NEOM: infrastructure layer in the Kingdom’s compute build
$500B Saudi megacity on Red Sea coast. Hosts DataVolt 1.5 GW AI factory at Oxagon.
Infrastructure is the slow-moving layer of the Saudi compute thesis. Where chip orders can be placed and rerouted in quarters and where models can be retrained in months, infrastructure — the dirt, concrete, switchgear, fiber, cooling plant, and substations that turn power and silicon into deployable AI capacity — is built on multi-year cycles, gated by permits, environmental approvals, civil-engineering capacity, and grid-connection schedules. Understanding NEOM means understanding the physical layer of the Kingdom’s bet on AI.
Physical scale and siting logic
The siting decisions visible across the Saudi build prioritize four overlapping criteria: power availability and cost, cooling efficiency, fiber connectivity, and proximity to demand or to specific natural advantages. NEOM’s position in that siting matrix is determined by its specific facility design, geographic anchor, and strategic mandate. Facilities anchored in Riyadh — including Hexagon, the Humain primary campus, and Center3’s flagship — privilege proximity to government and enterprise demand, paying a slightly higher cooling cost in exchange. Facilities anchored at NEOM Oxagon — including the DataVolt 1.5 GW factory — privilege access to dedicated renewables, deep-water cooling, and a clean-build greenfield environment, paying a higher fiber-distance cost in exchange. Facilities in Dammam — including Google Cloud’s announced US$10 billion AI hub — privilege proximity to Aramco’s energy infrastructure and to fiber landings into the Gulf. Jeddah-anchored facilities privilege solar-AI hybridization and access to Red Sea subsea cable landings.
The Saudi build’s physical signature is gigawatt-scale aggregation rather than the smaller-tier distribution typical of European or US deployments. Hexagon’s 480 MW, NEOM-DataVolt’s 1.5 GW, the Google Dammam hub’s announced multi-GW horizon, and Humain’s 6+ GW aggregate target reflect a deliberate strategy: rather than dozens of 30-50 MW colos, the Kingdom is building a small number of very large facilities that can be operated as integrated AI factories with on-site renewables, dedicated substations, and tightly engineered cooling.
Technical specifications
NEOM’s technical envelope — design IT load, PUE target, cooling topology, redundancy tier, network ingress/egress, security classification — defines the workloads it can host. Air-cooled facilities optimized for traditional cloud workloads are reaching practical limits at modern AI rack densities (50-150 kW/rack with current Blackwell deployments, projected to exceed 250 kW/rack with next-generation systems), forcing a transition to direct liquid cooling and rear-door heat exchangers for any facility intending to host frontier-AI workloads at scale. Saudi facilities being built greenfield in 2025-2026 are predominantly designed for liquid cooling from inception; facilities being retrofitted face longer transition timelines.
Power-usage-effectiveness (PUE) targets in the Kingdom benefit from cool-night/hot-day diurnal cycles in inland sites and from sea-water cooling at coastal sites. NEOM-Oxagon’s design PUE under 1.15 is among the most aggressive globally; Hexagon’s design PUE under 1.25 is competitive for an inland desert deployment. The Saudi PUE position is materially better than equivalent inland US facilities (typically 1.4-1.6) and within striking distance of Nordic-region best-in-class.
Connectivity and the fiber layer
Infrastructure cannot be considered separately from the fiber and subsea-cable connectivity that links it to user populations. Saudi Arabia’s geographic position — sitting between Africa, Asia, and Europe with both Red Sea and Gulf coastlines — gives the Kingdom unusual connectivity optionality. Subsea cable landings include 2Africa (Meta-led, Red Sea landings), SEA-ME-WE 6, the PEACE Cable (Chinese-affiliated, with sensitivities), Africa-1, and others. Domestic fiber backbones operated by stc, Mobily, and Zain link the cable landings to the inland data-center corridors. NEOM’s position relative to that fiber map determines its addressable user base for low-latency workloads.
Operating model and counterparties
NEOM operates inside a specific commercial-and-regulatory perimeter. The owner-operator counterparty (Humain, Center3, DataVolt, Google Cloud, AWS, Microsoft, Aramco Digital, ALAT, or a JV among them) sets capex, hires the engineering team, and signs PPAs, fiber, and customer contracts. The customer base — which is dominated by Humain-routed Saudi enterprise and government workloads, supplemented by hyperscaler-direct customer workloads where the facility is a hyperscaler-owned region — drives revenue. The regulatory counterparty — typically SDAIA at the AI-policy level, MCIT at the cloud level, NCA at the cybersecurity level, the Saudi Power Procurement Company at the energy level — sets compliance obligations and retains ultimate operational oversight.
The cross-cutting commercial relationships that define NEOM include: power-purchase agreements with ACWA Power and the Saudi Power Procurement Company; dark-fiber and IP-transit agreements with stc and Center3-affiliated carriers; silicon-supply agreements with NVIDIA, AMD, Qualcomm, Cisco, and the systems integrators (Lenovo, Supermicro, Dell, Foxconn); software and platform agreements with hyperscalers and with Humain Cloud.
Strategic implications
NEOM’s significance to the Saudi compute thesis is shaped by three strategic vectors. First, capacity expansion: each gigawatt of greenfield AI infrastructure brought online expands the Kingdom’s effective compute supply and either pulls workloads in from offshore or unlocks domestic demand previously constrained by capacity. Second, capability concentration: facilities like NEOM-DataVolt and Hexagon are being engineered to host frontier-class workloads (training runs, RLHF passes, agentic-AI orchestration) that require specific networking, cooling, and power architectures. Third, sovereignty positioning: facilities physically located in Saudi soil and operated by Saudi-controlled entities count toward the Kingdom’s compute-sovereignty score in a way that hyperscaler-region capacity does not.
Risks and constraints
The infrastructure layer faces a specific risk profile. Power-availability risk: the renewable-generation deployment driving the National Renewable Energy Program must keep pace with data-center load, and any meaningful slip would force either fossil-fuel bridging (compromising net-zero positioning) or load curtailment. Civil-engineering capacity risk: the Saudi construction sector is simultaneously absorbing the NEOM build, the Red Sea projects, the Riyadh metropolitan expansion, and the Vision 2030 megaprojects, creating cost inflation and schedule pressure. Climate-and-water risk: cooling-water availability and ambient-temperature profiles affect long-term operational economics. Cyber risk: large concentrated AI facilities are high-value targets, and the threat profile from regional state and non-state actors is non-trivial.
Mitigations are visible. ACWA Power’s renewable build is sized for over-procurement in early years to cushion ramp risk. The Saudi government is fast-tracking permitting for designated AI-infrastructure projects. NCA’s classification regime applies enhanced cybersecurity controls to gigawatt-class AI facilities. And the multi-site distribution across NEOM, Riyadh, Dammam, and Jeddah reduces single-site disruption risk.
What to track through 2027
Three observable indicators will tell whether NEOM is hitting its stride. First, energization milestones — the date and load at which each phase comes online. Second, customer disclosures — Humain, hyperscaler, and major-enterprise public statements about contracted capacity. Third, networking and fiber buildouts — specifically dark-fiber commissioning between facilities and to subsea-cable landings.
Construction-and-commissioning timelines
Gigawatt-class facilities like those associated with NEOM run on construction timelines that compress as much as possible against industry baselines but cannot defy physics. From greenfield site selection through full energization, a single gigawatt of AI infrastructure typically consumes 24-42 months: 6-9 months for permitting, environmental review, and final design; 12-18 months for civil works, structural buildout, and major-equipment delivery; 6-12 months for commissioning, testing, and progressive load ramp; and 3-6 months for tenant fit-out and customer onboarding. Saudi facilities are operating at the compressed end of that range, helped by the priority-permitting status that designated AI-infrastructure projects receive and by the coordinated timing of grid additions, fiber runs, and building construction.
The Saudi construction-and-commissioning advantage erodes at scale. Multiple simultaneous gigawatt-class projects — Hexagon, NEOM-DataVolt, Google Dammam, the Humain Riyadh primary, the Center3 expansion — compete for the same civil-engineering capacity, the same long-lead electrical equipment (transformers, switchgear, GIS), and the same specialized commissioning labor. Bottlenecks in any of those inputs create schedule slip that ripples across multiple projects. The macro pattern visible across 2025-2026 is that announced energization dates have slipped by an average of 4-7 months versus original announcement timelines, which is favorable by global benchmarks but still material for capital-deployment math.
Networking and the cluster-scale problem
Compute-bound AI workloads at frontier scale require not just power and cooling but also networking that supports the cluster-scale data movement of large training runs. NEOM’s networking architecture — the choice of InfiniBand NDR/XDR versus Ethernet-based fabrics like Cisco Silicon One or NVIDIA Spectrum-X, the topology design (fat-tree, dragonfly, hybrid), the cross-rack and cross-row bandwidth provisioning — determines whether the facility can host frontier-class training jobs or only inference and smaller-scale workloads.
The Saudi facilities being built greenfield in 2025-2026 are predominantly designed for InfiniBand-based fabrics with cross-cluster bandwidth provisioned at frontier-class levels (multiple terabits per second between rack rows, sub-microsecond latency targets within racks). That design choice locks in NVIDIA’s networking ecosystem at the cluster level even when the silicon mix includes AMD or Qualcomm components, which has implications for the long-arc lock-in of the Saudi compute stack.
Cooling architectures and the liquid transition
The transition from air cooling to direct liquid cooling is one of the most consequential infrastructure-engineering shifts of the 2025-2030 cycle. Air-cooled facilities are reaching practical limits at modern AI rack densities, and any facility intending to host next-generation accelerators (B300, MI355X, AI250) at scale must support liquid cooling. The specific liquid-cooling approach — rear-door heat exchangers, direct-to-chip cold plates, full immersion — affects construction cost, operational complexity, and ultimate density.
For NEOM, the cooling-architecture decision is among the most strategically dispositive engineering choices. Facilities designed for direct-to-chip liquid cooling can support 250+ kW racks; rear-door-only facilities are capped at roughly 150 kW; full-immersion facilities can theoretically scale higher but introduce operational complexity that few operators have mastered at gigawatt scale. The Saudi greenfield builds are predominantly opting for direct-to-chip plus secondary loop architectures, with NEOM-DataVolt experimenting with selected immersion cells.
Tenant mix and revenue model
Whether NEOM is a single-tenant facility (a captive build for a single customer like Humain or a hyperscaler), a multi-tenant colocation property (Center3-style), or a hybrid (anchor tenant plus secondary tenants) shapes the operational economics and risk profile. Single-tenant facilities have lower revenue volatility but higher concentration risk; multi-tenant facilities have higher volatility but more diversified counterparty risk. Hybrid models — increasingly common for Saudi greenfield projects — capture some of each.
Revenue models inside NEOM typically include: power-and-space wholesale (the colocation baseline); managed-services upsells (cooling, fire-suppression, physical security, remote-hands); networking and interconnection (cross-connects, peering, IP transit); and increasingly platform-layer services (managed AI infrastructure, sovereign-cloud overlays, model-hosting). The mix of those revenue streams determines the facility’s gross margin profile and its sensitivity to tenant churn.
Security architecture and threat model
Saudi gigawatt-class AI facilities face a sophisticated threat profile. Physical threats — drones, ground intrusion, supply-chain tampering — are addressed through layered perimeter security, drone-detection-and-defeat systems, controlled access at multiple checkpoints, and supply-chain vetting that draws on NCA classification standards. Cyber threats — both opportunistic and state-actor-grade — are addressed through air-gapped management networks for sensitive workloads, hardware roots of trust, encrypted-at-rest storage, and continuous monitoring keyed to the specific threat actors NCA tracks.
The Saudi-specific threat-model addition involves the geopolitical sensitivities of hosting US-origin advanced silicon under US export-control jurisdiction. Facilities like NEOM must demonstrate that the silicon cannot be diverted, accessed remotely by non-authorized parties, or repurposed in ways that violate the November 2025 framework’s anti-diversion commitments. That requirement drives specific architectural choices — physical isolation of frontier-silicon zones, enhanced access logging, audit-ready record-keeping — that are not standard in non-Saudi peer facilities.
Final analytical frame
Three closing points anchor the senior-analyst read on NEOM. First, the November 2025 US-Saudi compact reset the operating envelope inside which NEOM functions, and the durability of that reset through future US administration cycles is the single most important exogenous variable for NEOM’s 2026-2030 trajectory. Second, the institutional infrastructure surrounding NEOM — SDAIA’s policy throughput, Humain’s operating discipline, PIF’s capital deployment, the broader Saudi sovereign-architecture’s coordination capacity — is more sophisticated in 2026 than even informed observers expected as recently as 2023, and that institutional maturation is a compounding asset that should be priced into long-arc forecasts. Third, the gap between announcement and execution is real but narrowing, and the disciplined analyst tracks both vectors rather than treating them as equivalent.
For NEOM specifically, the cumulative read across capacity, capital, capability, sovereignty, and talent dimensions is positive on a base-case forecast, with material upside in scenarios where the post-November-2025 framework is extended, formalized, and supplemented by additional bilateral and multilateral arrangements. The principal downside scenarios involve geopolitical reversal, oil-price stress, or execution slippage on the underlying infrastructure builds — each is meaningful but each is also actively mitigated by visible Saudi-side policy and operational responses.
Cross-references in the saudicompute.com graph
NEOM interacts with a defined set of adjacent concepts and entities that working analysts should track in conjunction. The strongest cross-reference relationships connect NEOM to the sovereign-layer principals (SDAIA, PIF, Humain), to the operational counterparties (the major data-center operators, the major silicon vendors, the major cloud platforms), to the policy framework (BIS export controls, PDPL, the Major Non-NATO Ally framework, Vision 2030), and to the comparative reference points (G42, Mubadala, Stargate, the broader Gulf and OECD AI ecosystem).
The graph-based reading discipline — treating NEOM as a node with weighted edges to each of those adjacent entities — produces materially better analytical output than reading NEOM as a standalone unit. The saudicompute.com infrastructure is built around that graph-based reading, with the entity directory, the methodology page, the capital-flows page, and the policy tracker all operating as different views into the same underlying graph.
Closing on signal-vs-noise
The Saudi AI ecosystem in 2026 generates an enormous volume of public signal — press releases, conference announcements, vendor disclosures, analyst-firm reports, social-media coverage. The analyst’s task is not to consume more signal but to filter for the highest-quality data and to triangulate across independent sources. For NEOM, the highest-quality signal categories are: regulatory and customs filings (which lag announcement but reflect real flows); senior-counterparty financial disclosures (US 10-Q filings of major vendors, Tadawul disclosures of Saudi-listed counterparts); operational milestones (energization dates, customer-go-live dates, capacity-online dates); and the relationship-level intelligence available through serious engagement with the Saudi market over multiple cycles.
Practitioners who maintain that filtering discipline build a meaningfully better understanding of NEOM’s real position and trajectory than the broader market consensus reflects, and that informational edge is one of the principal value propositions of the saudicompute.com analytical infrastructure.
For deeper reading: Data center capacity tracker · NEOM compute thesis · ACWA Power and energy · Subsea cable infrastructure.