AI Model Providers

The Intelligence Layer

If silicon suppliers build the compute capacity and infrastructure operators build the physical facilities, AI model providers deliver the actual intelligence — the trained models, fine-tuned systems, and application platforms that turn raw compute into outputs that are useful to enterprises, governments, and citizens.

The AI model provider landscape in Saudi Arabia is more complex than it first appears. It includes companies that train and deploy foundation models, companies that provide data and model operations platforms, companies that offer AI-powered enterprise applications, and companies whose models are used in Saudi Arabia primarily through third-party distribution channels. With 10 entities tracked, an average SCS of 5.73, and $10.3 billion in committed capex, this sector sits at the intersection of commercial technology and sovereign ambition.

The sector average SCS of 5.73 is the lowest among all sectors, which reflects the nature of model providers in the Saudi context: most are foreign entities, many do not own or operate in-kingdom infrastructure at significant scale, and their Sovereignty scores are inherently lower than those of Saudi-controlled entities. But model providers’ influence on the Saudi AI ecosystem is disproportionate to their capital commitments — the models they develop and operate are the interfaces through which Saudi enterprises will interact with AI.

The Sovereign Arabic Model: Allam

Before examining international model providers, it is important to establish what is explicitly not in this sector: Allam, Saudi Arabia’s sovereign Arabic foundation model.

Allam is developed and operated by SDAIA — the Saudi Data and AI Authority — and is therefore tracked in the Sovereign Compute Operators sector, not the AI Model Providers sector. But understanding Allam is essential context for understanding why Saudi Arabia engages with international model providers in the way it does.

Allam is a 34-billion-parameter Arabic language model trained on 8 petabytes of Arabic-language data, using 5,000 NVIDIA Blackwell GPUs. It is the most capable Arabic-native foundation model in existence, and it represents Saudi Arabia’s strategic hedge against dependency on English-centric international models for Arabic-language workloads.

For Saudi AI applications involving Arabic text — government documents, Arabic-language citizen services, Islamic finance products, Arabic-language education — Allam provides a model that is culturally appropriate, linguistically precise, and sovereignty-compliant. No international model provider can offer the combination of Arabic training data depth, cultural specificity, and Saudi data governance compliance that Allam delivers.

The existence of Allam shapes Saudi Arabia’s engagement with international model providers in a specific way: international models are valued for their technical capabilities in English-language and multilingual contexts, and for specialized domain capabilities (code, mathematics, scientific reasoning) where Allam’s current version may not be the optimal choice. Allam handles the Arabic sovereign workloads; international models handle everything else. This division of labor is not a permanent architecture — SDAIA continues to develop Allam with increasing capabilities — but it describes the near-term reality.

Databricks: The Data and Model Platform

Databricks occupies the most strategically important position among international model providers in Saudi Arabia, and its role is distinctive: Databricks is not primarily a model company. It is the platform that allows Saudi enterprises to build, fine-tune, deploy, and govern their own AI models on their own data.

Databricks’ core platform — the Lakehouse architecture combining data storage, processing, and ML operations — solves the problem that every Saudi enterprise faces when trying to deploy AI: how do you train models on proprietary data, ensure that data never leaves your infrastructure, fine-tune foundation models for your specific domain, and maintain governance over AI outputs in a regulated industry context?

For Saudi financial institutions subject to SAMA cybersecurity frameworks, for healthcare providers under MOH data protection requirements, and for government agencies under SDAIA oversight, Databricks’ platform provides the operational control and auditability that running workloads on external model APIs does not. An Aramco geoscience team can fine-tune a geological analysis model on Saudi reservoir data without that data ever touching a US-based inference endpoint. A Saudi bank can build a credit risk model on customer transaction history without creating a cross-border data transfer compliance issue.

Databricks’ DBRX model and the MLflow toolchain for model lifecycle management position it as the platform for organizations that need more than API access to someone else’s model — they need to own and operate their own AI capability. This positioning resonates strongly with Saudi Arabia’s sovereign AI strategy.

xAI and the Grok-Humain Deal

Elon Musk’s xAI and its Grok model family occupy a politically visible but technically specific position in Saudi Arabia’s model landscape. The deal between xAI and Humain — announced in conjunction with Humain’s establishment in May 2025 — includes deployment of Grok models within Humain’s sovereign AI platform.

Grok’s technical characteristics make it particularly relevant for specific Saudi use cases. Grok’s integration with X (formerly Twitter) training data provides real-time information access that static models lack. For Saudi applications requiring current events, market intelligence, and social sentiment analysis, Grok’s real-time data integration is a meaningful differentiator.

The xAI-Humain relationship also has strategic dimensions beyond the technical. xAI’s founder’s prominent profile in US tech policy circles, and the specific timing of the deal announcement concurrent with Saudi-US AI framework agreements, suggests that the Grok-Humain partnership carries diplomatic weight alongside its commercial dimensions.

Grok’s deployment within Humain’s sovereign infrastructure — rather than accessed through an external API — means that the model’s operation will be subject to Saudi data governance requirements. This structural arrangement mirrors the broader pattern of Saudi AI: international model capability delivered through Saudi-controlled infrastructure with Saudi oversight.

G42 and Jais: The UAE Competitor in Saudi Markets

G42, the Abu Dhabi AI company, brings a distinctive complication to the Saudi AI model landscape. G42 is both a competitor to Saudi AI ambitions and a potential partner — a UAE sovereign entity that has built capabilities in Arabic AI that overlap with Saudi priorities.

G42’s Jais model is an Arabic-language LLM developed in partnership with Mohamed bin Zayed University of AI (MBZUAI). At 30 billion parameters, Jais is competitive with Allam in scale, and G42’s early mover advantage in Arabic AI development means that some Saudi enterprises and partners are already using Jais for Arabic-language applications.

The Saudi-UAE AI relationship is geopolitically complex. Both countries are pursuing sovereign AI strategies with overlapping objectives and competing for the same international technology partnerships. NVIDIA’s allocation decisions, hyperscaler partner priority, and US AI framework agreements are all contested between Riyadh and Abu Dhabi. At the same time, Saudi-UAE economic integration is extensive, and both countries benefit from coordinated approaches to US AI policy.

In practice, G42 and Jais occupy a niche in Saudi AI: they are used by Saudi enterprises and joint ventures that have existing UAE partnerships, for Arabic-language applications that predate Allam’s current availability. As Allam matures and SDAIA’s distribution capabilities expand, Jais’s Saudi presence will face competitive pressure from the sovereign alternative.

Salesforce Einstein: Enterprise AI Applications

Salesforce’s Einstein AI platform represents a different category of model provider — one that delivers AI capability as an embedded feature of enterprise applications rather than as a standalone model or platform.

Salesforce is deeply embedded in Saudi enterprise commercial operations. Major Saudi retailers, banks, and telecoms use Salesforce CRM as their customer engagement system. Einstein’s AI capabilities — predictive scoring, conversational service, document processing, workflow automation — are delivered within the Salesforce platform rather than requiring enterprises to build separate AI integration. For Saudi enterprises that are already Salesforce customers, Einstein is the path of least resistance to AI-enhanced customer operations.

Saudi Arabia’s Vision 2030 program places particular emphasis on customer experience improvement in sectors including tourism (Saudi Vision’s most ambitious non-oil revenue sector), retail, and financial services. Salesforce’s CRM plus Einstein proposition is directly aligned with the customer experience transformation agenda — and Salesforce’s presence in Saudi Arabia predates the AI buildout wave, giving it established relationships and a track record of enterprise delivery.

Einstein’s AI models run on Salesforce’s own infrastructure (with in-region data processing for PDPL compliance), and Salesforce has configured its Saudi cloud region to support the data residency requirements that Saudi enterprises require. The enterprise application delivery model — AI embedded in business workflows rather than a raw model API — makes Einstein particularly accessible to enterprises that lack dedicated AI engineering teams.

DeepSeek: The Open Weights Alternative

DeepSeek, the Chinese AI company that released its DeepSeek-R1 and DeepSeek-V3 models as open weights, represents an unusual category in the Saudi AI landscape: international model capability that is freely available without licensing, API costs, or geopolitical constraints.

DeepSeek’s technical achievement — training highly capable models at dramatically lower computational cost than US-developed models — was globally significant. DeepSeek-R1’s performance on reasoning benchmarks, competitive with GPT-4 class models at a fraction of the training compute cost, demonstrated that the conventional wisdom about the capital required for frontier model development was wrong.

For Saudi AI applications, DeepSeek’s open weights models offer specific advantages:

First, sovereignty compatibility. An open weights model can be deployed on Saudi-controlled infrastructure — Humain’s GPU clusters, SDAIA’s compute environment, an enterprise’s on-premise servers — without any ongoing dependency on a foreign model provider’s API. The weights are in-country and under Saudi control once deployed.

Second, cost efficiency for inference-at-scale. DeepSeek’s models achieve strong performance with lower computational requirements than comparable US models, which reduces the inference cost for Saudi enterprises deploying AI at scale.

Third, fine-tuning flexibility. Open weights models can be fine-tuned on proprietary Saudi data without requiring data to be sent to a model provider’s training infrastructure.

The absence of direct relationship or capital commitment from DeepSeek in the Saudi market is consistent with its open-source distribution model — DeepSeek does not need a Saudi office or partnership to be used by Saudi enterprises. But its influence on the Saudi model landscape is real, and its presence signals that Saudi AI model procurement includes open weights alternatives alongside commercial partnerships.

International Models Used in Saudi Arabia Through Intermediaries

A significant portion of Saudi enterprise AI model consumption happens through hyperscaler channels rather than direct model provider relationships. OpenAI’s GPT-4 and later models are deployed in Saudi Arabia through Microsoft Azure’s OpenAI Service, with Microsoft providing the data residency guarantees that direct OpenAI API access lacks. Anthropic’s Claude models are available through AWS Bedrock with in-region processing for PDPL compliance. Google’s Gemini models are available through Google Cloud’s Vertex AI platform.

These hyperscaler-mediated model deployments are tracked in the Hyperscaler Partners sector rather than the AI Model Providers sector, because the commercial and infrastructure relationship is with the hyperscaler rather than the model developer. But they represent significant consumption of international model capability by Saudi enterprises.

The implication is that the actual AI model usage landscape in Saudi Arabia is broader than the AI Model Providers sector alone captures. Saudi enterprises use Allam for Arabic sovereign workloads, GPT-4/Claude/Gemini through hyperscaler APIs for multilingual and general-purpose workloads, Databricks for data-grounded model development, and DeepSeek open weights for cost-sensitive inference — all simultaneously.

The Sovereign AI Trust Question

Saudi Arabia’s engagement with international model providers crystallizes around a fundamental question that every CIO, minister, and AI program manager in the kingdom must answer: which AI model do you trust for sovereign workloads?

The answer is neither simple nor uniform. It depends on the workload’s sensitivity, the data involved, the regulatory requirements, and the operational context.

For the highest-sensitivity workloads — defense, intelligence, critical national infrastructure, government decision-making — the answer is Allam or Humain-controlled models running on Saudi-sovereign hardware. Foreign model providers, regardless of their in-kingdom infrastructure arrangements, cannot provide the legal and operational control that classified or operationally critical government AI requires.

For government productivity and citizen services — Arabic-language document processing, public administration workflows, citizen-facing digital services — Allam is the preferred model, with international models as supplements for English-language or specialized technical content.

For enterprise AI in regulated industries — banking, healthcare, insurance — the answer combines PDPL-compliant hyperscaler model APIs (for general applications) with fine-tuned models on Databricks-style platforms (for proprietary data applications), with Allam for Arabic-native workflows.

For commercial applications in less regulated contexts — retail AI, marketing personalization, software development tools — international models accessed through hyperscaler APIs are acceptable and commercially efficient.

This segmentation of the model trust question reflects Saudi Arabia’s AI maturity: rather than a binary sovereign/non-sovereign choice, Saudi enterprises are developing nuanced model deployment architectures that use the right model for each workload’s trust and capability requirements. International model providers that understand this segmentation — and position their offerings appropriately within it — will find a receptive and sophisticated market.

The AI model provider landscape in Saudi Arabia will evolve rapidly over the next three to five years as Allam’s capabilities expand, as Humain’s sovereign model platform matures, and as international model providers deepen their in-kingdom operational presence. The companies that establish strong technical and commercial relationships in this period will be positioned for the larger market that follows as Saudi AI adoption accelerates across the economy.