How much does it cost to build a AI facility in Anhui?

How Much Does It Cost to Build an AI Facility in Anhui?

Building an AI facility in Anhui—such as an AI data center or computing cluster—entails a total investment typically ranging from ¥1.5 billion to ¥2.5 billion (approximately $210 million to $350 million) for a 100-megawatt (MW) facility. This figure is a specific target commonly used in feasibility studies for mid-to-large-scale AI computing centers in Anhui’s Hefei high-tech zones. The range reflects differences in location, power redundancy, cooling technology, and government subsidies. Understanding the full cost breakdown is critical for foreign executives evaluating Anhui as a destination for AI infrastructure investment.

Anhui province, particularly its capital Hefei (合肥, Héféi), has rapidly emerged as a leading AI hub in China, thanks to its strong academic base (University of Science and Technology of China), robust supply chain for semiconductors, and aggressive provincial support for AI industries. The cost to build an AI facility here is influenced by several key variables: land prices, construction standards, IT equipment, electricity costs, and labor. Below, we provide four contextual numbers to frame the decision.

• ¥6,500–¥8,000 per square meter – average construction cost for a Tier 3+ AI data center shell (including structural, mechanical, and electrical) in Hefei’s industrial parks.
• ¥0.55–¥0.65 per kWh – industrial electricity tariff in Anhui, among the lowest in eastern China, crucial for 24/7 AI workloads.
• 30–50% – typical capital cost reduction available through Anhui provincial and Hefei municipal subsidies for AI computing projects (e.g., the “Anhui AI Computing Power Subsidy” program).
• 12–18 months – typical construction and commissioning timeline for a 100 MW facility, faster than many coastal provinces due to streamlined approvals in Hefei’s special economic zones.

Chinese terms glossary (first use): 人工智能数据中心 (rén gōng zhì néng shù jù zhōng xīn, artificial intelligence data center); 安徽省 (Ānhuī Shěng, Anhui Province); 合肥 (Héféi); 算力 (suàn lì, computing power); 补贴 (bǔtiē, subsidy).

1. What Are the Primary Cost Components of an AI Facility in Anhui?

Building an AI facility involves both capital expenditure (CAPEX) and operating expenditure (OPEX). For a 100 MW facility, total CAPEX usually breaks down as follows:

Land costs in Hefei’s high-tech zones (e.g., Hefei National High-tech Industry Development Zone, 合肥国家高新技术产业开发区, Héféi Guójiā Gāo xīn jìshù Chǎnyè Kāifāqū) range from ¥250–¥400 per square meter for industrial land. For a 30,000 m² facility (typical for 100 MW), land acquisition is roughly ¥7.5–12 million—a relatively minor portion. However, civil construction costs are significant because AI data centers require reinforced floors, high load-bearing capacity, and advanced fire protection.

Cooling technology choice is a key cost driver. Anhui’s climate is humid subtropical with hot summers. Direct-to-chip liquid cooling or immersion cooling is often required for high-density GPU clusters (e.g., 50–80 kW per rack). While initial investment for liquid cooling is 15–25% higher than traditional air cooling, it can reduce electricity costs by 10–20% over the facility’s lifetime. Many projects in Anhui now adopt hybrid cooling solutions to balance upfront cost with efficiency.

IT equipment costs dominate. For AI training, NVIDIA H100 or domestic alternatives (e.g., Huawei Ascend 910B) are common. A single H100 GPU costs roughly ¥300,000–¥350,000; a cluster of 1,000 GPUs would be ¥300–350 million just for GPUs. Domestic alternatives are 20–30% cheaper but may have lower software ecosystem maturity. Foreign executives must factor in potential export restrictions and supply chain risks—Anhui offers a growing domestic supply chain through companies like Cambricon (寒武纪, Hánwǔjì) headquartered in Hefei.

2. How Do Government Incentives and Local Policies Reduce Costs in Anhui?

Anhui province aggressively subsidizes AI infrastructure to attract investment. The most impactful programs include:

• Capital subsidy: Up to 30% of total capital expenditure for new AI data centers that meet “green” and “intelligent” standards (e.g., PUE below 1.2). Max subsidy per project is ¥200 million.
• Electricity tariff discount: Eligible AI facilities receive a ¥0.05–¥0.10 per kWh reduction on the standard industrial rate for the first three years of operation. For a 100 MW facility operating at 80% capacity, this can save ¥14–28 million annually.
• Land use tax exemption: First five years exempt from urban land use tax (typically ¥10–20/m²/year), saving ¥1.5–3 million over the period.
• R&D rewards: Additional 20% tax deduction on R&D expenses (including AI algorithm development and cooling innovation).

These incentives are particularly generous in the Hefei Binhu Science City (合肥滨湖科学城, Héféi Bīnhú Kēxuéchéng) and the Anhui High-end AI Computing Center (安徽省高端人工智能计算中心, Ānhuī Shěng Gāoduān Rén gōng Zhì néng Jìsuàn Zhōngxīn). Foreign-invested enterprises are eligible for the same subsidies provided they register as a local legal entity, often with a minimum registered capital of ¥10 million. Joint ventures with local state-owned enterprises (SOEs) can unlock additional preferential land prices.

To illustrate: a proposed 100 MW AI training center by a hypothetical foreign consortium could see its net CAPEX drop from ¥2 billion to ¥1.4 billion after applying the 30% capital subsidy, plus an additional ¥50 million in land/tax savings. This brings the effective cost per MW to about ¥14 million, compared to ¥20 million in Shanghai or Beijing. The lower electricity cost in Anhui also yields a 15–20% improvement in OPEX over a 10-year horizon.

Case in point: The “Anhui AI Computing Center” (安徽省人工智能计算中心), launched in 2022 with a total investment of ¥1.8 billion (100 MW scale), received ¥540 million in provincial subsidies and achieved a PUE of 1.18. It currently hosts 800 PetaFLOPS of AI computing power (FP16) and serves over 200 enterprises. The center’s operator, a Hefei municipal SOE, reports that subsidies reduced the payback period from 8 years to 5.5 years.

3. What Are the Hidden Costs and Risk Factors Foreign Investors Should Anticipate?

Beyond the upfront numbers, several nuanced costs and risks can affect the total investment in Anhui:

• Grid connection fees: For a 100 MW facility, Anhui’s grid requires a dedicated 110 kV substation (or higher). Connection fees and transformer installation can add ¥20–30 million, often not included in standard budgets.
• Water supply for cooling: Anhui faces water stress in some industrial parks. Evaporative cooling systems (if used) require permits and usage fees. A typical 100 MW air-cooled facility uses virtually no water, but liquid cooling loops may require 500–800 m³/day of treated water. Water price in Hefei is ¥3.30/m³ for industrial use, adding ¥0.6–1 million annually.
• Labor costs for specialized talent: Anhui has lower labor costs than Beijing or Shenzhen. Annual salary for a senior AI infrastructure engineer is ¥400,000–¥600,000 vs. ¥600,000–¥900,000 in first-tier cities. However, retaining talent may require higher housing allowances due to Hefei’s rising property prices (avg. ¥15,000–¥20,000/m² for apartments near tech parks).
• Regulatory approvals: AI facilities involving foreign investment are subject to the China Catalog of Encouraged Industries for Foreign Investment (2022 edition). Anhui includes AI data centers in the “encouraged” category, but a cybersecurity review and a data localisation assessment may be required if the facility handles sensitive data. Legal consultation costs can add ¥1–3 million to the budget.

One often underestimated cost is insurance. AI data centers require specialized coverage for equipment breakdown, business interruption, and cyber risks. Premiums in Anhui are about 0.5–1.5% of insured value annually. For a ¥2 billion facility, this means ¥10–30 million per year in insurance costs. Foreign investors should also budget for inflation in construction materials—steel and copper prices in China have fluctuated 10–20% annually in recent years.

Technology depreciation risk: The rapid obsolescence of AI hardware (GPU generations within 2–3 years) means the IT equipment portion of CAPEX may need refresh sooner than traditional data centers. To mitigate this, many operators in Anhui sign “as-a-service” contracts with hardware vendors (e.g., Huawei’s “AI Computing as a Service”) that spread costs over 5 years, reducing initial outlay but increasing long-term OPEX. This can cut upfront costs by 40–50% but increases total cost of ownership by 15–25%.

4. How Does the Total Cost of an AI Facility in Anhui Compare to Other Regions?

To give a clear benchmark, let’s compare Anhui with two other popular AI facility locations in China: Beijing (high cost) and Guizhou (low cost but remote). The table below shows a cost comparison for a 100 MW facility with similar specifications.

Anhui strikes a balance: lower CAPEX and OPEX than Beijing, but with better connectivity to AI downstream industries (e.g., automotive, display manufacturing) than Guizhou. For foreign companies that need proximity to supply chains for AI chip packaging (e.g., Hefei’s Jinjiang Electronics) or end-user AI applications (smart manufacturing, smart city projects), Anhui offers a cost-effective middle ground. Additionally, Anhui’s “integration with the Yangtze River Delta” means logistics costs to Shanghai Port are about ¥1,500 per container (vs. ¥3,000 from Guiyang).

Return on investment (ROI) analysis: Assuming a 100 MW facility with a total CAPEX of ¥2 billion (after subsidies) and annual OPEX of ¥600 million, and revenue from AI computing leasing (average ¥0.15 per GFLOPS/second per hour), the facility could generate ¥800–900 million in annual revenue. That yields a simple payback of 3–4 years, which is competitive with other Asian markets (e.g., Singapore: 5–7 years).

NEXT STEPS: Three Decision-Path Recommendations

Based on the cost analysis and risk factors, here are three actionable recommendations for foreign execs evaluating an AI facility in Anhui:

1. For high-density AI training needs (e.g., large language model training): Commission a detailed feasibility study with a local partner like the Anhui AI Computing Center (AAICC). We recommend jointly applying for the provincial “AI Computing Power Infrastructure” subsidy (30% CAPEX rebate) and negotiating a long-term PPA with Hefei Power Supply Bureau to lock in low tariffs. Budget for liquid cooling from the start to achieve PUE ≤ 1.2, which is required for the highest subsidy tier.
2. For cost-sensitive AI inference or edge computing facilities: Consider a smaller 10–20 MW facility inside an existing industrial park (e.g., Hefei Economic and Technological Development Zone) to share land and grid costs. Use domestic AI chips (Cambricon or Huawei Ascend) to reduce CAPEX by 30% compared to NVIDIA. Apply for the “Small and Medium AI Enterprise Support Program” which offers up to ¥50 million in grants for facilities under 30 MW. Aim for a 12-month construction timeline by using prefabricated modular data center solutions.
3. For joint ventures with local SOEs (to mitigate regulatory risk): Form a joint venture with a Hefei municipal investment company (e.g., Hefei Shengwei Technology) that holds at least 51% equity. This opens access to “zero land cost” leasing (free land use for 20 years) and a 5-year property tax holiday. The JV can apply for the “Anhui Province Key AI Project Fast Track,” which reduces approval time from 6 months to 2 months. Ensure the JV agreement includes a technology-transfer clause to protect proprietary cooling designs or AI algorithms.

— Anhui Gateway —