The World’s Largest Government Data Center
The Hexagon Data Center, located in Riyadh and operational from early 2026, is sized at 480 megawatts — the largest sovereign government data center anywhere. The facility is operated under SDAIA authority and supports the National Data Lake, the integrated database covering 430+ Saudi government systems that anchors the Kingdom’s sovereign AI strategy. The name is literal: viewed from above, the facility’s architectural footprint is hexagonal — an unusual departure from the rectangular floor plans that data-center design conventionally favors for modular expansion, and a signal that this building was conceived as a statement as much as a plant.
In most national AI strategies, government data infrastructure is a footnote — plumbing maintained by a public-sector IT function that trails commercial technology by years. Hexagon inverts that pattern deliberately. It is not national IT infrastructure by default but national AI infrastructure by design: the physical anchor for the Kingdom’s most sensitive workloads, the training substrate for its government AI models, and the demonstration piece for the sovereign-compute architecture Saudi Arabia intends to show the world.
The Scale Claim, Interrogated
The 480 MW capacity is roughly 4-5x the size of comparable government facilities in other major economies. The US government’s largest dedicated data center facilities — the NSA Utah Data Center is the most-discussed — are estimated at roughly 100-150 MW. The UK’s GCHQ data center capacity is below 50 MW. EU member-state government compute is fragmented across smaller facilities. Hexagon’s scale is genuinely unprecedented for sovereign-owned infrastructure, and most national government data centers in developed economies operate in the tens of megawatts, which makes the gap not incremental but categorical.
Within Saudi Arabia’s own pipeline, Hexagon is approximately 2.4 times the size of Humain’s Riyadh Phase 1 campus (200 MW), more than 1.5 times Humain’s planned Dammam campus (300 MW), and 2.4 times the GDH Riyadh campus (200 MW). Among Saudi facilities, only DataVolt’s planned 1.5 GW NEOM project exceeds it — and DataVolt targets 2028 while Hexagon comes online in early 2026, which makes Hexagon the largest operational facility in the Kingdom for the critical 2026-2027 window when the buildout converts from announcement to operation.
One honest caveat belongs in any analysis: the 480 MW figure likely describes full-buildout rating rather than day-one operational load. Early-stage data centers typically run at a fraction of rated capacity while halls are fitted out and workloads migrate. The distinction does not diminish the statement of intent — the power, land, and cooling architecture are sized for 480 MW — but analysts tracking actual Saudi compute capacity should distinguish rated from utilized megawatts here as everywhere in the pipeline.
What It Hosts
The facility’s primary workload is the National Data Lake — the SDAIA-operated integration layer covering 430+ Saudi government IT systems. The Data Lake serves as the substrate for the SDAIA AI initiatives: ministry-level analytics, cross-agency search, AI-driven decision support, and (increasingly) sovereign-controlled foundation model training using Saudi government data.
Beyond the Data Lake, the facility hosts the SDAIA sovereign AI factory — up to 5,000 NVIDIA Blackwell GPUs deployed for government-controlled AI training and inference. The GPU allocation is separate from Humain’s commercial fleet and operates under different access controls; SDAIA workloads run on SDAIA-controlled compute, with Humain’s commercial customers using Humain-operated facilities elsewhere. The separation is structural, negotiated as a distinct component of the broader NVIDIA partnership: a dual-track architecture in which the commercial Humain fleet and the sovereign SDAIA fleet draw from the same silicon pipeline but never share an operator, an access-control regime, or a customer base.
Inside the National Data Lake
The Data Lake is the more strategically significant and less discussed half of the facility’s mission. Categorically, it hosts demographic data (citizen records, residency information), economic data (tax records, employment, business registrations), services data (healthcare, education, social services), infrastructure data (energy consumption, transportation flows, water utilization), and operational data (ministry budget execution, procurement, government employment). Aggregate volume runs to multiple petabytes and grows continuously as systems are integrated.
The integration is the value. Pre-Data-Lake, a question like how many citizens receive housing subsidies and also use government healthcare required separate queries to multiple ministries, manual reconciliation, and weeks of staff time; post-Data-Lake it is a single analytical query. For AI specifically, cross-ministry training data produces capabilities single-ministry models cannot match — a citizen-services agent trained on the full Lake can route citizens between ministries and surface cross-cutting policy patterns that siloed data hides. The engineering achievement should not be understated: integrating 430+ systems built across decades, with incompatible formats and proprietary standards, is a multi-year program that most developed-economy governments have attempted and quietly abandoned. SDAIA has been at it since its 2019 establishment, and Hexagon is the physical consolidation of that work. The open analytical question — whether the 430-systems figure reflects genuine interoperability or ingestion without normalization — is the single most important unknown in assessing the Lake’s real AI readiness.
The Sovereign AI Factory
The 5,000-Blackwell deployment makes Hexagon the answer to a question every AI-ambitious state eventually faces: where do you train models on data you cannot let leave your control? Training government-specific models — legal Arabic, regulatory analysis, citizen-services assistants, defense-adjacent applications — on proprietary state data requires that compute and data co-reside in a controlled environment. Sending sensitive training corpora to a commercial cloud, even one with Saudi data residency, introduces risk vectors a sovereign facility eliminates. A properly networked 5,000-GPU Blackwell cluster is a training resource comparable to large commercial AI research labs; Saudi Arabia has, in effect, equipped its government with frontier-grade training capacity as a permanent institutional capability.
The hardware generation matters to the claim. Blackwell — the GB200/GB300 line cleared for Saudi delivery under the November 2025 export approval — is NVIDIA’s most advanced training and inference architecture, with per-GPU performance on transformer workloads substantially above the prior Hopper generation. Most government compute worldwide runs years behind the commercial frontier; Hexagon’s factory runs the same silicon generation the US hyperscalers are deploying, procured through the same pipeline that feeds Humain’s commercial campuses. The 5,000-unit allocation is a fraction of the 35,000-system clearance and smaller still against the 600,000-GPU multi-year pipeline, but it is the fraction with the strictest custody: pre-approved deployment sites, personnel access authorization, and the Chinese-equipment exclusion that conditions the entire export framework apply with full force inside a facility built explicitly for state data.
The factory sits atop an accumulating SDAIA stack. The SambaNova partnership ($140M, 2024) supplied earlier AI training infrastructure and supported Arabic model development; the Databricks-SDAIA platform deal ($500M) supplies the enterprise data-and-AI tooling layer for government analytics; Allam — the Kingdom’s flagship Arabic LLM, developed under SDAIA before its commercial deployment through Humain — is the proof that the stack produces models, not just architecture diagrams. Hexagon consolidates these threads under one roof and one authority.
Why Sovereign Compute at This Scale
Three considerations drove the Hexagon scale. First, data integration: combining 430+ government systems into a unified analytical substrate requires substantial compute, both for integration ETL and for the analytics on top. Second, AI training: government-specific models touching sensitive data cannot be outsourced for sovereignty reasons, and the compute budget for a serious national modeling program is measured in hundreds of megawatts once inference deployment across ministries is counted. Third, demonstration effect: the facility’s scale signals to international observers — and to Saudi citizens — that the Kingdom operates at compute parity with global powers. The demonstration logic is not vanity; in a region where the UAE’s G42 program competes for the same third-country partnerships, visible sovereign capacity is a diplomatic asset with measurable commercial consequences.
The Compliance Architecture
Sovereign designation is an operational regime, not a label. Government-grade hosting at Hexagon’s tier implies physical security to state standards — perimeter controls, biometric access logging for data-hall entry, CCTV with government-specified retention — and network architecture that physically isolates government tenant infrastructure: separate switching fabrics rather than logical separation, with controlled and audited interconnection points to government networks.
The personnel dimension is the operationally hardest. Sensitive government workloads can require that all staff with physical or logical access be Saudi nationals holding applicable clearances — and at 480 MW, operations require hundreds of staff, which means building a large cleared national workforce through multi-year recruitment and training. That requirement doubles as Vision 2030 Saudization policy: the sovereign facility is also a forced-investment program in Saudi technical labor. Audit obligations complete the regime — access events, change management, and incident records retained in regulator-specified formats with custody chains that allow full reconstruction of any data-processing event. This is the machinery that makes claims about PDPL and KSA-RoD compliance demonstrable rather than rhetorical.
Why Riyadh
The geographic placement in Riyadh, rather than Dammam or NEOM, reflects the proximity-to-government calculation: most ministries are headquartered in Riyadh, and reducing latency to government users matters for the Data Lake’s integration role. Integration workloads are chatty — continuous synchronization with hundreds of ministry systems — and the operational relationship between SDAIA and its ministry counterparts is administrative as much as technical; co-location in the capital serves both. The placement also concentrates the Kingdom’s compute geography deliberately: Riyadh for government and commercial compute, Dammam for the Google Cloud hub on Eastern Province energy, the Red Sea coast for NEOM-DataVolt’s renewable scale play. Hexagon is the capital-city anchor of that three-node map.
The Operational Frame
Hexagon coming operational in early 2026 marks one of the structural milestones of the Year of AI 2026 — the cabinet-decreed program year that commits every ministry to AI deployment milestones. The transition from Saudi government compute fragmented across ministry-level data centers to Saudi government compute unified at hyperscale under SDAIA is a sovereignty step on par with the establishment of SDAIA itself in 2019. It is also the enabling condition for the ministry milestones: cabinet mandates for AI deployment in finance, health, education, and transportation presuppose somewhere for those workloads to run under acceptable controls, and Hexagon is that somewhere.
The facility’s significance extends beyond raw compute capacity. It establishes the architectural pattern for sovereign AI: a single, large, government-controlled facility hosting both the integration layer and the AI training infrastructure, with Humain’s commercial fleet operating in parallel for the private-sector compute market. The two-tier structure — sovereign compute and commercial compute under separate operators — is the template Saudi Arabia is exporting to other markets.
What to Watch
Four indicators will show whether Hexagon delivers its mission rather than its ribbon-cutting. First, utilization against rating: how quickly actual IT load climbs toward the 480 MW design figure, which is the honest measure of the facility’s operational reality. Second, integration depth: whether the Data Lake’s 430+ systems progress from ingestion toward genuine normalized interoperability — observable indirectly through the sophistication of cross-ministry services the government ships during the Year of AI 2026. Third, workforce: the cleared-Saudi-national staffing requirement is the binding operational constraint, and recruitment pace will gate how much of the facility can carry the most sensitive workloads. Fourth, model output: the SDAIA factory justifies itself through what it trains — successors and specializations of Allam, ministry-domain models, and the government AI services the cabinet decree demands.
The comparison to watch abroad is Abu Dhabi. The UAE AI Office is SDAIA’s institutional analog, but the Emirates has announced no sovereign government facility in Hexagon’s class; its government AI capacity runs substantially through G42’s commercial infrastructure and the Microsoft relationship. If Hexagon works, it becomes the strongest argument in the Saudi export pitch — evidence that full-depth sovereign compute is buildable, operable, and worth the cost. If it stalls at low utilization or shallow integration, the UAE’s leaner partner-dependent model gains the argument instead.
The Exportable Template
The template deserves the last word, because it is the part of Hexagon with consequences beyond the Kingdom. Most states confronting AI sovereignty face a false binary: build everything domestically at prohibitive cost, or rent from hyperscalers and accept the control loss. The Saudi answer is a structured middle path — sovereign facility for state data and state models, commercial partner fleet for the open market, hard operator separation between them, and shared access to the same silicon supply chain negotiated at national scale. The AWS Humain AI Zone applies the same philosophy inside a hyperscaler region; Hexagon applies it at full sovereign depth.
For the dozens of governments now drafting sovereign AI programs without Saudi capital, the full 480 MW expression is out of reach — but the pattern is not, and patterns travel. If the Hexagon architecture becomes the reference design for sovereign AI infrastructure the way GovCloud became the reference for government cloud, Saudi Arabia will have exported something more durable than capacity: the standard itself, authored in Riyadh, running since early 2026, at a scale no other government has matched.