Arabic-language data, evaluation infrastructure and engineering expertise specific to the problem, rather than HUMAIN needing to build that base of knowledge from a standing start. Why enterprises care about this specifically For enterprise customers — banks, government bodies, telecoms and retailers serving Arabic-speaking populations — the practical stakes of the Arabic AI gap are direct: a customer-service AI tool that misunderstands Arabic dialect, or a document-processing system that fails on Arabic-script contracts, is not simply a lesser product, it is one that cannot be deployed at all for large parts of its intended customer base. That is why closing this gap has become treated less as a research problem and more as a commercial precondition for AI adoption across the region, and why companies solving it specifically — rather than general AI platforms hoping Arabic support improves incidentally — have drawn strategic attention from an entity as central to the ecosystem as HUMAIN. The stakes are compounded by the sheer size of the addressable market: any enterprise operating across Saudi Arabia and the wider Gulf and Levant is, by definition, serving an overwhelmingly Arabic-speaking customer base, which means an AI product's Arabic performance is not a secondary feature but close to the core requirement for adoption at scale. What "a new phase" actually looks like What marks 2026 as a new phase in this race is the shift from research announcements about Arabic-language model benchmarks toward capital being deployed directly into companies with working Arabic AI products and, in MOZN's case, an existing base of paying customers. The race to build Arabic AI is no longer only about whose model scores highest on a language benchmark — it is increasingly about whose Arabic AI capability is actually embedded inside products enterprises are already buying. That shift also changes how the race should be judged going forward: benchmark scores on academic Arabic-language tests matter less than which vendors' Arabic