Analysis Frames AI Data-Center Competition Around Power, Infrastructure and Supply Chains
Редакция ЦОД:Главное
A comparative analysis of the United States, the European Union and China argues that competition for AI infrastructure is increasingly constrained by power availability, grid and physical infrastructure, chips, water and supply chains. It cites planned 2026 capital expenditure of about US$725 billion by Amazon, Alphabet, Microsoft and Meta combined, alongside the EU’s goal of tripling data-center capacity within five to seven years.
- Amazon, Alphabet, Microsoft and Meta are cited as planning about US$725 billion of combined CAPEX in 2026, although the data-center-specific allocation is not disclosed.
- The analysis places grid access, power availability, cooling, equipment and critical materials alongside compute demand as potential constraints on AI data-center deployment.
- The EU’s objective is to triple data-center capacity over five to seven years; it is a target, not installed or commissioned capacity.
- The source describes China’s approach as directing computing workloads towards regions with available capacity and stronger renewable-energy resources.
- The article cites potential gallium supply exposure, with demand projected to exceed current supply by 10% by 2030 and China holding 99% of refined-gallium supply.
The analysis characterises AI infrastructure as a strategic competition in which electricity, physical infrastructure, semiconductor supply, water and critical-material supply chains are becoming binding constraints.
For the United States, it highlights a private-capital-led model centred on hyperscalers. According to the source, Amazon, Alphabet, Microsoft and Meta plan combined capital expenditure of about US$725 billion in 2026, compared with US$410 billion in 2025—an indicated increase of 77%. The source says a significant, but unspecified, share of the 2026 total is directed towards AI data centers, semiconductors and energy resources; it does not disclose the portion allocated specifically to data centers.
The article also relays an IEA view that nearly half of US data-center capacity is concentrated in five regional clusters. It further cites an IEA forecast that data centers could account for nearly half of growth in US electricity demand by 2030. If realised, that concentration could heighten the importance of local generation, transmission and distribution capacity, grid interconnection and equipment availability in determining the pace of AI-campus delivery.
In Europe, the EU is described as pursuing a regulatory and industrial-policy-led approach, with a target to triple data-center capacity over five to seven years. This is a policy objective rather than commissioned capacity, and the analysis provides neither a baseline capacity figure nor details of the projects intended to meet the target.
China is presented as pursuing centrally coordinated compute-infrastructure planning. Its national computing network is intended to direct workloads towards regions with available capacity and more substantial renewable-energy resources. The source does not specify the network’s capacity, locations or delivery timetable.
Critical-material exposure is also identified as a potential constraint. The source states that gallium demand could exceed current supply by 10% by 2030, while China controls 99% of the global refined-gallium market. More broadly, the analysis suggests that the ability to secure power connections, grid equipment, cooling infrastructure and material supply may increasingly constrain AI-infrastructure deployment across markets.
The analysis points to a shift in the data-center market: deployment of AI capacity could increasingly be determined by grid interconnection, power supply, cooling, land and equipment availability—not solely by demand for compute or access to capital. Large hyperscaler investment plans could support demand across the infrastructure supply chain, while European and Chinese build-outs may be more shaped by policy, regulation and the geographic distribution of energy resources.
- TechNews 科技新報 / КитайскийПервоисточник →
Материал подготовлен редакцией ЦОД: Главное на основе указанных первоисточников.