“There is no money there”

Hyperscalers are committing around US$700 billion this year to AI infrastructure. Yet, it seems that the investment to secure the required resource materials for this buildout has surprisingly not been addressed.

To illustrate this, a few months ago, I was talking with a group of investors in London, U.K. about circular economy strategies. One of them turned to me and said: “We do not invest in the circular economy because there is no money there.”

This sentiment has been troubling me ever since. One might conclude that, as a circular proponent and consultant for over a decade, I am envious that six of the largest U.S. hyperscalers are expected to spend around US$700 billion in capital expenditure this year, most of it driven by AI and data-centre expansion rather than investing in circular initiatives.

This isn’t a matter of envy over the attention AI receives. Rather, it is a matter of analysing two overlooked market realities.

AI is currently being widely celebrated for productivity gains, economic growth and technological advancement. Hence, investors from around the world, driven by FOMO, are throwing money at AI while, frankly, the returns required to justify this scale of AI investment have yet to be proven. So far, there is no money there either.

The second reality? As we all likely know, data centres are quite literally the physical foundation that AI requires. Perhaps our use of the term “cloud” for saving digital data outside of our computers has created a subconscious impression that our stored data, composed of immeasurable ones and zeros, are literally “in the clouds”.

But no, nothing floats. Every one and zero has an address, and that address is a data centre built of concrete, steel and a great deal of copper.

50,000 tonnes before the first server hums

In reality, every data centre begins as a simple hole in the ground well before the first server ever begins to hum. According to current 2026 industry data, a single, one-gigawatt (GW) AI-ready data centre campus can require up to 50,000 metric tonnes of copper due to its dense power distribution, heavy busbars, and advanced liquid cooling systems. And there are also the transformers composed of hundreds of kilometres of cables designed to turn electricity into computation.

Multiply that by the recent news of this US$700 billion a year hyperscaler infrastructure buildout and we can only conclude that the most important enabler of all, investment in the materials themselves, is simply not there. Even the world’s most well-capitalised tech giants cannot simply conjure raw materials out of thin air. When physical supply cannot meet deployment speed, the financial viability of these massive tech investments begins to fracture.

Allow me to explain.

Simple arithmetic tells us that we cannot get the required natural resources from the ground to meet the speed of data centres’ boom. For example, S&P Global projects that copper demand will rise from the present use of 28 million tonnes a year up to 42 million by 2040, resulting, if nothing changes, in a supply shortfall of at least 10 million tonnes annually. Analysts call this gap a systemic risk to many global industries. This grave concern is already being addressed by the markets with the price of copper touching an all-time high above US$13,000 per tonne in January of this year.

Copper, which is not even a critical raw material, is perhaps the most obvious resource depletion concern. E-mobility needs it. Grid expansion needs it. Wind turbines, drones and even the current rearming defence sector needs it. The data centre boom did not start this competition for copper, but it has exacerbated it. Hyperscalers are now competing with traditional utilities for transformers and other grid equipment already subject to multi-year lead times. In one of my in-company trainings in Dubai this year the competition for scarce resources was loudly reported as the most serious concern for their company. They are already experiencing delivery delays for some of critical technology required to run the grid.

Nvidia saves it, the grid spends it

To be fair, tech-sector promoters point out that hardware giants like Nvidia are now actively designing their way out of this shortfall. By shifting to high-voltage 800V DC power architectures, and co-packaged optical fibre interconnects, they are successfully slashing internal rack-level copper demand.

Yet, this possible efficiency gain will clearly not be enough. Bank of America and other energy analysts note that these internal data centre savings are completely wiped out by the staggering grid infrastructure expansion required to plug these clusters into the actual power grid. Generating and delivering that electricity acts as a 3.5x multiplier, demanding three to four times more raw copper for regional substations and heavy transmission lines than the actual data centres themselves. In essence, it is akin to the folly of sitting atop a tree and cutting the one tree branch upon which one is sitting.

This brings us to a profound irony: these enormous AI projects, rather than starving circular initiatives of capital, will ultimately become a major driver for the widespread adoption of circular economy principles.

To illustrate, let us return to the issue of our copper shortfall. The means to address this shortfall theoretically already exists as the required copper is already plentiful “above” ground rather than “below” it. Of the roughly 550 million tonnes of copper produced since 1900, an estimated two-thirds still remains in productive use. This enormous above the ground stock is often referred to as the “urban mine”. We wrote about Urban Miners in 2024: the second edition and the fifth one.

In its report “Recycling of Critical Minerals”, the International Energy Agency projects available end-of-life copper scrap rising to 27 million tonnes by 2050. This would be equivalent to around 70% of projected copper demand in its Announced Pledges Scenario. Hence, copper mining in our near future can easily be found in our already existing built environment. As we don’t like terms like “scrap”, due to its negative connotation, this resource is often referred to as “secondary” copper.

However, “theoretically” is the operative word, because as the London investor mentioned earlier informed me, “We do not invest in the circular economy”. His comment is borne out by recent statistics.

The consequence of this should not be surprising. The share of secondary copper in total demand fell from 37% in 2015 to 33% in 2023. The Circularity Gap Report (CGR) has also tracked this same direction globally. Currently, just 6.9% of materials entering the world economy are secondary, down from 9.1% when measurement began in 2018. This year’s CGR edition puts a price on this troubling trend. Its initial estimate of the Value Gap is €25.4 trillion in avoidable economic value loss annually to the persistent, traditional linear material resource extraction.

This is almost 31% of global GDP.

Or, to put it more simply, for every three euros of value the world creates, it discards one.

Moreover, the Circularity Gap Report highlights the lost resource pathways precisely attributed to the current digital economy boom by specifically identifying end-of-life waste as well as the resulting premature deterioration of fixed capital. A data centre server routinely being retired after only four years of use is not progress. By ignoring design from the outset, a fundamental circular economy strategy, the fallout consequence is premature capital deterioration.

Let me underline this further for our always well-informed Circular Times readers. The sources for the very materials that AI infrastructure urgently requires, as with copper, and that these AI investments will need to be paid off, are being ignored. It seems that our recent worshipping of the AI Data Centre god has indeed “clouded” recent circular economy progress accomplished over the last decade.

However, copper shortfalls, as with other resource needs, as it turns out, are not attributable to technology. After all, our current circular inspired recovery processes routinely extract over 90% of the copper in electronic waste. Rather, this shortfall consequence is attributable to both collection practices and initial design. In the United States, where the biggest data centres boom is happening, only 6% of copper having reached its end of life is actually recovered, and globally, only 25% of e-waste is documented as collected. The e-waste generated in 2022 contained metals worth an estimated US$91 billion. Yet only around US$28 billion in secondary raw-material value was recovered through documented urban mining.

US$60 billion a year goes into landfills while procurement teams are left worrying about supply security.

Products that have components glued, potted, and miniaturised rendering them to be unrecoverable are design decisions and these consequences can only be reversed by adhering to circular principles. For example, modular racks, standardised components, aluminium substituted for copper, where and when engineering allows, and using motors redesigned to need fewer rare earths are circular-inspired options.

The mine above our heads

What could fix this unsustainable cost?

Ironically, it will be the circular economy itself despite sentiments that “we do not invest in the circular economy because there is no money there”.

The IEA puts the entire investment needed in mining and refining, all critical minerals, through 2040, at over US$750 billion. That is roughly one year of hyperscaler capital expenditure. Recycling shrinks that bill, cutting required mining investment by about 30% and reducing new copper mine development needs by up to 40% by 2050. The circular infrastructure itself, the collection systems, secondary smelters, and e-waste recovery capacity, comes in even less expensive, and with less embodied carbon, because the ore grade of a secondary material is superior to original ground mine extractions. The IEA estimates that around US$600 billion of mining investment will still be required through 2040 under our announced climate pledges. Without greater recycling, the bill will be around 30% higher.

Moreover, there is also the cost related to the concept of time itself.

I always cover this factor in all my corporate training. A hyperscaler converts capital into an operating data centre in 18 to 36 months. A new copper mine takes more than 20 years to develop. No solution bridges these gaps. Only material already above ground can provide the required resources on the same schedule as the buildout itself. This ultimately, and ironically, makes circularity not only a sustainability nicety appended to the investment case, but the only supply strategy that can match these new US$700 billion AI hyperscaler infrastructure resource requirements.

So, when an investor says “we do not invest in the circular economy because there is no money there”, ask them directly whether they have ever considered the security of their AI investments. (In my case, it was a man.)

And as a proponent of the circular economy, I confidently feel that this recent data centre investment boom will actually build demand for the circular economy.

P.S. For full disclosure, and in the spirit of this subject, this article was, in part, researched with the help of an AI system running from within one of these new data centres. Whether this is actually ironic we will leave up to you to decide.

Sources

S&P Global (2026). Copper supply shortfall analysis: demand rising from 28 to 42 million tonnes by 2040, 10 million tonne deficit.Via SDxCentral. https://www.sdxcentral.com/news/ais-appetite-for-copper-poses-a-systemic-risk-to-data-center-buildout/

Circle Economy and Deloitte Netherlands (2026). The Circularity Gap Report 2026: The Value Gap. EUR 25.4 trillion annual value loss, 6.9% global circularity.https://dashboard.circularity-gap.world/report/2026/cgr-2026-overview

IEA (2024). Recycling of Critical Minerals: copper scrap growth to 27 Mt by 2050, secondary share 17% to 40%, e-waste metal values, mining investment savings.https://www.iea.org/reports/recycling-of-critical-minerals/executive-summary

IEA (2026). Global Critical Minerals Outlook 2026: $750+ billion investment need through 2040, recycling rate outlook, rare earth diversification cost.https://www.iea.org/reports/global-critical-minerals-outlook-2026/executive-summary

IEA (2024). News release: $600 billion mining investment through 2040, 30% higher without recycling; recycling market value of $200 billion by 2050.https://www.iea.org/news/policy-momentum-behind-critical-minerals-recycling-gathering-pace-but-greater-uptake-required

Wood Mackenzie / Financial Times (2025). Hyperscalers outbidding grid suppliers for transformers; refined copper deficit forecasts. Via Tom’s Hardware. https://www.tomshardware.com/tech-industry/ai-data-center-buildout-pushes-copper-toward-shortages-analysts-warn

Industrial Info Resources (2026). Data centre expansion plans of $600+ billion for 2026; US Project Vault strategic reserve.https://www.industrialinfo.com/iirenergy/industry-news/article/how-will-tight-copper-market-affect-data-center-growth–353673

Data Center Knowledge (2026). Critical minerals supply chain strain; 50,000 tonnes of copper per hyperscale data centre; offtake agreement trends.https://www.datacenterknowledge.com/supply-chain/the-critical-minerals-crisis-ai-data-centers-face-supply-chain-strain

Yahoo Finance / Cryptoprowl (2026). Copper price above $13,000 per tonne; US critical mineral designation, November 2025.https://finance.yahoo.com/news/copper-shortage-threatens-data-centre-200100745.html

Resources, Conservation and Recycling (2023). Sector-level estimates for global future copper demand: US above-ground stocks 1.8 times underground reserves; 6% end-of-life recycling rate.https://www.sciencedirect.com/science/article/abs/pii/S0921344923000782

Sverdrup and Ragnarsdottir. Copper stock in use in society, approximately 550 million tonnes; primary production peak projections. Via Phoenix Refining. https://www.phoenixrefining.com/blog/the-approaching-era-of-copper-scarcity

World Economic Forum (2024). Global copper recycling rates and secondary supply potential.https://www.weforum.org/stories/2024/12/mining-s-new-frontier-developing-copper-recycling-for-a-more-sustainable-future/

This article is part of our work on circular economy and artificial intelligence. MGT OPEN’s executive programme Circular Economy and Artificial Intelligence: Leading the Twin Transition takes place on the Crikvenica Riviera, Croatia, 14 to 18 June 2027. Details and registration on MGT OPEN.

Author: Dr Gordana Kierans