ALKHOBAR: When unveiled Allam, its homegrown Arabic large language model, it sent a clear signal: the Kingdom is no longer content to simply consume global AI technologies.
It intends to build its own. For many, this was a moment of pride — a proof that the Arab world can produce tools designed to understand its own languages, cultures, and contexts.
But experts caution that Allam is only the first step in a much longer journey. Success will not be determined by the models alone, but by the invisible foundations that support them: data, infrastructure, governance, and trust.
“You can’t capture the intent, emotion, and cultural depth of Arabic through translation,” said David Barber, director of the UCL Centre for Artificial Intelligence and Distinguished Scientist at UiPath. “You need systems that think in Arabic from the ground up.”
David Barber, director, UCL Centre for Artificial Intelligence; distinguished scientist at UiPath. (Supplied)
Barber highlights a stark reality: only about 15 percent of Arabic text online is clean enough for training a large language model, compared with over 50 percent for English — a huge head start for models like GPT or Claude. Complicating matters further are Arabic’s complex grammar, diverse dialects, and the common mixing of English and Arabic in a single sentence.
“When you train on noisy or shallow data, the system learns shortcuts,” Barber explained. “It can mimic fluency, but it misses the depth, the idioms, the cultural nuances, the rhythm of thought that makes Arabic distinct.”
For Barber, this underscores the importance of ’s push for locally sourced, high-quality datasets. Without them, any Arabic LLM risks becoming a shallow copy of English-language AI: competent at generic tasks but unable to capture the soul of the language it claims to represent.
Even the best data is ineffective if it cannot be properly organized, secured, and delivered to the model. Seema Alidily, regional director at Denodo, said Gulf enterprises still face major challenges here.
“Without localized infrastructure, AI systems risk misunderstanding user intent or producing irrelevant outputs,” she said. “Data virtualization is one of the few ways to unify governance and access across cloud and on-site systems without moving sensitive information.”
Seema Alidily, regional director, Denodo. (Supplied)
Practically, this means investing in platforms that can pull data from dozens of scattered sources — from ERP systems to IoT sensors— and present it in a unified view for AI to use. In , where Vision 2030 projects depend on massive, real-time datasets, this approach is critical, especially given strict regulations on handling citizen data.
Alidily warned that merely replicating Western infrastructure may not suffice. “In the Gulf, centralized visibility and compliance must come first,” she noted. “It is not just a technical issue, it is about aligning with the legal, cultural, and regulatory expectations of the region.”
For Bader AlBahaian, country manager for at VAST Data, the stakes go beyond efficiency — they touch on independence and security.
“If we depend exclusively on external platforms, we risk importing their policies and their priorities, often at the expense of regional needs,” he said.
Bader AlBahaian, country manager, , VAST Data. (Supplied)
AlBahaian advocates for “sovereign-by-design” systems: storage and compute architectures that keep sensitive data within national borders, encryption and access controls that satisfy local regulators, and AI models trained under rules set by the Kingdom rather than a foreign vendor.
“It is not just about where the data sits,” he added. “It is about who gets to define how it is used, who takes responsibility when something goes wrong, and who has the power to switch the system off if necessary.”
This question of sovereignty is becoming urgent as AI begins to shape decisions in finance, healthcare, education, and public policy. A misaligned model trained on foreign data could issue recommendations that contradict local priorities — or worse, expose the region to economic or political risks.
But building perfect infrastructure is only half the challenge. Success ultimately depends on how AI is deployed.
“Digital labor will allow businesses to have much deeper relationships with their customers,” said Ibrahim Alseghayr, managing director of Salesforce . “And by taking on so much of the routine work, AI frees humans to focus on collaboration, creativity, and critical thinking.”
Ibrahim Alseghayr, managing director of Salesforce . (Supplied)
Alseghayr points to Agentic AI — systems that can act on a company’s behalf — as already transforming service centers, financial operations, and citizen engagement platforms. In , he sees huge potential for digital labor in scaling mega-projects like Neom, automating logistics networks, and delivering smarter healthcare services.
He cautioned that this transformation must be carefully managed. “We need strong governance, testing environments, and continuous oversight,” he said. “Otherwise, we risk building tools we do not fully understand, and that could erode trust instead of building it.”
Across all four experts, one theme is clear: global rules and imported frameworks will not suffice. The Arab world must craft its own AI governance models, rooted in its cultural and legal realities.
For Barber, Allam is a test case. “This is the Kingdom’s chance to prove that it can build systems that are not only technically powerful but also aligned with its values,” he added.
DID YOU KNOW?
• Arabic’s complex grammar, dialect diversity, and frequent English–Arabic mixing make it one of the hardest languages for AI to master.
• ’s Allam is the first homegrown Arabic large language model, designed to think in Arabic rather than translate from English.
• Vision 2030 projects depend on real-time data, but regulations require strict handling of citizen information.
“Agentic AI can create personalized treatment plans, autonomously monitor patients, and detect early signs of health deterioration before a doctor ever enters the room,” he said.Alidily agrees, emphasizing that governance frameworks must reflect the Gulf’s unique data protection requirements, with regulators working closely with technology providers to define shared standards.
AlBahaian is even more direct. “Trust is earned through systems, not slogans. People need to know where their data is, who is using it, and for what purpose. That is the only way to build confidence at scale.”
The message is clear: Arabic AI’s future will not be decided by model size alone. It will depend on investments in infrastructure, sovereignty, and governance.
has taken the first step with Allam. What comes next — the data pipelines, virtualized infrastructure, sovereign controls, and digital labor deployments — will determine whether the Kingdom becomes a true AI creator or remains a buyer of foreign-built intelligence.
