The Net-Zero AI Factory
DataVolt’s NEOM AI factory, sited in NEOM’s Oxagon industrial zone on the Red Sea coast, is a $5 billion, 1.5 gigawatt facility scheduled for operational status in 2028. It is the single largest renewable-powered AI facility committed anywhere in the world, and the single largest project of any kind in Saudi Arabia’s AI infrastructure pipeline. DataVolt, a Saudi infrastructure company aligned with the broader Vision 2030 portfolio, is building the facility on infrastructure provided by NEOM’s energy and connectivity utilities.
The 1.5 GW scale is unprecedented for renewable-powered compute. Most large AI facilities globally are gas- or grid-powered with renewable offsets; DataVolt’s facility is being constructed with direct renewable supply — solar, wind, and green-hydrogen-derived power sourced from NEOM’s broader energy portfolio. The net-zero claim is structural rather than offset-based: the facility is designed to operate on renewable supply from day one, through dedicated generation and power purchase agreements rather than certificates bought against a fossil grid. That distinction — physical net-zero versus accounting net-zero — is what separates this project from nearly every green data-center claim in the industry, and it is the claim on which the project’s global significance rests.
What 1.5 GW Actually Means
Capacity figures deserve interpretation. At modern AI training densities — 80-100 kW per rack for liquid-cooled Blackwell-generation systems — 1.5 GW of IT load supports roughly 15,000-18,750 high-density AI racks, implying capacity on the order of a million or more Blackwell-class GPU chips at full buildout. For historical scale: the training cluster reported for GPT-4 ran approximately 25,000 A100s; the DataVolt facility could host dozens of clusters of that class simultaneously.
Within the Saudi pipeline, the facility occupies its own order of magnitude. It is 7.5 times the size of Humain’s Riyadh Phase 1 campus (200 MW), more than three times the 480 MW Hexagon government data center, three times the planned 500 MW xAI campus, and fifteen times the current Center3 colocation footprint (100 MW). The comparisons matter because they establish what kind of project this is: not an incremental addition to Saudi capacity but a step-change that would, if delivered, make Saudi Arabia one of the world’s most significant AI compute locations by a considerable margin — and make every other facility in the Kingdom a rounding error against the Oxagon number.
The $5 billion price sits in the same band as the Kingdom’s largest partner commitments — above AWS’s $5.3 billion only nominally, half of Google Cloud’s $10 billion Dammam program, more than three times the Groq-Aramco Digital inference buildout — but it buys a categorically different asset. The hyperscaler figures fund multi-year service platforms; the DataVolt figure funds a single physical plant. Dollars-per-megawatt, it is among the most capital-efficient line items in the Saudi pipeline, precisely because the energy architecture substitutes NEOM’s generation portfolio for the gas turbines and grid-reinforcement costs that burden equivalent projects elsewhere. Capital efficiency at announcement, of course, is a projection; the EPC execution decides whether it survives contact with the site.
Why NEOM
NEOM, the $500B megacity project on Saudi Arabia’s Red Sea coast, provides three operational advantages for hyperscale AI infrastructure. First, abundant renewable energy: NEOM sits in the Tabuk region, one of the world’s highest solar-irradiance zones — reliably above 2,000 kWh per square meter per year — with consistent wind resources off the Gulf of Aqaba corridor; the combination supports renewable generation at low marginal cost across more hours of the day than either source alone. Second, desert cooling and seawater access: the coastal location provides natural cooling options that reduce the energy overhead of mechanical cooling, a material line item at gigawatt scale. Third, connectivity: NEOM is anchored to the Trans-Asia subsea cable system, providing low-latency links to Asia, Europe, and Africa — the traffic triangle the Saudi hub strategy is built to serve.
For NEOM itself, the DataVolt facility is the most operationally significant tenant in Oxagon, the megaproject’s floating-and-coastal industrial complex near the Saudi-Egyptian border. Oxagon was conceived as NEOM’s manufacturing and logistics hub; AI compute has become its anchor industrial use, and the credibility exchange runs both ways. NEOM needs a flagship tenant that validates its industrial thesis after years of scope-reduction headlines; DataVolt needs NEOM’s energy portfolio and utility infrastructure to make the net-zero architecture physically possible. Neither side of that dependency is trivial.
The Contrarian Site Selection
On the traditional data-center site-selection model, Oxagon is an implausible choice. The world’s major compute concentrations cluster near population centers and enterprise demand — Northern Virginia, Silicon Valley, Dublin, Singapore, Tokyo — or at minimum near mature grid and fiber infrastructure, as in Iceland, Sweden, and Oregon. Oxagon is remote, still under construction, and far from Saudi Arabia’s own population and enterprise centers in Riyadh and Jeddah. A conventional colocation developer would not shortlist it.
The AI-factory workload profile is what rewrites the calculus. Training is latency-tolerant: a frontier training run does not care whether its users are twenty milliseconds or two hundred milliseconds away, because it has no users — it has a job queue measured in weeks. What training cares about is the cost and continuity of power, which is precisely what Oxagon’s dedicated renewable generation offers and what demand-adjacent sites increasingly cannot. The site-selection inversion — power-adjacency over demand-adjacency — is the same logic that put xAI’s Colossus in Memphis rather than Silicon Valley, applied at three times the scale and with a decarbonized supply chain. For latency-sensitive serving, the Saudi portfolio has Riyadh, Dammam, and the subsea-connected coastal routes; Oxagon does not need to win that mission to justify itself. It needs to be the cheapest clean gigawatt-scale training location on earth, which is a mission with exactly one entry requirement Oxagon meets and Virginia does not: available power.
The Power Architecture
The engineering problem that defines the project is firming: solar stops at night, wind is variable, and AI training jobs require uninterrupted power measured in days and weeks. DataVolt’s answer is architectural rather than contractual. ACWA Power — Saudi Arabia’s dominant renewable developer, itself partially PIF-owned — is the primary power supply partner, providing solar and wind generation dedicated to the Oxagon facility under direct power purchase agreements rather than grid-mix procurement. NEOM’s broader green-energy infrastructure, including the Helios green hydrogen project, supplies the storage and balancing layer: hydrogen produced from surplus daytime renewable generation becomes dispatchable power for the hours the panels and turbines cannot cover.
Green hydrogen as the firm-power solution is the project’s most distinctive engineering bet and its least proven one. If it works at gigawatt scale, Oxagon demonstrates a template for 24/7 renewable compute that no operator in Virginia or Ireland can currently match. The honest caveat, noted in the project’s public record, is that the detailed energy-storage plans have not been fully disclosed — and until they are, the 24/7 renewable claim rests on NEOM’s broader energy program delivering on its own aggressive schedule.
DataVolt as Delivery Risk Owner
DataVolt’s role is EPC — engineering, procurement, and construction — which means DataVolt bears the project delivery risk: designing the facility, procuring equipment, managing construction, and delivering commissioned infrastructure. That is a different risk allocation from conventional data-center development, where the owner-operator carries development risk with construction management support. DataVolt is the specialist responsible for turning NEOM’s specifications into an operating plant.
The delivery challenge is not a scaled-up version of a normal build. A 1.5 GW facility requires utility-scale power infrastructure — multiple dedicated substations, potentially new transmission — civil engineering across millions of square feet, and construction logistics at a remote coastal site in northwestern Saudi Arabia that lacks the mature contractor ecosystems of Riyadh or Dammam. The rational execution path is phased delivery: early tranches of 200-300 MW coming online in 2026-2027, with subsequent phases completing through 2028. Phasing also de-risks the capital: early phases can demonstrate operational and commercial credibility before the full $5 billion is committed to steel and silicon.
The Global Power Constraint
The NEOM-DataVolt facility addresses one of the most-discussed constraints on global AI scaling: power. Mainstream AI compute growth is projected to consume electricity at rates that strain grid capacity in established hyperscaler markets — Virginia, Ireland, Singapore — where interconnection queues, community opposition, and transmission limits now gate new capacity more tightly than capital does. Saudi Arabia’s energy abundance, both hydrocarbon and renewable, positions it as one of the few markets where 1.5 GW of dedicated compute can be added without grid stress.
This is the deep logic of the entire Saudi compute thesis — energy-to-compute conversion — expressed in its purest form. The Kingdom’s structural advantage is not engineering talent or software ecosystems; it is the ability to bring cheap electrons and capital to the same site at scale, quickly. Most Saudi facilities exercise that advantage through hydrocarbon-backed grid power. DataVolt exercises it through the renewable portfolio, which converts the same land-and-sun endowment into a product the hydrocarbon story cannot offer: compute that satisfies a carbon audit.
The Green Compute Market
That product has a customer base that is growing structurally. Global enterprises face tightening carbon-accounting obligations, and the International Energy Agency’s 2024 data-center report flagged AI as a primary driver of data-center emissions growth — regulatory and investor pressure that turns genuinely carbon-neutral compute from a nice-to-have into a competitive requirement. Hyperscalers and major AI companies carry public sustainability commitments that their AI expansion is actively straining; every gigawatt of gas-fired AI capacity added elsewhere deepens the deficit that offset markets are decreasingly permitted to paper over.
DataVolt’s Oxagon facility is positioned as the supply-side answer: hyperscale AI capacity whose net-zero claim survives scrutiny because it is physical, not financial. If the claim holds, the facility can command a premium tenant mix — the workloads of companies that need both scale and a defensible emissions story — and Saudi Arabia gains a differentiated export product in a compute market that otherwise competes on price and latency alone. The UAE has nothing equivalent announced at this scale; neither, for that matter, does the United States.
The Saudi Cadence
The 2028 timeline aligns with a broader Saudi cadence. Humain’s 1.9 GW capacity target by 2030 includes the NEOM-DataVolt facility, and the 6.6 GW decade-end target across Saudi compute infrastructure is contingent on renewable-powered facilities like DataVolt scaling on schedule. The sequencing is deliberate: the Riyadh and Dammam campuses plus Hexagon carry the 2026-2027 load while the Oxagon phases build, and DataVolt then carries the capacity story from 2028 onward. In portfolio terms, the Kingdom’s near-term facilities are the execution proof and DataVolt is the scale proof; the program needs both, in that order.
The facility also diversifies the portfolio’s geography. Riyadh concentrates government and commercial compute; Dammam anchors the Google Cloud hub against Eastern Province energy and Gulf subsea routes; Oxagon opens the Red Sea corridor with Trans-Asia connectivity toward European and African traffic. Three coasts, three cable systems, three energy profiles — a national compute footprint designed like a network rather than a campus.
The Risks
Two risks bound the project, and both deserve sober weighting. First, construction execution: 1.5 GW facilities are operationally complex, and 2028 represents an aggressive timeline given typical hyperscale construction cycles of 3-5 years — compressed further here by remote-site logistics and a first-of-kind power architecture. Second, renewable supply integration: matching 1.5 GW of demand with renewable supply on a 24/7 basis requires either substantial battery storage or grid backup, and the undisclosed storage detail is the load-bearing unknown in the net-zero claim.
There is a third, contextual risk the project cannot escape: NEOM’s well-documented track record of scope reductions and timeline delays. Every NEOM-sited project pays a credibility discount inherited from the megaproject’s history, fairly or not. DataVolt’s mitigations are real — ACWA Power’s delivery record in Saudi renewables, the phased build path, and the fact that an AI factory is a conventional asset class compared to a linear city — but the discount only clears when megawatts are serving workloads.
The Benchmark Test
If the NEOM-DataVolt facility hits its 2028 target with full renewable supply, it sets a precedent globally — the model for net-zero hyperscale compute, proven at a scale an order of magnitude beyond anything comparable, in a location most of the industry dismissed. If it slips or compromises on the renewable framing, it becomes another conventionally-powered facility with an aspirational green narrative, and the burden of the Saudi green-compute story shifts to smaller projects. The outcome is binary in narrative terms even though it will be incremental in engineering terms, which is why the 2026-2027 phase deliveries — not the 2028 headline date — are the milestones worth watching. They will reveal, years before the ribbon-cutting, whether the world’s largest net-zero AI factory is a construction schedule or a press release.