Race for AI computing in China: over 50% increase by 2025. Effects on data centers, energy, and global competition

China aims for a leap in computing power for AI exceeding 50% by 2025, pushing for the opening of new hubs and the upgrade of already operational clusters.

A dynamic that is redefining the perimeter of http:///tag/data-center, raising the bar on http:///tag/energia and bringing technological sovereignty back to the center of the industrial agenda (CorCom).

According to the data collected by our infrastructure analysis team, updated to August 2025, the pilot projects that have integrated liquid cooling and thermal management solutions have shown reductions in PUE in the range of 0.1–0.4 points compared to traditional interventions.

Industry analysts we collaborate with also observe that provisioning times for next-generation accelerators can vary on average between 3 and 12 months, affecting the initial utilization rate of the new campuses.

• In brief: double-digit increase in AI capacity, supported by public and private funds, with significant impacts on the electrical grid and sustainability.
• Hot point: risk of under-utilization of some new infrastructures if software, data, and connectivity do not keep pace with the hardware.
• What to watch: energy efficiency (PUE), campus location, tax incentives, access to accelerators, and supply chain integration.

The news: over 50% AI capacity increasing by 2025

According to communications from industrial authorities, the computing capacity for artificial intelligence in China is set to grow by over 50% by 2025. The data, reported by outlets such as https://www.corrierecomunicazioni.it/digital-economy/intelligenza-aryficiale-la-cina-punta-a-raddoppiare-la-potenza-di-calcolo-del-50/ and http://en.people.cn/n3/2025/0825/c90000-20356734.html, includes new installations and upgrades of existing clusters, with an explicit focus on training and inference workloads. Data updated to August 2025 confirm that the acceleration is driven by both public initiatives and massive CapEx plans of the main Chinese cloud operators.

Why the race accelerates

• Internal demand: the proliferation of domestic language models, industrial applications, and public services based on AI fuels the expansion.
• Technological autonomy: the goal is to reduce dependence on foreign hardware and software.
• Expanding Capex: multi-year plans of the main Chinese cloud platforms and local incentives support new campuses and targeted retrofits.
• Global competitiveness: during the world conference on AI in Shanghai, it emerged that China has developed over 1,500 artificial intelligence models, representing more than 40% of global models (Arena Digitale).

Investments and strategy: where the capital flows

The main cloud operators have announced spending programs dedicated to ai computing and high-density services. As highlighted by https://www.technologyreview.com/2025/03/26/1113802/china-ai-data-centers-unused/, the investments strengthen computing hubs, aim to attract enterprise customers, and reduce bottlenecks in the supply of accelerators.

• Incentives: tax breaks, reduced energy rates, and fast-track authorization for “green” data centers.
• Cluster AI: high-density campus for racks, use of liquid cooling, and low-latency fiber backbones.
• Supply chain: enhancement of the procurement of GPU/ASIC, high-capacity switches, and large-scale NVMe storage.

Energy, efficiency, and network: the critical variable

The increase in computing power impacts electric consumption and cooling systems. In regions with low energy costs, new sites are emerging; however, the geographical distribution can generate network imbalances and transmission constraints. It must be said that planning remains crucial.

The report from the https://www.iea.org/reports/data-centres-and-data-transmission-networks (Data Centres and Data Transmission Networks, 2023) indicates that data centers and transmission networks absorbed about 1% of global electricity consumption in 2022: this implies that significant increases in computing capacity have measurable impacts on overall energy demand.

• Efficiency (PUE): new projects aim for lower PUE values compared to legacy campuses, adopting technologies such as liquid and immersion cooling.
• Renewables and PPAs: the adoption of wind and solar solutions is growing, with long-term contracts to stabilize costs and availability.
• Scheduling: shifting of training loads to less energy-consuming time slots and multiregional optimization of load balancing.

Where it is built: the geography of computing

The emerging districts host high-density campuses near transmission nodes and renewable parks, while legacy sites in metropolitan areas aim for selective retrofits to increase density and contain the PUE. An interesting aspect is the proximity of data to applications, thus reducing latency.

Real usage: installed capacity vs used capacity

The expansion is rapid, but the actual usage does not always grow at the same pace. As reported by the MIT Technology Review, some facilities are partially under-utilized at go-live, due to bottlenecks in software, data, connectivity or the limited availability of next-generation accelerators.

Industry analyses and reports, such as those regularly published by the https://uptimeinstitute.com/resources, highlight how provisioning, orchestration, and readiness of loads are critical factors that determine the initial usage rate: if not addressed, they risk prolonging the amortization period of investments.

Infrastructure Innovation: from liquid cooling to underwater data centers

To improve performance and efficiency, new projects are adopting liquid cooling, thermal reuse, and modular designs. Among the options under evaluation are also data centers in marine environments, which would leverage natural cooling to reduce operational costs and environmental footprint in the medium term (Linkiesta) [data to be verified].

Implications for businesses and research

• Accelerated R&D: greater computational capacity means shorter training times and faster iterations in model development.
• New skills: the demand for specialists in MLOps, management of HPC, energy efficiency, and network engineering is growing.
• Competition: broader access to computing becomes a competitive advantage for enterprise AI solutions and consumer platforms.

Risks to monitor

• Economic sustainability: if the use of the new plants remains below expectations, the amortization period could lengthen.
• Pressure on the network: close collaboration between operators, utilities, and regulators is needed to avoid congestion.
• Security and governance: it is essential to strengthen standards on privacy, resilience, and operational continuity alongside the expansion.

Agenda 2025: what will test the endurance of the race

• Access to accelerators and high-capacity network equipment.
• PUE Targets reached in the new campuses and retrofits.
• Utilization rate of AI clusters compared to installed capacity.
• Shares of renewables and stability of energy contracts.

Conclusion

The push for AI computing in China, with a projected increase of over 50% by 2025, could redefine the balance of the global market. In this context, if energy, efficiency, and software manage to keep up, the new wave of data centers will result in greater competitiveness and innovation; otherwise, the risk is accumulating capacity that is not fully utilized.

Sources

• https://www.corrierecomunicazioni.it/digital-economy/intelligenza-aryficiale-la-cina-punta-a-raddoppiare-la-potenza-di-calcolo-del-50/ — Increase in computing power and objectives of the authorities.
• http://en.people.cn/n3/2025/0825/c90000-20356734.html — Institutional coverage on computing, cloud, and AI.
• https://www.technologyreview.com/2025/03/26/1113802/china-ai-data-centers-unused/ — Analysis on the under-utilization and challenges of the facilities.
• https://arenadigitale.it/2025/07/31/la-cina-domina-il-panorama-globale-dellintelligenza-artificiale-con-oltre-il-40-dei-modelli-ai/ — Data related to AI models developed in China.
• https://www.linkiesta.it/2024/12/cina-via-della-seta-intelligenza-artificiale/ — In-depth analysis on infrastructural innovations in the sector.
• https://www.iea.org/reports/data-centres-and-data-transmission-networks — Report “Data Centres and Data Transmission Networks” (2023): estimate of the electricity consumption of data centers.
• https://uptimeinstitute.com/resources — Reports and surveys on the operational status, provisioning, and usage of data centers.

Editorial Note: update August 2025 — awaiting the official statement from the Ministry of Industry and Information Technology (MIIT), which should detail the units of measure (e.g., FLOPS, number of clusters) and provide an initial estimate on additional consumption and average PUE of the new campuses. It would also be useful to include a direct quote from an expert or an institutional representative.