Last verified: 16 October 2026. The network-expansion statement is attributed to SPPC’s chief executive. It is not an independently audited measure of completed grid works.
Saudi Arabia’s push into artificial intelligence is becoming a power-delivery question. At the WPC Energy Congress in Riyadh, Saudi Power Procurement Company chief executive Mazin Al-Bahkali said about half the grid was being scaled up to meet growing demand, including industry, AI and data. Arab News reported the remarks on 13 October. [S1]
The statement is significant because it places digital development inside the national electricity investment programme. It does not disclose which assets make up the quoted half, the amount already completed, a budget or a commissioning calendar. Those details are necessary to connect a national ambition with a particular data centre’s operating date.
The next meaningful milestone for an AI facility is dependable electricity at its site, under terms that support commercial operation. A generation award, a transmission upgrade, a connection agreement and an energised facility contribute to that outcome at different stages. Counting them as equivalent would overstate readiness.
What the 50% statement can support
The reported remarks identify rising demand and an active expansion programme. Al-Bahkali described work to forecast consumption and attract Saudi and foreign investors into generation. His comments included AI and data within a broader set of electricity users. They did not assign half of national electricity demand or half of grid investment exclusively to AI. [S1]
| Statement or measure | Relevant denominator | Evidence still needed |
|---|---|---|
| About half the grid being scaled up | Unspecified in the reported remarks | Asset list, definition and project stages |
| A data centre announced at a stated MW size | IT load, facility load or connection capacity | Consistent capacity definition |
| Generation capacity awarded | Nameplate MW or GW | Completion, fuel, availability and grid access |
| Connection agreement | Contracted electrical service | Energisation date and delivery conditions |
| Computing service available | Usable IT capacity | Installed systems, uptime, customers and utilisation |
The missing denominator is not a reason to dismiss the announcement. It defines the next reporting task. An expansion measured by route length can differ substantially from one measured by substations, transfer capability or expenditure. Without a common base, the percentage cannot become a national completion score.
The same caution applies to renewable targets. A share of generation, a share of installed capacity and a portion of the network being upgraded are different measures. They should never be substituted for one another merely because they use a similar percentage.
A data centre needs a chain of power assets
Electricity must be generated, transported through the network and delivered at the required voltage and reliability. Within the site, transformers, switchgear, uninterruptible supplies, cooling and backup arrangements keep computing equipment operating. A shortage or delay at any necessary point can constrain the facility.
The IEA’s Energy and AI report describes both the computing equipment and the auxiliary systems that consume electricity. It also highlights a timing mismatch: digital facilities can move faster than the broader energy infrastructure needed to serve them. That is a planning issue for countries expanding both systems together. [S2]
National generation adequacy does not prove local connection readiness. A country can have sufficient total production capacity while a particular site requires additional network works. Conversely, a site connection can exist before enough computing hardware or customer demand is available to use it fully.
A delivery programme needs the dependencies in sequence. The useful date is when the site can sustain its contracted operating load, rather than the date of the earliest upstream announcement. Manufacturing lead times, civil works, testing and integration should be represented in that schedule.
The megawatt label needs to be defined first
A 100 MW announcement can describe several different things. It may refer to IT equipment, total facility demand, a utility connection, an initial phase or a longer-term campus plan. Those definitions produce different electricity requirements and different amounts of usable compute.
Power usage effectiveness, or PUE, expresses total facility energy relative to IT equipment energy over the same period. It captures the additional electricity needed for supporting systems. It is an efficiency ratio, not a measure of the usefulness of the calculations performed by the servers.
For illustration, assume 100 MW of IT load, a PUE of 1.30 and an average utilisation factor of 90%. If the ratio and utilisation remain constant, average facility demand would be 117 MW and annual electricity use approximately 1.025 terawatt hours. The calculation is 100 × 1.30 × 0.90 × 8,760 hours. These are selected assumptions, not disclosed specifications for a Saudi project.
If the original 100 MW instead meant total facility power, multiplying it by PUE again would double count the support load. That is why a national pipeline assembled from incompatible capacity labels can materially overstate both computing availability and required electricity.
Peak capacity and annual energy answer different questions
Megawatts describe a rate of power use. Megawatt hours and terawatt hours describe energy over time. A facility may need a large firm connection even while its average utilisation is lower during an initial ramp-up. Network planners must consider the maximum credible load, its location and its timing.
Annual energy procurement also has a time profile. An annual amount of renewable electricity can match annual consumption without providing the same output in every hour. Reliability depends on the complete supply arrangement, including flexibility and any firming or backup resources.
The right commercial questions include the connection date, maximum demand, expected load profile, redundancy and the conditions under which service can be curtailed. An attractive average tariff cannot compensate for an operating arrangement that does not meet the customer’s reliability requirements.
The infrastructure cost also needs allocation. Developers, utilities and other users can bear different shares of new connection and network expenditure. A subsidised or publicly funded component can improve the project’s private economics while leaving a wider public cost. That allocation should be disclosed before a tariff is presented as an inherent national advantage.
Procurement is advancing, with commissioning still ahead
On 16 October, MEED’s public report summary identified a tender for HUMAIN’s 100 MW Al-Saad AI data centre in Riyadh, with an option to expand to 200 MW. Its estimated contract value was $1 billion. The visible summary establishes a reported tender and estimate, not an award, financial close or operating facility. [S3]
That is a useful new point in the delivery chain. It indicates work moving into a specified procurement stage. It should not be combined automatically with earlier capacity announcements as though each referred to a separate additional asset. Project names, sites, phases and counterparties need reconciliation to prevent double counting.
The site’s September power-agreement analysis provides the earlier context. Electricity agreements and wider cooperation frameworks established coordination. The October reporting increases the importance of matching those agreements with construction, equipment and energisation milestones.
A sound ledger would distinguish announced, tendered, contracted, financed, under construction, energised, commissioned and commercially available capacity. It would also identify partial phases. A campus can begin operations with a small portion of its eventual capacity without validating the complete masterplan total.
Power cost matters through the useful workload
Electricity is one component of computing economics. Hardware cost, financing, cooling, networking, software, staffing and utilisation also determine the cost of a useful AI task. Cheap energy can support competitiveness, but cannot guarantee it.
Utilisation is particularly important for expensive equipment. A facility with low power prices and too few paying workloads can have worse economics than a more expensive site with high productive use. An available GPU is not the same as a contracted customer, and a contracted customer is not the same as collected revenue.
The efficiency frontier also has several dimensions. Better cooling reduces supporting energy. Better hardware or software can reduce energy per useful task. More efficient models can make services less expensive, while expanding adoption can increase total demand. The IEA therefore uses scenarios rather than a single deterministic forecast for future AI electricity consumption. [S2]
For Saudi Arabia, that uncertainty argues for phased infrastructure with credible demand commitments and room to adapt. Building every conceivable load immediately can create idle assets. Waiting for every customer to arrive can create a connection bottleneck. The planning task is to coordinate commitments and options across both systems.
Industrial users share the system
The CEO’s remarks describe a demand base extending beyond AI. Manufacturing, water-related infrastructure, buildings and other activities also require dependable electricity. A grid programme should be evaluated against the complete demand forecast rather than one technologically prominent customer group. [S1]
Different users place different demands on the network and create different economic benefits. A concentrated digital campus can need a large, reliable connection; a dispersed industrial programme can require upgrades across several locations. The best allocation depends on costs, flexibility and expected value, rather than the visibility of the announcement.
The assessment should also distinguish private returns from domestic economic contribution. A data centre can earn commercial revenue with imported hardware and limited direct employment while supporting a wider digital ecosystem. Whether that ecosystem develops depends on local customers, technical skills and service activity. Electricity consumption itself is not the developmental objective.
The HUMAIN–Napster application partnership provides a complementary demand-side question: can the technology support useful services that organisations will pay to operate? Infrastructure and applications need to be assessed together without treating either announcement as proof of the other’s commercial success.
Reliability needs its own evidence
Critical digital facilities require an operating record. The evidence should include service availability, interruption frequency, restoration time and the performance of backup arrangements under relevant conditions. A promised level of resilience is different from a tested operating history.
October’s infrastructure attacks make continuity more salient, but they do not establish a specific power interruption at a named AI site. A responsible assessment should avoid inferring damage or downtime from geographic proximity. It should examine operator disclosures and actual service records.
Likewise, cybersecurity and physical resilience are distinct from annual energy sufficiency. A system can have enough generation while remaining exposed through particular assets or controls. The appropriate response is evidence-based redundancy and operating discipline, not an unsupported declaration that an entire national grid is insecure.
The assessment
The reported expansion of about half the grid indicates that Saudi Arabia recognises the scale of the electricity task accompanying industrial and digital growth. The next useful disclosures are asset-specific: what is being built, at what cost, for which loads and by which commissioning dates.
AI competitiveness will depend on the conversion of that programme into reliable power and commercially used computing capacity. The strongest evidence will be an integrated delivery schedule, verified operating performance and customers paying for useful work. A large network percentage establishes ambition and activity; the economic return requires the chain to function end to end.
Related Vision 2030 context
- September’s AI data-centre power agreements
- HUMAIN and MIS’s planned 250 MW expansion
- HUMAIN’s Riyadh and Dammam operating-status assessment
Sources
- [S1] Ghadi Joudah, Arab News, reporting SPPC chief executive Mazin Al-Bahkali’s WPC Energy Congress remarks, 13 October 2026. The 50% statement remains attributed, with no denominator or completion schedule supplied in the report. Arab News.
- [S2] International Energy Agency, Energy and AI, 2025, “Energy demand from AI.” Used for data-centre components, infrastructure timing and demand uncertainty. IEA.
- [S3] Yasir Iqbal, MEED, “Humain tenders 100MW Riyadh AI data centre,” 16 October 2026. Only the publicly visible title and summary are used; the quoted value is an estimate. MEED public summary.