Author: Dong Jing
Source: Hard AI
&#34;Without electricity, there is no AI.&#34; Bank of America stated that the essence of the AI &#34;computing power competition&#34; is a &#34;power competition,&#34; and it is expected that investments in non-IT infrastructure supporting AI operations will experience explosive growth.
On November 3rd, according to Hard AI, Bank of America stated in its latest research report that the investment boom in AI infrastructure is expanding from traditional IT equipment such as chips and servers to infrastructure fields like electricity, cooling, and metals. These &#34;non-IT&#34; investments will become an important support for AI development.
The report pointed out that by 2030, China's total capital expenditure (Capex) on AI will grow to 2-2.5 trillion yuan, of which investments in non-IT infrastructure (such as electricity, heat dissipation, metal materials, etc.) supporting AI operations will account for one-third, forming a huge market of up to 800 billion yuan.
The bank believes that China has significant advantages in this round of AI infrastructure construction, including ample power generation capacity, relatively low electricity prices (30-60% lower than developed markets), a dominant position in renewable energy supply chains, and a relatively young and well-developed power grid network, laying a solid foundation for the development of AIDC (AI Data Centers).
<h2>The essence of AI competition has become a power competition</h2>
Bank of America stated that with the explosive growth in demand for AI model training and inference, the energy consumption of data centers is sharply rising.
The report predicts that China's electricity consumption by data centers will grow at a compound annual growth rate (CAGR) of 18%, from 102 terawatt-hours (TWh) in 2024 to 277 TWh in 2030, accounting for 29% of the total electricity consumption of global data centers.
The main driving forces behind this are three points:
<blockquote>
Accelerated proliferation of AIDC (AI Data Centers): Their power consumption is far higher than that of traditional data centers.
Deployment of high-performance chips: Taking NVIDIA as an example, its GB200 chip has a power consumption of up to 2.7kW, far exceeding the H100's 700W.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/d23bbedf-f2d0-4162-ac99-22f610a4ad74.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/d23bbedf-f2d0-4162-ac99-22f610a4ad74.png"/>
Surge in server cabinet power density: Future cabinets are expected to reach astonishing power consumption levels of 200kW or even 600kW.
</blockquote>
Against this backdrop, Bank of America expects that by 2030, total investment in AI in China will reach approximately 2-2.5 trillion yuan, with the key focus being on the 800 billion yuan non-IT infrastructure investment, where the power system (generation and transmission) will dominate with 38%, followed by metals needed for data center construction (12%) and advanced cooling systems (10%).
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/ddb255bc-d4c2-473d-8357-17c3e93120b3.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/ddb255bc-d4c2-473d-8357-17c3e93120b3.png"/>
<h2>Power supply: Five new opportunities in nuclear power, equipment, energy storage, and generators</h2>
The report emphasizes that compared to the aging power grids and electricity shortages faced by Europe and the United States, China has significant advantages in AI power infrastructure: ample power generation capacity, industrial electricity prices that are 30-60% lower, a dominant position in renewable energy supply chains, and a younger, more robust power grid.
The bank believes this provides a solid foundation for accommodating the explosion of AI computing power and has given rise to five major investment opportunities:
<blockquote>
Nuclear Power: As a stable, efficient, and clean 24/7 baseload power source, nuclear power is an ideal partner for AIDC.
The report predicts that China's nuclear power installed capacity will grow from 60GW in 2025 to 100GW in 2030, accounting for 60% of the global nuclear power capacity under construction.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/dfb2e55b-4e8b-46cd-835e-2d8f1c901b80.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/dfb2e55b-4e8b-46cd-835e-2d8f1c901b80.png"/>
Power Equipment: The global upgrade of power grids and the construction of AI data centers are driving a surge in demand for equipment such as transformers.
The report estimates that China's transformer export value will grow by more than 30% and 20% in 2025 and 2026, respectively.
Energy Storage Systems (BESS): Energy storage systems are essential to ensure stable power supply for data centers.
Bank of America expects that by 2030, the newly installed capacity of battery energy storage systems (BESS) related to AIDC will reach 55GWh, with a compound annual growth rate of up to 22% from 2024 to 2030.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/04b3dab5-8916-4bdf-8144-2a2ec58e1424.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/04b3dab5-8916-4bdf-8144-2a2ec58e1424.png"/>
Diesel Generators: As emergency backup power sources during data center outages, their demand rises with the construction of data centers.
The report predicts that the compound annual growth rate of sales for large-bore diesel engines used in data centers will reach 28% between 2024 and 2027.
AI Power Supply: The power systems within AI servers are also critical. With the significant increase in chip power consumption, emerging power technologies such as high-voltage direct current (HVDC), Panama power supply, and solid-state transformers will see strong growth.
</blockquote>
<h2>&#34;Cooling&#34; and &#34;Materials&#34;: The Physical Foundations of AI</h2>
Bank of America points out that in addition to electricity, efficient heat dissipation and basic raw materials are also the physical foundations for building the AI infrastructure, containing enormous investment value.
Liquid Cooling Technology: The report states that for every 10°C increase in server temperature, equipment reliability can decrease by up to 50%. Liquid cooling technology is 20-50 times more efficient in thermal conduction than traditional air cooling and can save up to 30% of electricity. As AI computing power density increases, liquid cooling becomes a necessary choice.
Bank of America predicts that China's liquid cooling market will grow rapidly at a compound annual growth rate of 42% from 2025 to 2030, reaching a market size of 79 billion yuan by 2030, with penetration rates jumping from 10% in 2025 to 45%
Key Metals - Copper and Aluminum: Copper plays a critical role in signal transmission, power delivery, and heat management in the construction of AI data centers.
The report estimates that by 2030, the copper directly consumed by AI data centers in China will reach approximately 1 million tons, accounting for 5-6% of the total national demand. Meanwhile, as a key material for lightweight and thermal conductivity, the demand for aluminum is also strong, with data centers expected to drive a demand of 695,000 tons of aluminum by 2030.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/9e954cf2-19f4-4b6b-b6d6-271a0ae30c7e.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/9e954cf2-19f4-4b6b-b6d6-271a0ae30c7e.png"/>

601985

Bank of America stated that the essence of the AI competition is essentially a "power competition," with the investment frenzy shifting from traditional "chip" technology to non-IT infrastructure such as power, cooling, and materials. It is expected that by 2030, China's related non-IT investment scale will reach 800 billion yuan, with power systems (generation and transmission) occupying an absolute dominant position of 38%, followed by metals required for data center construction (12%) and advanced cooling systems (10%)

From "Chip" to "Electricity", Bank of America: The scale of non-IT investment in China's AI infrastructure will reach 800 billion yuan