The $500 Billion Urban-Scale AI Layer

NEOM is not a city in the conventional sense, and that distinction matters enormously for how artificial intelligence gets deployed there. The $500 billion masterplan that the Public Investment Fund unveiled in 2017, and that has since expanded to encompass The Line, Oxagon, Trojena, Sindalah, and a half-dozen subsidiary destinations, is best understood as a single integrated computing substrate onto which urban services are layered as software. Where Riyadh, Jeddah, and Dammam grew organically over decades and now retrofit AI onto legacy infrastructure, NEOM is being instrumented at the foundation. Every meter of fiber, every chilled-water loop, every elevator shaft, every facade panel is being specified with sensing, actuation, and inference budgets in mind. This is the most ambitious greenfield smart-city deployment ever attempted, and it sits at the center of Saudi Arabia’s $77 billion AI compute buildout.

The operating company that translates the masterplan into running software is Tonomus, a wholly owned NEOM subsidiary chartered in 2022 to be the cognitive layer of the region. Tonomus is unusual among smart-city operators because it owns the data, the platforms, and the customer relationship simultaneously. Where Singapore’s Smart Nation initiative coordinates dozens of agencies and where Songdo in South Korea relies on Cisco and LG as integrators, Tonomus consolidates the role of platform owner, system integrator, and service operator into a single PIF-backed entity. That consolidation collapses procurement cycles and accelerates deployment, but it also concentrates the technical risk: every architectural decision Tonomus makes propagates across the entire NEOM portfolio.

Tonomus and the Cognitive Stack

Tonomus describes its architecture in terms of a cognitive stack with four layers: physical sensing, data fabric, AI services, and citizen-facing experiences. The physical sensing layer is built on a dense mesh of LiDAR, multispectral cameras, environmental sensors, sub-metered electrical and water infrastructure, and a 5G/6G radio access network engineered for sub-millisecond edge round-trips. The data fabric layer normalizes telemetry from those sensors into a unified semantic model. The AI services layer hosts the inference workloads — vision, time-series forecasting, optimization, and generative — that convert telemetry into decisions. The experience layer surfaces those decisions to residents through a personal AI agent, branded as the NEOM cognitive companion, and to operators through a control center analogous to a network operations center for the city itself.

Compute capacity for the AI services layer is provisioned through a combination of on-region data centers in Oxagon and shared capacity at the Hexagon facility outside Dammam operated under the Humain umbrella. Tonomus has publicly disclosed partnerships with Google Cloud for the public-facing experience layer and has been reported to be in advanced conversations with NVIDIA, AMD, and Cerebras for accelerator capacity. The on-region footprint is sized in the low hundreds of megawatts initially, scaling to a multi-gigawatt envelope across the NEOM region by 2030 — a figure consistent with the broader Humain mandate to bring 1.9 GW of AI compute online by the end of the decade.

Digital-Twin Deployment as the Organizing Primitive

The single most important architectural decision Tonomus has made is to treat the digital twin not as a visualization tool but as the system of record for the city. Every physical asset — a chiller, a tram, a building facade, a pedestrian crossing — has a corresponding digital entity that holds its design specification, its real-time telemetry, and its predicted future states. Operations and maintenance, security, public safety, and resident services all read from and write to the same twin. This is a sharp departure from the BIM-plus-GIS approach that most smart cities adopt, where the twin is a downstream artifact of the as-built drawings.

The twin is rendered in NVIDIA Omniverse and Unity for visualization, but its ground truth lives in a graph database that connects asset entities to their telemetry streams, their controlling software, and their service-level commitments. When a resident files a complaint about elevator wait times in a specific tower, the complaint is routed to the twin entity for that tower’s vertical-transportation system, which in turn knows which AI scheduling model is responsible, which maintenance contractor holds the SLA, and which sensor stream provides ground truth for verification. The twin is the lingua franca that lets a single AI agent reason across the operational domains of the city.

Urban-Scale AI Infrastructure Patterns

Three infrastructure patterns recur across NEOM’s AI deployments and are worth naming explicitly. First, edge-heavy inference: latency-sensitive workloads such as autonomous mobility, crowd flow analysis, and access control run on edge nodes co-located with the radio access network, with only model updates and aggregated telemetry flowing to the regional data center. Second, federated learning across districts: each NEOM destination — The Line, Oxagon, Trojena — trains local models against its own population and operational profile, with periodic federation to share generalizable patterns without surfacing identifiable resident data. Third, sovereign generative tier: any LLM-mediated citizen interaction is served from Saudi-resident infrastructure, fine-tuned on Arabic and the regional dialect mix, and gated by SDAIA-aligned safety policies before responses reach residents.

These patterns are not unique to NEOM in principle, but the scale and the integration are. Tonomus is the only operator in the world deploying all three simultaneously across a region that is simultaneously a sovereign jurisdiction, a private development, and a Vision 2030 priority project.

Oxagon — The Industrial AI Twin

If The Line is the residential and commercial showcase, Oxagon is the industrial spine. Oxagon’s port and manufacturing zone is being instrumented as an industrial digital twin from day one, with AI-driven optimization across container handling, warehouse robotics, advanced manufacturing cells, and the hydrogen and renewables value chains that anchor the masterplan. The Oxagon twin is operationally distinct from the resident-facing NEOM twin but federates with it on cross-cutting concerns such as workforce mobility, environmental monitoring, and emergency response.

Oxagon’s AI deployment leans heavily on the patterns proven at SABIC’s Yanbu and Jubail complexes — predictive maintenance on rotating equipment, advanced process control on chemical reactors, and computer-vision quality inspection on packaging lines — but extends them with greenfield capabilities such as autonomous yard tractors and robotic last-mile transfer between manufacturing cells and the port. The Royal Commission for Jubail and Yanbu, which has decades of operational experience running heavy industrial zones, is an informal benchmarking partner for Oxagon’s industrial AI stack.

Autonomous-Systems Integration

NEOM has committed to fully autonomous mobility as the default mode for The Line, with no private vehicles permitted within the linear city. That commitment forces the autonomous-systems stack to be production-grade from opening day, not an experimental overlay. The vehicle fleet — including pod transit, autonomous shuttles, and last-mile delivery robots — runs on a unified motion-planning and traffic-orchestration platform that ingests the digital twin as its world model. Tonomus has been actively benchmarking autonomy stacks from Mobileye, Waymo’s enterprise licensing arm, and Pony.ai, with localization fine-tuning performed on Saudi street-furniture and signage corpora.

Vertical mobility — elevators, vertiports for eVTOL services, and the linear-city’s high-speed rail spine — is treated as part of the same autonomy domain. A resident’s trip from a 200th-floor apartment to an Oxagon factory floor is planned end-to-end by a single mobility orchestrator, with handoffs between vertical, horizontal, and air modes negotiated automatically. This level of integration has no precedent in any operating city, and it creates substantial new categories of AI risk that Tonomus is working to address through extensive simulation in the Omniverse twin before any human passenger is moved.

Citizen-Services AI

The resident-facing experience is anchored by a personal cognitive companion that handles interactions across utilities, healthcare, education, retail, mobility, and government services. The companion is an Arabic-first agent fine-tuned on the dialect mix expected in NEOM’s international resident base, with a sovereign control plane that ensures personal data never leaves Saudi territory. Interactions are persisted in a personal data store that the resident controls, consistent with the spirit of the Personal Data Protection Law (PDPL) administered by SDAIA.

The companion is integrated with Absher and Tawakkalna for any service that crosses the boundary between NEOM-internal and national-government domains. It also brokers commercial interactions with on-region merchants, with payment settlement running on rails that integrate Mada, mada Pay, STC Pay, and international card networks. Tonomus has indicated that approximately 80 percent of routine resident interactions are expected to be mediated by the companion within five years of full operations.

Regulatory Considerations

NEOM operates under a special regulatory regime granted by Royal Decree, which gives it latitude to pilot AI-driven services that would face longer approval cycles in the rest of the Kingdom. SDAIA, CITC, and the National Cybersecurity Authority retain oversight, but NEOM’s regulatory desk negotiates faster sandboxing pathways for novel deployments such as autonomous air mobility and biometric access control. This regulatory posture is one of NEOM’s structural advantages and is a significant factor in vendor selection — vendors that can move at NEOM’s tempo are favored over those whose compliance cycles are pegged to slower jurisdictions.

The PDPL applies to all resident data generated within NEOM, with NEOM operating as both controller and processor for most service flows. Cross-border transfer of personal data is heavily restricted, which is a substantial reason that Tonomus has prioritized on-region compute capacity rather than relying on hyperscaler regions outside Saudi Arabia.

Deployment Timeline and Success Metrics

Tonomus’s public deployment timeline targets initial residential occupancy in The Line in 2026, with full Stage One operations by 2030. Oxagon’s port and manufacturing zones are phased in earlier, with limited container operations underway and major manufacturing tenants — including a planned hydrogen export terminal anchored by ACWA Power — coming online progressively through 2027. The compute capacity buildout is paced ahead of the resident-occupancy curve to ensure that AI services are stable when populations arrive.

The success metrics that matter to PIF and to Vision 2030 stakeholders are not the technical KPIs that vendors emphasize. They are: cost per resident-service interaction, retention of international talent, energy productivity per resident, and contribution to non-oil GDP. Tonomus reports these metrics into the broader NEOM operating dashboard that the NEOM Board reviews quarterly, and they ultimately roll up into the Vision 2030 progress reports presented to the Council of Economic and Development Affairs.

Vendor Selection and Common Pitfalls

Vendors selling into NEOM should expect a procurement posture that values architectural fit over price. Tonomus is willing to pay premium pricing for vendors that can demonstrate clean integration with the digital twin, the sovereign data fabric, and the cognitive-companion experience layer. Vendors that arrive with point solutions — a parking-management AI, a building-energy optimizer — without an integration story are routinely deferred to the integrators (Accenture, Deloitte, KPMG, NTT Data Saudi Arabia) who package them into compliant offerings.

The common pitfalls are predictable but persistent. The first is treating NEOM as a normal smart-city engagement and proposing solutions designed for retrofit deployments. The second is underestimating the data-residency requirements and assuming that hyperscaler regions in Bahrain, the UAE, or Europe will be acceptable. The third is bringing a Western-trained model and assuming it will perform adequately on Saudi populations and Arabic content without substantial fine-tuning. The fourth, and the most consequential, is mistaking the regulatory speed of NEOM for an absence of regulation — SDAIA, CITC, and NCA all retain veto authority, and a deployment that ignores them will be unwound.

Energy and Sustainability AI Across the NEOM Footprint

NEOM’s commitment to operate as a fully renewable region creates a substantial AI agenda around energy management that intersects with the broader Saudi grid-AI program. The NEOM Green Hydrogen Company joint venture between NEOM, ACWA Power, and Air Products is one of the world’s largest committed green-hydrogen export projects, and its operating envelope depends on AI-driven coupling between solar and wind generation, electrolyzer scheduling, ammonia synthesis, and the export-shipping coordination that runs through the Oxagon port. The integrated optimization is non-trivial because the renewable resource at the NEOM site is unusually high-quality but still variable, the electrolyzer fleet has substantial ramp constraints, and the downstream ammonia synthesis and shipping book have their own scheduling realities. The AI substrate that holds this together is being built jointly by NEOM, ACWA Power’s operations team, and a tier of specialist energy-AI vendors that has been steadily consolidating around the project.

Within The Line, the building-level energy-management AI is being deployed at the scale of an integrated district-energy system rather than as building-by-building optimization. The cooling load — which is the dominant residential energy consumer in the Saudi climate — is being served through district cooling from a small number of central plants, with AI-driven dispatch that combines weather forecasting, occupancy sensing, and chiller-fleet optimization. The waste-and-water flows are similarly orchestrated as a district utility rather than as building-level systems, with AI-driven demand prediction and treatment-plant optimization tying the layers together. The combined energy-water-waste AI is one of the more sophisticated district-utility deployments anywhere in the world, and it serves as a benchmark for how integrated district-utility AI can be designed when the underlying physical infrastructure is greenfield.

Workforce and Talent Considerations

The talent strategy for NEOM’s AI footprint is consequential because the operational scale and technical complexity of the deployments require a workforce that does not yet exist in the Saudi labor market in sufficient depth. Tonomus and the broader NEOM operating companies have been recruiting aggressively from the global AI talent base, with substantial recruitment from the major US and European tech centers, from the leading Asian AI hubs, and from the broader GCC and South Asia talent base. The compensation packages are highly competitive, the work-permit and residency framework is structured to attract families on multi-year horizons, and the Saudi-citizen-development pipeline is being built in parallel through partnerships with KAUST, KFUPM, Princess Nourah bint Abdulrahman University, and Prince Mohammad bin Fahd University.

The Saudization framework requires progressive Saudi-citizen employment shares across the operating companies, and the AI-talent track is one of the more challenging Saudization targets given the depth of specialist talent required. Tonomus has been investing in graduate-and-early-career programs, in mid-career re-skilling for Saudi engineers transitioning from adjacent technical disciplines, and in senior-leadership Saudization through the substantial returnee population of Saudi technical leaders who built careers in international hubs and are now relocating back to participate in Vision 2030.

For deeper reading: see Tonomus and the NEOM cognitive stack, Oxagon industrial AI, SDAIA regulatory posture, and Hexagon data center capacity.