Zuckerberg Warns of Energy Challenges in AI Data Center Expansion: Which Stocks to Watch?

In a recent interview, Mark Zuckerberg of Meta stated that energy limitations will impede the expansion of AI data centers.

Previously, there were also views expressing that energy is a key factor limiting AI development. In January, OpenAI CEO Sam Altman also pointed out at the World Economic Forum in Davos that artificial intelligence will consume more electricity than people anticipate, and future development will require breakthroughs in energy. In March, Tesla CEO Elon Musk predicted that following a chip shortage, the next constraint for AI computation would be electricity supply, and suggests that there may not be sufficient power for all chips by next year. Moreover, Jensen Huang has also previously suggested that the culmination of AI development lies in the advancement of solar energy technology and the enhancement of energy storage solutions.

Last week, Bank of America Merrill Lynchequity strategist Thomas (T.J.) Thornton anticipates that AI workloads' electricity consumption will grow at an annual compound growth rate of 25-33% over the next few years. By 2030, the power demand for data centers in the US could be approximately 250 TWh, equivalent to 8% of the total US power demand by 2030.

GPU performance enhancements lead to higher power consumption. Newer GPUs have more cores and higher clock speeds, increasing energy use. Furthermore, the number of transistors in GPU chips is continually rising to support stronger processing capabilities and better efficiency. Despite smaller transistors reducing power per unit, the growing number of transistors can still lead to greater overall power consumption.

Large-scale AI models, characterized by their computational complexity and extensive iterative processes, exhibit a substantial increase in power demand. Such models, which may encompass billions of parameters, necessitate considerable computational resources, correlating closely with energy consumption. Ian Bratt, Senior Technical Director at Arm, notes: "The efficacy and problem-solving capacity of neural networks enhance with scale, resulting in a proportional escalation in energy expenditure."

Tech companies are leveraging nuclear energy to strategize a new wave of data center deployments. Amazon Web Services (AWS) recently acquired a 960MW data center campus in Pennsylvania from Talen Energy, which sources power from the adjacent 2.5GW Susquehanna nuclear power station. AWS is not alone. Last year, Microsoft also took a step toward nuclear power by signing an agreement with the private nuclear fusion company Helion Energy in the United States. Helion is expected to provide power to Microsoft in about five years.

Nuclear power's high energy density allows it to generate substantial electricity in a small footprint, ideal for powering large data centers. Bank of America highlighted that data center growth is crucial for nuclear industry expansion. As the digital economy grows, data centers' rising power needs positively affect nuclear producers, potentially leading to higher electricity prices and profit margins.

Additionally, the long-term demand for nuclear power and the short-term constraints on uranium supply will continue to drive up uranium prices. According to the World Nuclear Association, there is not enough uranium supply in the market to meet the growing demand for nuclear energy, a trend driven by the global shift towards clean energy. The organization believes that by 2040, the supply may only meet 50% of the potential demand, even including the restart of mines and the full development of currently planned mines.

Bank of America Merrill Lynchforecasts that the direct increase in copper demand from data centers will reach 500,000 tons by 2026, taking into account only the copper used in equipment. Additionally, the increase in power generation necessitates the upgrade of the associated power grids. With its excellent conductivity and low energy consumption advantages, copper is also extensively used in power grids, further augmenting copper utilization. However, copper faces pressure on long-term supply due to declining ore grades, limited capital expenditure, and so on.

The closure of key mines has also garnered attention. Panama's government ordered First Quantum to halt operations at the Cobre Panama mine, accounting for 1% of global copper. The Canadian miner is seeking arbitration, but the mine may remain closed until 2029 or potentially not reopen. "This is a major catalyst for market tightening," says Jonathan Beigle of Ridgeline Royalties.

Rio Tinto's plan to open one of North America's largest copper mines in Arizona has become entangled in complex litigation. The state-owned Codelco, which accounts for a quarter of Chile's copper output, has been plagued by operational issues, causing its production to drop to the lowest level in 25 years. These factors also affect the supply of copper mines.

Citi analyst Maximilian Layton expects demand to outstrip supply by 1 million metric tons during the next three years. "Explosive price upside is possible over the next two to three years."

Nuclear power companies can establish suitable pricing within contracts with data centers and have certain cost advantages. Generators who sign new power purchase agreements with data centers can expect a premium on their contract prices, which will benefit revenue generation. Entities providing clean and reliable nuclear power generation hold a competitive edge in this respect. The ready access to cooling water and the typically expansive site areas of nuclear power plants undoubtedly reduce costs. These features not only facilitate the co-location of data centers on-site—saving on grid connection expenses—but also provide inherent operational efficiencies, as stated by Morgan Stanley.