Copper, AI, and Data Centers: Analyzing the Demand-Supply Paradox

Copper, AI, and Data Centers: Analyzing the Demand-Supply Paradox

Several years ago, noted artificial intelligence leader Andrew Ng predicted that AI would be like electricity. It would be fundamental to all facets of our lives, but we wouldn't really notice it running in the background.

An interesting thing about electricity is that it would not have the impact it does on our lives without copper. The metal conducts electricity beautifully and is the core component of the wires that bring power to our homes from the grid, our rooftop solar panels, generators, or wherever else we're getting electricity.

What's funny about Ng's prediction is that is that AI too would not have the impact it is increasingly having on our lives without copper. AI requires enormous amount of data to do the things we ask it to do poan significant amount of energy to do the things we ask it to do. The numbers are wild:

  • On average, a ChatGPT query needs nearly 10 times as much electricity to process as a Google search.
  • A single ChatGPT query requires 2.9 watt-hours of electricity, compared with 0.3 watt-hours for a Google search, according to the International Energy Agency. Consider that about 40kW are used in a day by a 3,000 square foot home.
  • A traditional server rack in a data center typically needs between 10-15kW of power. Compare this to the 40-60kW of power needed for AI-ready racks – 4-6x more power and cooling usage.
  • The average size of data center facilities globally has reached 412,000 square feet, a nearly 5x increase since 2010.
  • Globally, over 7,000 data centers are either operational or at various stages of development, double the 3,600 in 2015.
  • Data centers have the capacity to consume a combined 508 TWh of electricity annually if operated continuously, more than the total annual electricity production of Italy or Australia.
  • Between 2022 and 2030, the demand for power will rise roughly 2.4%. Roughly 1 percent points of that will be tied to data centers. That kind of demand increase apparently has not been seen since the early 2000s.
  • Utility companies are being asked to increase the size of their generation fleets by 10-30% to accommodate the growing energy demands. US utilities will need to invest around $50 billion in new generation capacity just to support data centers alone.

Growing Demand for Copper

That new investment in generation capacity has me very curious to see what happens to copper demand over the coming decades if AI truly does become like electricity in our lives, not to mention the other big thing driving demand - the energy transition. Currently, Goldman Sachs says that data centers in the US only require about 100,000 metric tons of copper on an annual basis. That is less than 10% of how much copper China uses across all applications per month. I don't see how that projection makes sense considering the growth figures I outlined above. It seems to me $50B in new generation capacity, just in the US, will require quite a bit more copper. The data back me up:

  • It is estimated that it takes close to 65,000 tonnes of copper per every GW of applied power. The potential incremental copper demand just from the US could range between 0.5-1.5% of global copper demand.
  • Global copper demand is projected to increase by 20% by 2035, reaching 30 million mt/year, driven by electric vehicle production, electricity transmission grid expansion, and renewable power generation.
  • The demand for copper in power transmission and distribution networks is projected to rise to 1.8 million mt by 2035, up from the current 100,000 mt/year.

Challenges in Copper Production

Right now, projections for copper supply to meet demand say that things are pretty balanced. Look out over the next several years and you're looking at copper production needing to double. But there's a problem. Mining companies pulled back on their investment in copper production many years back for a number of reasons. Investors want to see dividends. Communities reject the development of projects. Companies view the political environment as unstable. The list goes on, and the journey to make up for the underinvestment in projects is a very long one.

  • Global development and expansion capital for primary copper mines peaked at $26.13 billion in 2013.
  • Capital spend on copper projects is estimated to have been only $14.42 billion in 2022.
  • There was an estimated 0.1% reduction year over year in development capital expenditure for copper projects in 2022.
  • The copper market is projected to have a deficit of 54,000 mt for 2024.
  • Total development capex is likely to hover under $8 billion for the time being.
  • New primary copper mines starting up between 2019 and 2022 had an average lead time of 23 years from discovery to commercial production. The time to conduct feasibility studies averaged nearly 6.6 years. Bankable projects averaged two years before construction began. The time between the start of construction and commercial copper production averaged 3.2 years.

The Tension and Potential Solutions

There's a real tension here as it feels like a perfect storm of rising demand and lagging supply. I have so many questions about the role innovation plays all along the mining lifecycle, what the future looks like for the future of our data center demand.

Prospecting and evalution - What role does AI play in accelerating that work. Wilson, I have questions!

Development - How do you ensure you've really got buy-in from the community in developing the mine? Are there opportunities to more deeply understand the world view of the community and craft a vision for a potential mining operation within that context? Could you more tangibly ensure incentives are aligned? What does it look like for a community to have a real ownership stake in an operation, beyond a new school and job prospects? Perhaps considerations like these prevent large mines from being shutdown? In the same breath, perhaps you get faster clarity on whether there's alignment to continue development, so you can move on.

Extraction - What additional technologies are out there that build on the shift to underground mining, in efforts to reduce the environmental impact of the mine?

Closing/Reclamation - How do we leverage those same considerations in the development phase to craft a vision with the community on what happens to the mine once it is closed? How do we build a vision that goes beyond extraction?

Data centers - While they are getting more efficient, that efficiency likely leads to boosted workflows and more data centers. The latest version of the hottest chip in the market uses three times more power than its predecessor, and customers show no sign of slowing down on their orders. We have data centers today that can fit eight 767s parked nose-to-tail. How much further are we going on expanding the footprint of these? How creative can we get in building out future data centers? What lessons can we learn from nature in coming up with those new designs?

Conclusion

For AI to become electricity, it will need a significant amount of electricity. Providing that electricity will require a significant amount of copper. Our copper production is already slipping into a deficit in an environment where capital expenditures , but copper has been on the back burner when it comes to investment. It will take significant investment to catch up. Those new copper mines that come online need to come online quicker. They and the new data centers that come online will need to be built with more excellence - better for the environment and more efficient. The challenge is real, as is the opportunity for innovation to help alleviate the pressure. I'm excited to learn about the people and organizations driving those innovations.

Readings:

O'Keeffe, Dunigan, Grant Dougans, David Crawford, Aaron Denman, Anne Hoecker, Jue Wang, Matthew Crupi, and Jim Wininger. “How to Feed AI’s Hunger for Power.”?Bain & Company,?www.bain.com/insights/how-to-feed-ais-hunger-for-power/.

Crooks, Ed. “The Energy Transition Sparks New Interest in Copper Mining.”?Wood Mackenzie, 9 Feb. 2024,?www.woodmac.com/news/opinion/the-energy-transition-sparks-new-interest-in-copper-mining/.

Mills, Richard. “Biggest Copper Mines Produced 20% Less Copper in 2023.”?Ahead of the Herd, 6 Apr. 2024, aheadoftheherd.com/biggest-copper-mines-produced-20-less-copper-in-2023-richard-mills/.

Energy Perspectives 2024. Equinor ASA, 2024,?www.equinor.com/en/sustainability/energy-perspectives.html.

Kalinic, Stjepan. “Vale To Boost Copper, Nickel Production With $3.3B Investment - Vale (NYSE:VALE).”?Benzinga, 21 June 2024,?www.benzinga.com/news/large-cap/24/06/33359723/vale-to-boost-copper-nickel-production-with-3-3b-investment.

“Views from the Floor - The Copper Bottleneck: What’s the Impact on AI and Net Zero?”?Man Institute | Man Group, 26 Mar. 2024,?www.man.com/maninstitute/views-from-the-floor-the-copper-bottleneck-whats-the-impact-on-ai-and-net-zero.

“Global Copper Demand to Rise 20% by 2035 to 30 Million Mt/Year: Nornickel.”?S&P Global Commodity Insights, 10 Oct. 2023,?www.spglobal.com/commodityinsights/en/market-insights/latest-news/metals/101339-global-copper-demand-to-rise-20-by-2035-to-30-million-mtyear-nornickel.

“AI Is Already Wreaking Havoc on Global Power Systems.”?Bloomberg.com, 28 May 2024,?www.bloomberg.com/news/features/2024-05-28/artificial-intelligence-is-already-wreaking-havoc-on-global-power-systems.

Li, Ying. “Copper Miners Enjoy High Profits, but Development Capital Expenditure Lag.”?S&P Global Market Intelligence, 26 Apr. 2023,?www.spglobal.com/marketintelligence/en/news-insights/research/copper-miners-enjoy-high-profits-but-development-capital-expenditure-lag.

Berman, Gregory. “Hedge Funds’ Bullish Copper Bets Run Into China’s Slowdown.”?MINING.COM, 22 May 2024,?www.mining.com/hedge-funds-bullish-copper-bets-run-into-chinas-slowdown/.

“AI Is Poised to Drive 160% Increase in Data Center Power Demand.”?Goldman Sachs, 14 May 2024,?www.goldmansachs.com/insights/pages/ai-is-poised-to-drive-160-percent-increase-in-data-center-power-demand.html.

Owen, Will. “Copper Must Scale up Production Within Sustainable Limits.”?Global Mining Review, 24 Apr. 2024,?www.globalminingreview.com/mining/24042024/copper-must-scale-up-production-within-sustainable-limits/.

The Markets: Copper Hits All-Time Highs. Goldman Sachs, 17 May 2024,?www.goldmansachs.com/insights/podcasts/episodes/17-may-2024-the-markets-copper-hits-all-time-highs.html.

要查看或添加评论,请登录

Kwame Som-Pimpong的更多文章

社区洞察

其他会员也浏览了