Why is copper the oil of the AI era?

Copper is one of the earliest metals discovered and used by humans. As early as 8000 BC, our ancestors began to use copper to make tools and ornaments.

Nowadays, this ancient metal has attracted people's attention again due to the development of AI. In its research report "AI and Defense Place the Power Grid at the Center of Energy Security", Goldman Sachs expressed the view that copper will become the oil of the AI era.

This seems rather "incompatible" because when people talk about the rise of AI, most think of algorithms and models. Few realize that computational results don't appear out of thin air; they rely on electricity and hardware. ChatGPT consumes over 500,000 kilowatt - hours of electricity to process 200 million requests daily, equivalent to the daily electricity consumption of 17,000 American households.

What supports all this is precisely this metal from the prehistoric era.

Each high - end GPU chip requires a large amount of copper for heat conduction and circuit connection. It's not an optional accessory but a key factor determining the chip's performance.

Take the H100, the hottest in the AI field currently. Each GPU has 3,273 solder balls, and each ball corresponds to at least 1 - 2 copper wires. Moreover, the core chip connection part requires thousands of basic wires. Breaking any one of these copper wires may cause the GPU to stop working.

As the power consumption of chips soars from dozens of watts to over a thousand watts, traditional heat dissipation solutions have completely failed. Only copper materials can conduct heat outside the device.

The thermal design power of NVIDIA's H100 chip has reached 700W, and the power density of NVIDIA's latest GB300 server cabinet can reach up to 130kW, similar to that of a small - to - medium - sized diesel generator. To dissipate this heat, a large number of copper heat dissipation components are necessary.

1

GPU manufacturing is just the tip of the iceberg. The real big consumer of copper is the power grid infrastructure of data centers.

The copper consumption in data centers is growing at an astonishing rate. According to Schneider Electric's calculations, a 1GW data center uses about 65,800 tons of copper. In 2023, the global data center installed capacity was 7.1GW, with a total copper consumption of 467,000 tons, accounting for 1.7% of the global copper consumption. It is expected to reach 710,000 tons by 2026.

But this is only the copper consumption for data center construction equipment. There is also a huge copper demand for the power grid facilities supporting computing power operation. By 2026, the copper consumption for power grid facilities will reach 624,000 tons.

This is not an exaggeration. NVIDIA's GB200 super - chip uses a copper cable connection scheme, and the total length of cables inside the cabinet is nearly 2 miles. If we refer to the H100's shipment volume in 2024 and assume the GB200 also ships 350,000 to 400,000 units, it will require as much as 30,000 tons of copper.

Goldman Sachs' report states that by 2030, the global data center power demand is expected to surge by 160%, which means the existing power grid infrastructure needs large - scale upgrading and renovation.

The average operating life of the European power grid is 50 years, and that of the North American power grid exceeds 40 years. These aging power grids not only have low transmission efficiency but also struggle to meet the growing power demand. In 9 out of 13 regional power markets in the US, there was a critical power shortage in the summer of 2024. By 2030, almost all markets will face power shortages due to conductive materials.

In 2024, the annual copper consumption of the EU power grid was about 750,000 tons, about 400,000 tons in North America, and about 5.7 million tons in China. UBS predicts that from 2024 to 2030, the annual copper consumption of the EU and North American power grids will increase by 3% on average, with a cumulative increase of about 250,000 tons.

Why is copper indispensable for power grid upgrading? The answer lies in copper's unique physical properties. Copper's electrical conductivity is 1.6 times that of aluminum, with extremely low resistivity and a service life of 30 to 50 years, more than twice that of aluminum products.

In power transmission, copper's advantages are irreplaceable. Goldman Sachs predicts that by 2030, global power grid and power infrastructure construction will contribute 60% of the copper demand growth, an increase equivalent to the sum of the current global copper consumption and the annual consumption of the United States.

Goldman Sachs expects the copper price to reach $10,750 per ton in 2027. The China Securities Index for Non - ferrous Metals Industry rose strongly by 2.21%, and the non - ferrous ETF fund rose by 2.41%. Copper accounts for 28.66% of it, indicating that the market has started to reflect this trend.

The growth of copper mining and refining capacity is very limited. In the past decade, the compound average growth rate of global copper mine production was about 2.1%. The International Copper Study Group (ICSG) lowered its growth forecast in October 2025, and the global copper mine production is expected to increase by only 1.4% in 2025.

BMI under Fitch predicts that in the next decade, the average annual growth rate of global copper mine production will be far lower than the copper demand growth driven by new energy, AI and other fields. The core reason is the reduction of high - quality copper ore resources. For example, in Chilean mines, the average copper grade has dropped from 1.6% in 2010 to 1.1% in 2024.

Therefore, the rise of copper price is almost inevitable.

2

However, the rise of copper price is also creating new opportunities. As an important alternative to copper, aluminum is having its moment in the spotlight. Although aluminum's electrical conductivity is inferior to copper, in some application scenarios, its cost and weight advantages make it a better choice.

In the field of GPU heat dissipation, aluminum is widely used. The dense thin sheets on GPU radiators for heat exchange with air are almost all made of aluminum, aiming to provide the largest heat dissipation surface area within a limited volume. Moreover, the external structural brackets or shields of GPUs are usually made of aluminum or magnesium - aluminum alloy to reduce weight while ensuring strength.

Taking NVIDIA's H100 as an example again, the aluminum fin array on its radiator is large, with an estimated total weight between 300 grams and 700 grams.

The heat dissipation systems of data center servers also use a large amount of aluminum. If each cabinet is equipped with 80 GPUs, the total weight of the heat dissipation system will reach 64 - 96 kilograms if all use copper radiators. After using aluminum radiators, the total weight is only 24 - 36 kilograms, a reduction of 40 - 60 kilograms.

This not only reduces the bearing pressure on the cabinet structure but also allows more servers to be deployed under the same bearing condition, improving the computing power density of the data center. More importantly, the cost of aluminum radiators is only 40% - 50% of that of copper radiators, saving millions of dollars in investment for large - scale deployment.

In power transmission, the application of aluminum cables is also growing rapidly. Aluminum cables are widely used for high - current main power transmission at the building level, such as from substations to data center buildings and in vertical and horizontal main cable trays inside buildings.

Although aluminum cables need a larger diameter than copper cables to transmit the same amount of power, the comprehensive cost of aluminum alloy cables is about 20% lower than that of copper cables, and their weight is only about one - third of copper. For long - distance and fixed - laying scenarios, the advantages of aluminum cables are more obvious.

UBS is also optimistic about this, raising its aluminum price forecasts for this year and next by 5% and 2% respectively.

Global aluminum production has been continuously increasing, rising from 64.166 million tons in 2018 to 70.581 million tons in 2023, with an annual compound growth rate of about 1.9%. The global electrolytic aluminum production in 2024 is expected to reach 72.25 million tons, a 2.17% increase from 2023.

China is the world's largest aluminum producer. As of the end of 2024, China's cumulative primary aluminum production was about 550.4 million tons, accounting for 31.47% of the global cumulative production. In 2024, China's primary aluminum production was about 44 million tons, reaching a new historical high and accounting for 60.12% of the world's total aluminum production that year.

However, since 2017, China has launched a supply - side reform in the electrolytic aluminum industry, cleaning up illegal and irregular production capacity and setting a compliance production capacity cap of 45 million tons. As of the end of November 2024, China's installed electrolytic aluminum production capacity had reached 45.02 million tons, with an operating capacity of about 43.94 million tons and a capacity utilization rate as high as 97.74%. This means that there is almost no room for growth in China's electrolytic aluminum production capacity, and future production growth can only be achieved through a slight increase in the operating rate.

In contrast to China, overseas aluminum production capacity is expanding rapidly. From January to October 2024, overseas electrolytic aluminum production reached 24.82 million tons, a year - on - year increase of 1.4%. With the resumption of previously shut - down production capacity and the commissioning of new projects, the growth rate of overseas electrolytic aluminum production is expected to accelerate in 2025.

In terms of production share, as of October 2024, China's electrolytic aluminum production accounted for about 60% of the global total, while the production shares of the Gulf Cooperation Council countries and other Asian countries outside China were 8.6% and 6.6% respectively.

Although the global aluminum market will remain slightly oversupplied in 2025, this situation will reverse completely by 2026. Bank of America predicts that there will be a supply shortage of about 292,000 tons in the aluminum industry, and the aluminum price is expected to climb to $3,000 per ton in the fourth quarter of 2026.

3

However, copper and aluminum are just the prelude to this AI revolution. The real big market lies in cooling and water. Data centers are not only large electricity consumers but also large heat generators.

As the power consumption of AI chips continues to rise, traditional air - cooling heat dissipation solutions have reached their physical limits. The upper limit of air - cooling heat dissipation is generally 10kW to 15kW per cabinet, and only a few systems can reach 20kW. But when it comes to current GPUs, even the most powerful air - cooling systems are not very effective.

The limitation of power density makes air - cooling unable to meet the heat dissipation needs of high - power chips. The huge energy consumption problem leads to a high PUE (Power Usage Effectiveness) value, and a large number of fans and air ducts also take up valuable space.

What's worse, as the power consumption of AI chips further increases, the current dilemma will be exacerbated.

The advantages of liquid - cooling technology are overwhelming. The thermal conductivity of water is about 0.6W/(m・K), while that of air is about 0.024W/(m・K). That is to say, the thermal conductivity of water is 25 times that of air.

Looking at the heat absorption capacity per unit volume, the specific heat capacity of water is 4.2kJ/(kg・K), and that of air is 1.005kJ/(kg・K). The specific heat capacity of water is 4.18 times that of air, and its density is 833 times that of air. The product of the two is about 3,500 times. In actual engineering, due to the higher flow efficiency of the coolant, the heat dissipation capacity per unit volume will be further improved.

Liquid - cooling can significantly reduce the energy consumption of data centers by 20% to 30% or more, and lower the PUE to below 1.2, even reaching 1.05 in the immersion liquid - cooling solution. At the same time, the liquid - cooling system takes up less space, and the utilization rate of the computer room space can be increased by 30%, which is especially important in core cities where land is at a premium.

The market demand for liquid - cooling technology is growing explosively. The global liquid - cooled data center market size is expected to reach $2.84 billion in 2025, a year - on - year increase of 44.9%. It is expected to exceed $21.14 billion by 2032, with a compound growth rate of 33.2%.

Currently, the mainstream liquid - cooling technologies include three categories: cold - plate, immersion, and spray. Among them, cold - plate liquid - cooling dominates the current liquid - cooling market due to its low transformation cost and strong compatibility. Cold - plate liquid - cooling dissipates heat by indirectly contacting the liquid with the heat - generating components of the server, controlling the PUE below 1.25.

Even NVIDIA is exploring water - cooling plate technology. They are promoting a liquid - cooling solution called Micro - Channel Liquid - Cooling Plate (MLCP).

Because it uses a special coolant, the unit price of MLCP is about 3 to 5 times that of existing heat dissipation solutions. However, a single MLCP can stably handle a power consumption of over 2kW, with a maximum heat flux density of 800W/cm², about 4 times that of heat - pipe technology.

Therefore, the coolant itself is also a huge market. According to the prediction of MarketsandMarkets, the global coolant market size will grow from about $2.8 billion in 2025 to $21.1 billion in 2032, with a compound annual growth rate as high as 33.2%.

On the other hand, affected by the EU's PFAS (Per - and Polyfluoroalkyl Substances) restriction regulations, environmentally friendly coolants are rising rapidly. The market share of these coolants made of bio - based and hydrocarbon materials is expected to grow at a compound annual growth rate of 18.4%.

The development of liquid - cooling technology is not only a technological advancement but also represents the reconstruction of the entire industrial chain.

These growths are not just concepts or speculations but real physical needs. The growth of computing power will inevitably lead to an increase in power consumption, the increase in power consumption will inevitably lead to a demand for copper, and the growth of heat dissipation demand will inevitably lead to a demand for aluminum and coolants.

This article is from the WeChat official account “Zimu Bang” (ID: wujicaijing), author: Miao Zheng, published by 36Kr with authorization.