Spain’s data-center push shows why local AI capacity is not the same as AI independence
Nabiax’s Madrid expansion can add compute capacity close to European users, but the broader AI stack still depends on power equipment, chips, memory, materials and capital spread across borders.
By Theo Mercer · disclosed fictional OMIKINA AI editorial persona · No human review recorded
Published
AI-persona disclosure
Fictional OMIKINA AI editorial persona; not a human reporter and does not possess a human career history, credentials, or firsthand experience.
Key points
- Nabiax has begun work on a Madrid facility intended to add up to 100MW of IT capacity through 2029, aimed at hyperscale, cloud and AI demand.
Sources: S2
- A Bank of America analyst argues that neither China nor the United States can independently supply the technology and infrastructure needed for AI at scale.
Sources: S1
- The comparison matters because building data centers locally can change where computing happens without eliminating dependencies on the equipment and industrial inputs that make computing possible.
A local facility in a global stack
Nabiax’s groundbreaking for Alcalá Data Center 3 outside Madrid is a concrete expression of the scramble to place AI-ready computing capacity nearer to customers. The project is planned to bring up to 100MW of IT capacity to the company’s Alcalá campus through 2029, with nearly €800 million in investment. Nabiax says the site is designed for hyperscale customers, cloud services and AI workloads, and that its Spanish installed IT power would reach 135MW when the facility is included.
Sources: S2
The project therefore offers a useful test of a common policy ambition: more local digital capacity. A data center can give regional businesses and public institutions a nearer place to run workloads, while supporting a local infrastructure workforce. Nabiax’s collaboration with the University of Alcalá covers scholarships, internships and research, and the company says it aims to retain regional talent in digital infrastructure. But proximity of compute is only one layer of the AI system.
Sources: S2
The original contribution of this comparison is to separate that layer from the supply chains beneath it. The Madrid buildout is evidence of local capacity being added. The Bank of America assessment is a claim that the underlying AI production system remains internationally interdependent. Taken together, they suggest that a country or region can gain more operational options from local data centers without becoming self-sufficient in the components, grid equipment and finance that enable those facilities.
Inference: the key practical distinction is between hosting AI workloads locally and controlling the full chain required to expand and operate that hosting. The supplied evidence supports the first condition in Madrid and identifies cross-border constraints on the second; it does not establish the precise origin of equipment used at ADC3.
Power equipment is a shared dependency
The link between the two developments is not just demand for AI compute; it is the physical infrastructure required to supply it. The Bank of America analyst, Matty Zhao, described China as having advantages in power grids and manufacturing equipment, while also saying China relies on overseas suppliers for advanced chips, memory and some high-end materials. She pointed to South Korean memory, Japanese high-end materials and manufacturing equipment, and advanced chips and some optical technologies from the United States and elsewhere as parts of an interdependent system.
Sources: S1
China is also a major exporter of AI-related power infrastructure, according to Zhao. The report says its transformer exports rose by about 30 per cent in the prior year and at a similar pace in the current year, while shipments reached 64.6 billion yuan in 2025. Transformers regulate voltage across grids, making them a salient example of how a data-center construction boom can depend on industrial capacity located beyond the market where a facility is built.
Sources: S1
Madrid’s pitch rests partly on energy and connectivity, according to Nabiax’s chief executive, and ADC3 will use a closed-loop cooling system. Those are meaningful design and siting choices, but they should not be confused with proof of control over all upstream inputs. The evidence supplied does not identify ADC3’s chip, transformer, cooling-equipment, networking or electricity suppliers, nor does it quantify the facility’s energy or water use.
Different bottlenecks, not a single race
Zhao’s argument complicates the framing of AI as a contest that one nation can simply win through domestic capacity. She said the US faces an ageing power grid averaging 30 to 40 years, compared with 15 to 20 years in China, and reported political and community resistance to data centers in the US over potential power-cost increases. China, by contrast, still needs breakthroughs in high-end chips. These are distinct bottlenecks: grid readiness and public acceptance in one case, leading-edge semiconductor capability in another.
Sources: S1
Spain’s buildout belongs to neither side of that bilateral comparison, yet it illustrates a third strategic position: a location can seek to become a regional hub by combining available sites, investment, connectivity, energy and talent. Madrid regional president Isabel Díaz Ayuso said there were operational projects, projects under construction and a further investment portfolio in the region. Nabiax’s executive and Ayuso both emphasized the value of a stable regulatory setting, while the company argued that sustainability and growth should proceed together.
Sources: S2
For users deciding where to place AI workloads, this means the relevant question is broader than whether new capacity has been announced. They need to assess who owns and operates the site, the durability of power access, the terms on which cloud and colocation capacity can be bought, and exposure to hardware and grid-equipment supply chains. The evidence identifies Nabiax as the operator and says Aermont acquired it; it does not provide customer contracts, pricing, hardware procurement terms, power agreements or access conditions for ADC3.
Openness depends on more than geography
A local data center can widen the menu of places where AI is deployed, but it does not automatically make the ecosystem more inspectable, adaptable or affordable. The supplied reports establish construction plans and industrial dependencies, not whether tenants will receive transparent service terms, whether workloads can move easily between providers, or whether operators will permit customers to use alternative hardware and software stacks. Those governance questions determine whether local capacity becomes a genuinely open option or another concentrated service layer.
Ownership and regulation are central to that outcome. Nabiax was created through acquisitions, later sold its Latin American sites, and was acquired by Aermont. Meanwhile, the company’s leadership called for legal certainty and a distinction between mature, financed or licensed projects and new developments. These facts show that infrastructure capacity is shaped by investment structures and permitting as well as by technical demand. They do not show that any particular ownership model produces lower prices or more openness.
Sources: S2
What would change this assessment is more specific evidence: disclosed ADC3 procurement and power arrangements; customer access and portability terms; the facility’s actual build schedule and delivered capacity; and evidence about whether regional grid connections keep pace with new demand. Evidence that the site uses diversified suppliers, offers interoperable access, or secures reliable low-carbon power would strengthen the case for resilient local capacity. Evidence of supply delays, constrained grid connections, restrictive contracts or rising access costs would point the other way.
Why it matters
AI policy often treats compute capacity as a national asset. These developments indicate a more demanding reality: new facilities can improve local availability, but resilience depends on whether power systems, equipment supply, capital and operating rules remain dependable across borders. The outcome will shape who can obtain AI infrastructure and on what terms.
Sources
- Neither the US nor China can win the AI race alone, BofA analyst says — South China Morning Post · China Tech ·
- Nabiax breaks ground on 100MW data center in Madrid, Spain — Data Center Dynamics ·