DataVolt: Building the World’s Largest Net-Zero AI Factory at NEOM
DataVolt’s Oxagon AI factory project represents one of the most ambitious infrastructure bets in the global AI compute buildout — and one of the most uncertain. A $5 billion commitment to 1.5 gigawatts of AI compute capacity in a NEOM industrial zone powered entirely by renewable energy, positioned as the world’s largest net-zero AI data center, is the kind of project that either anchors a new era of sustainable AI infrastructure or becomes a cautionary study in megaproject execution risk. The case for DataVolt’s success rests on genuine structural advantages in site selection, power economics, and strategic positioning within Saudi Arabia’s AI ambitions. The case for skepticism rests on NEOM’s well-documented track record of scope reductions and timeline delays. Both cases deserve analytical rigor.
The Project: Scale, Location, and Power
DataVolt’s Oxagon AI factory is sited at NEOM’s Oxagon development — a floating and coastal industrial complex on the Red Sea near the Saudi-Egyptian border in the Tabuk region of northwest Saudi Arabia. Oxagon is designed as NEOM’s manufacturing and logistics hub, intended to blend advanced manufacturing with port logistics, supported by NEOM’s green energy infrastructure and connected to the planned industrial ecosystem NEOM is developing.
The project parameters are exceptional by any standard. The $5 billion investment target funds construction of AI compute infrastructure targeting 1.5 GW of capacity by 2028. To put this in context, a typical hyperscale data center development runs 100–400 MW over multiple years; DataVolt’s Oxagon facility would be five to fifteen times that scale in a single location. The SDAIA Hexagon data center in Riyadh, Saudi Arabia’s flagship government AI facility, operates at 480 MW — DataVolt’s Oxagon target is more than three times larger.
Power sourcing is the project’s defining characteristic and its most important differentiator. The facility is designed to run on 100% renewable energy from the first day of operation — not via renewable energy certificates offset against grid power, but through direct power purchase agreements with dedicated renewable generation. ACWA Power, Saudi Arabia’s dominant renewable energy developer (itself partially PIF-owned), is the primary power supply partner, providing solar and wind generation dedicated to the Oxagon facility. NEOM’s broader green energy infrastructure, including its Helios green hydrogen project, provides additional renewable capacity and the energy storage and grid balancing services needed to maintain 24/7 operations on intermittent renewable sources.
This is genuinely different from most data center “green” claims. The majority of data center operators claiming carbon neutrality are doing so via renewable energy certificate purchases or grid-scale carbon offsets, not through dedicated renewable generation. DataVolt’s architecture — AI factory co-located with dedicated renewable generation in a purpose-built industrial zone — is designed to be net-zero on a physical energy basis, not just on an accounting basis.
Why Oxagon? The Strategic Site Logic
The site selection logic for an AI factory at Oxagon is worth unpacking, because on the surface it seems counterintuitive. The world’s major AI compute concentrations are near major population centers (Northern Virginia, Silicon Valley, Dublin, Singapore, Tokyo) or at least near major grid infrastructure and fiber connectivity (Iceland, Sweden, Oregon). Oxagon is remote, still under construction, and located far from Saudi Arabia’s major population and enterprise centers in Riyadh and Jeddah.
The case for Oxagon as an AI factory site rests on several factors that the traditional data center site selection model does not fully capture.
First, power cost and sustainability. Oxagon’s dedicated renewable generation eliminates the power cost volatility and carbon risk that increasingly constrains data centers built on conventional grids. As regulatory pressure on AI compute’s carbon footprint intensifies globally — the International Energy Agency’s 2024 data center report highlighted AI as a primary driver of data center emissions growth — having genuinely carbon-neutral AI compute is shifting from a nice-to-have to a competitive requirement. Hyperscalers and major AI companies face pressure from investors, regulators, and their own sustainability commitments to decarbonize their compute. DataVolt’s Oxagon facility offers that capability at gigawatt scale.
Second, land availability and cost. NEOM has essentially unlimited land at effectively zero acquisition cost compared to any data center location near major cities globally. Gigawatt-scale facilities require hundreds of acres; in Northern Virginia or Singapore, land for this scale simply does not exist at any reasonable price. Oxagon’s remote industrial zone removes land as a constraint entirely.
Third, the Saudi AI ecosystem alignment. DataVolt’s facility is directly embedded in Saudi Arabia’s flagship AI infrastructure buildout. Proximity to Humain’s AI ambitions, NEOM’s digital infrastructure requirements, and the Kingdom’s sovereign AI compute agenda means DataVolt has a natural primary customer pool. The facility does not need to compete for generic hyperscale customers globally — it can serve the Saudi AI demand that is being created by the very policy environment that made the project viable.
Fourth, thermal cooling advantages. Oxagon’s Red Sea coastal location provides access to seawater cooling, which substantially reduces the energy overhead for data center cooling — one of the largest single power drains in a conventional facility. Seawater at the Red Sea surface runs 25–30°C, cooler than ambient air in the region, enabling evaporative and direct cooling approaches that reduce overall facility Power Usage Effectiveness (PUE) compared to air-cooled facilities in hot climates.
The AI Factory Model: Not a Generic Data Center
DataVolt is explicitly positioning the Oxagon facility as an “AI factory” rather than a general-purpose data center. This distinction matters operationally and strategically.
A generic hyperscale data center is designed to run diverse workloads — web serving, databases, storage, some compute. Its network architecture, power density per rack, and cooling infrastructure reflect a range of use cases. An AI factory is purpose-built for AI training and inference workloads, which have very different technical requirements: extreme GPU density, high-bandwidth interconnect (InfiniBand or Ethernet at 400Gb/s+ per node), power densities of 100–200 kW per rack (versus 10–20 kW for generic data centers), and thermal management designed for sustained maximum-load operation.
Designing from the ground up for AI workloads allows DataVolt to optimize its facility architecture in ways that retrofitting a conventional data center cannot. The building layout, power distribution, cooling architecture, network topology, and facility management systems can all be co-optimized for GPU cluster operation. This matters economically: a well-designed AI factory achieves better GPU utilization, lower operational cost per GPU-hour, and better thermal management than an AI overlay on a conventional facility.
The AI factory model also has implications for the GPU supply chain. DataVolt will require large quantities of NVIDIA Blackwell (GB300 or successor) GPUs for the facility’s planned 1.5 GW AI compute capacity. Saudi Arabia’s overall GPU allocation from NVIDIA — 18,000 GB300 units in the initial tranche, scaling to 600,000 over three years, primarily through Humain — creates a question about how DataVolt’s GPU procurement interacts with the Kingdom’s centralized GPU acquisition strategy. DataVolt is not Humain and is not directly within the PIF-Humain procurement pipeline, which means it must either secure its own NVIDIA allocation (requiring BIS export license approvals under the AI Diffusion framework) or operate in partnership with Humain for GPU supply.
NEOM’s Execution Risk: The Unavoidable Overhang
Any analysis of DataVolt’s Oxagon project that does not centrally address NEOM’s execution track record is incomplete. NEOM is simultaneously Saudi Arabia’s most ambitious infrastructure project and one of the most significantly scope-reduced megaprojects in recent history.
The original vision for The Line — NEOM’s 170km linear city — projected a city of 1.5 million residents by 2030. Credible reporting in 2024 indicated that NEOM’s management had internally revised that target to approximately 300,000 residents, a reduction of 80%, with infrastructure construction proceeding on a far smaller initial footprint than originally planned. Significant numbers of workers have been laid off as NEOM’s construction pace slowed. The broader NEOM project has been through multiple strategy reviews as the Saudi government has grappled with the gap between Vision 2030’s infrastructure ambitions and the practical execution capacity of the Saudi construction industry and government procurement systems.
DataVolt, as a tenant in the Oxagon development, inherits NEOM’s execution risk whether it wants to or not. Several specific risks are material:
Infrastructure availability timeline: If Oxagon’s port facilities, road connectivity, and utility grid connections are delayed, DataVolt cannot operate the facility even if its own construction proceeds on schedule. The shared infrastructure dependencies of a megaproject mean that a single-site AI factory is exposed to the weakest-link dynamics of its host development.
Customer availability: If NEOM’s own digital infrastructure buildout (managed by Tonomus) proceeds more slowly than planned, a significant portion of DataVolt’s natural customer base — NEOM’s own AI workloads — may not materialize on the timeline that the facility’s business case requires. A 1.5 GW AI factory needs customers; if NEOM is running at 20% of planned scale in 2028, the demand assumption that justified the facility’s size requires revision.
Capital commitment durability: The $5 billion investment commitment requires sustained capital availability over a multi-year construction period. NEOM has in the past experienced funding pauses and priority reallocation as PIF has managed its overall investment portfolio against Saudi Arabia’s fiscal position. DataVolt is a private entity, not directly PIF-funded, but the broader investment environment it operates in is shaped by PIF’s strategic priorities.
Regulatory and permitting: Building at Oxagon requires navigating NEOM’s own special economic zone regulatory framework, which is still being developed and may have gaps or ambiguities that create construction or operational delays.
Renewable Energy Infrastructure: ACWA Power and the Power Supply Chain
ACWA Power’s role as DataVolt’s renewable energy provider deserves specific attention. ACWA Power is one of the world’s largest independent power producers focused on renewable energy, with a particularly strong position in Saudi Arabia and across MENA. The company is partially owned by PIF and has been a primary delivery vehicle for Saudi Arabia’s renewable energy targets under Vision 2030.
ACWA Power’s relationship with Oxagon extends beyond DataVolt — ACWA is developing renewable generation for the broader NEOM ecosystem, including the Helios green hydrogen project that aims to produce green hydrogen for export and domestic use. DataVolt’s power supply arrangement with ACWA is thus embedded in a larger renewable energy ecosystem rather than being a standalone bilateral deal. This is both a strength (ACWA has scale and development experience) and a dependency (if ACWA’s Oxagon-area renewable projects face delays, DataVolt’s power supply is affected).
The green hydrogen connection is worth noting for the facility’s sustainability case. Green hydrogen produced from excess renewable generation can be stored and used to generate power when solar and wind generation is insufficient, providing the “firm” renewable capacity that a 24/7 data center operation requires. If NEOM’s Helios project delivers on its production targets, it could provide DataVolt with genuinely dispatchable renewable power rather than requiring large-scale battery storage — a potentially significant cost and reliability advantage.
Strategic Significance and Market Positioning
If DataVolt’s Oxagon facility delivers on its specifications, it would be the world’s largest single-site net-zero AI compute facility by a substantial margin. This is not merely a record — it is a proof point for a new model of AI infrastructure: gigawatt-scale, purpose-built for AI workloads, powered entirely by dedicated renewable generation, and located in a low-cost land market with strong geopolitical alignment to the world’s largest AI spending program.
The implications for the broader AI infrastructure industry would be significant. As AI training workloads continue to scale (next-generation frontier model training runs are projected to require 100s of MW of sustained GPU power for months), the ability to concentrate large GPU clusters in a single location with firm renewable power and competitive economics becomes a genuine differentiator. DataVolt’s Oxagon facility, if complete, would be a natural venue for the largest AI training runs in the world.
For Saudi Arabia’s Sovereign Compute Score, DataVolt’s facility contributes primarily to the Capacity dimension — it adds raw AI compute capacity to the Kingdom’s infrastructure base. Its renewable power commitment adds a sustainability dimension that is increasingly relevant for geopolitical positioning. And its location within NEOM adds to the execution risk that is one of the persistent discounts applied to Saudi AI compute ambitions by outside observers.
DataVolt’s Oxagon project is a genuinely important infrastructure bet — the right idea, in the right location for the right strategic reasons, with a credible renewable energy partner. It is also a bet with non-trivial execution risk that ultimately depends on NEOM delivering on at least a portion of its industrial hub ambitions. Investors and analysts tracking Saudi AI compute should watch the Oxagon construction timeline, ACWA Power’s renewable generation delivery milestones, and NEOM’s revised population and activity targets as the leading indicators of DataVolt’s probability of hitting its 2028 capacity targets.