What an AI Spending Slowdown Could Mean for Copper, Energy and Construction
The AI spending boom reaches well beyond technology stocks. For investors in copper, energy and construction, the question is how much of their business depends on that spending, and what could support demand if it slows.
A copper miner, a power supplier and a construction company can all benefit from the same new data center.
That doesn't mean they would all lose business in the same way if its owner decided to build fewer of them. The contractor could have fewer projects to bid for. Copper suppliers could see weaker orders for the next round of buildings and equipment. A power plant could keep supplying electricity to facilities that are already running.
For investors, that difference matters. Owning the physical businesses behind artificial intelligence (AI) is a different exposure from owning the technology companies themselves. But some of their growth may still depend on the same spending plans.
Major buyers are still planning heavy capital spending. This article examines what could happen if they pulled back.
Building fewer data centers isn't the same as using less AI
A company can stop building and still keep its existing data centers running.
Amazon, Alphabet, Microsoft and Meta have guided to between $720 billion and $740 billion of capital spending in 2026 between them1. These are company-wide plans, with differing spending definitions, and include investment beyond AI. Now imagine one of them postpones its next campus, a site with several data centers, but keeps its existing facilities busy. It needs less construction work and fewer new electrical systems than planned. It still needs electricity for the servers it already has.
Steven Feldman, Wealthion's co-founder and chief executive, made the distinction in a recent interview: a data center earning less than expected is different from one being shut down. Even when returns disappoint, he said, “they still need electricity”2.
That doesn't make electricity demand immune to a slowdown. Existing facilities could become less busy, and a slower buildout could reduce future growth. But neither follows automatically from a smaller construction budget.
The International Energy Agency (IEA) projects that data centers worldwide will use about 950 terawatt-hours of electricity in 2030, up from 485 in 2025. A terawatt-hour is a billion kilowatt-hours. This April 2026 projection covers all data centers, including those serving conventional computing. It isn't an AI-only forecast3.
There is also no fixed conversion from dollars of AI spending to tons of copper or units of electricity. A budget buys different combinations of servers, buildings and other equipment, so cutting one part affects suppliers differently from cutting another.
Copper has many customers, but prices can still feel a slowdown
Copper has many customers beyond data centers. Their demand could cushion an AI-specific slowdown, but it would not necessarily prevent prices from falling.
Copper enters a data center through power cables, electrical connections and cooling equipment. Much of that demand arrives when facilities are built or expanded. The copper already installed doesn't have to be bought again each time a server runs4.
That makes AI an extra customer for a metal with a much broader market. In its 2025 assessment, the IEA estimated that copper use in conventional and AI data centers could reach 250,000 to 550,000 metric tons in 2030, roughly 1% to 2% of global copper demand. That's a dated estimate of one use, not a current measurement of AI's entire copper footprint. The electricity networks serving those facilities add another link to demand4.
Lyn Alden made the broader point in an interview published in October: AI matters at the margin, while construction elsewhere and other uses still account for much of commodity consumption. She also pointed to competing forces, including weaker growth in China's demand5.
The IEA's 2026 minerals report supports that wider picture. It lists electricity networks, transport, construction and industrial uses among copper's markets, and reports almost 28 million metric tons of global refined-copper demand in 20256. Those customers could cushion a slowdown limited to AI. They wouldn't necessarily cushion a recession that also weakens factory activity and building work.
A small share of demand doesn't mean a small move in price, either. If buyers and available supply were closely balanced, losing some expected orders could change what buyers were willing to pay. Inventories, mine disruptions and production decisions would matter too.
Supply also responds slowly. In an interview published in August, mining specialist Steven Enders described companies cutting budgets after a downturn becomes apparent and hesitating to invest when conditions improve7. So a slowdown can weaken today's demand while delaying tomorrow's supply. That adds uncertainty. It doesn't protect against falling prices.
Energy depends on which business is being paid
A gas producer, a power plant and a utility earn their income in different ways.
A gas producer sells fuel. A generator turns fuel, sunlight, wind or nuclear energy into electricity. A utility may supply customers and operate the network that connects them.
For natural gas, slower growth in data-center electricity use could mean less extra demand from gas-fired power plants. But heating, industry and exports also need gas. The United States (US) Energy Information Administration (EIA) forecasts US exports of liquefied natural gas, which is gas chilled for transport by ship, rising from 15.1 billion cubic feet per day in 2025 to 19.0 billion in 2027 as export capacity expands8.
That November 2026 forecast shows a source of demand beyond AI. It doesn't put a floor under gas prices. In the same outlook, the EIA expects US production to grow faster than domestic demand and exports. More customers can exist alongside ample supply8.
For electricity suppliers, location and contract terms matter. A provider serving a campus that is already operating faces a different risk from one financing a new plant or substation for a campus that hasn't opened. Other customers can't necessarily use the spare capacity in the same place, at the same time.
The IEA's October 2026 report highlights that risk. If demand is overestimated, grid investment can be left underused. Minimum-payment commitments, exit fees and other contract protections can shift some of that risk to the data-center customer, depending on the arrangement9.
So rising electricity use doesn't tell the whole investment story. Revenue depends on what is sold and on what terms, and returns on what the infrastructure cost to build and finance.
Construction orders can slow before current work ends
A contractor can stay busy on current jobs while customers approve fewer new ones.
A slowdown may show first in new awards and start dates, then in revenue as existing work is completed. Competition for fewer projects could also put pressure on pricing.
Exposure varies widely. Comfort Systems USA, which installs mechanical and electrical systems, reported that technology customers generated 60.8% of its revenue in the quarter ended September 30, 2026. That category is broader than AI data centers, so it isn't an AI revenue percentage10.
An order book offers some visibility, but its terms matter. EMCOR reported $18.57 billion of remaining performance obligations, meaning contracted revenue not yet recognized, at September 30. The total covers all its businesses, not just data centers. EMCOR said many construction contracts could be canceled, but that substantial economic penalties made the risk of cancellation low in its assessment11.
Existing contracts give some protection, though not a guarantee of timing. And a fall in new awards can matter even when they hold.
History offers one example. When telecommunications companies cut spending after 2000, revenue at Quanta Services, a contractor, fell 13% from its 2001 peak by 2002. Its operating profit of $222 million in 2000 became a loss of $159 million, after a write-down of acquired businesses12,13. Today's customers and funding differ, so this isn't a forecast.
Other construction work could help. Healthcare, manufacturing, public facilities and maintenance still need contractors. But winning replacement work takes time, and its profitability may differ. A crew, location or specialist skill suited to one project may not transfer at once to another.
The same slowdown can produce very different revenue losses
A market can have a small AI share while one supplier in it depends heavily on AI.
Copper as a whole has a broad market, with data centers as one small part. A contractor with a high share of revenue from AI-related projects could be more exposed to a slowdown in that spending. Profit could also fall by a different percentage from revenue, because some costs continue even when revenue declines.
Where a slower buildout could reach each business, and what else it sells to
| Business | Where a slower buildout could hurt | Demand beyond AI |
|---|---|---|
| Copper producers | Fewer new installations and weaker expected metal demand | Electricity networks, vehicles, buildings and industry |
| Natural gas producers | Less additional gas-fired generation than expected | Heating, industrial use and exports |
| Power suppliers | Slower growth in sales; underused new capacity | Existing customers, wider use of electricity (such as electric vehicles) and industrial activity |
| Construction contractors | Fewer awards, delayed starts and tougher bidding | Manufacturing, healthcare, public facilities and maintenance |
These are possible channels, not a forecast or ranking. None of the other customers guarantees that lost AI demand will be replaced.
When could each business feel a slowdown?
The timing depends on which projects are delayed, what has already been ordered or built, and how busy existing facilities remain.
New construction awards. Fewer approvals could reduce new work while contractors continue existing jobs.
Copper and electrical-equipment orders. Changes would depend on when materials were ordered and whether contracts allow deferral. Some equipment is ordered before construction teams mobilize14.
Growth in gas and power demand. Delayed openings could reduce or postpone the extra demand suppliers expected from new facilities.
Electricity for operating data centers. Demand could continue if facilities remain busy. It could weaken if their use falls.
What would make a slowdown more damaging?
The cause matters most.
A power-connection delay can postpone work without removing the customer's need for it. Disappointing AI revenue can cause a deeper rethink of expansion plans. A broad recession can weaken both AI spending and the other demand that supports these sectors.
The evidence would look different in each case. Fewer announcements are different from canceled contracts. Delayed openings are different from lower electricity use at operating facilities. Share prices can also react to changed expectations before those changes reach reported revenue.
What this means for investors
Copper, energy and construction are three different ways to take part in the physical buildout, with three different weak points.
How well each would hold up depends on the work already committed, the customers already being served and the demand beyond the next data center. A tangible asset can remain useful while the business that owns it earns less than investors expected.
Our free AI guide explores overlapping holdings and the physical buildout behind AI, helping explain where the same spending cycle can appear in different investments.
We are not making any recommendations.
Sources & Disclosure
1 Capital spending guidance for 2026: Amazon.com Announces Third Quarter Results, October 29, 2026; Alphabet 2026 Q3 Earnings Call, October 28, 2026; Microsoft Fiscal Year 2027 First Quarter Earnings Conference Call, October 28, 2026; Meta Reports Third Quarter 2026 Results, October 21, 2026. 2 Wealthion, What to Own if the Tech Boom Falters, Steven Feldman interview, video published October 30, 2026; distinction between disappointing returns and a facility shutting down. 3 International Energy Agency, Key Questions on Energy and AI, April 16, 2026, Executive summary; central electricity-demand projection for all data centers, 2025 to 2030. 4 International Energy Agency, Global Critical Minerals Outlook 2025, May 21, 2025, p. 111, Box 2.1; copper uses and estimated 2030 demand in conventional and AI data centers under the Stated Policies Scenario. 5 Wealthion, The Treasury’s Gold Revaluation Play, Lyn Alden interview, video published October 19, 2026; AI-linked commodity demand and demand beyond AI. 6 International Energy Agency, Global Critical Minerals Outlook 2026, revised July 2026, pp. 126 to 127 and Outlook; global refined-copper demand and markets. 7 Wealthion, Soft Jobs. Hard Talent: Mining’s Hidden Bottleneck, Steven Enders interview, video published August 12, 2026; mining investment responses through the cycle. 8 US Energy Information Administration, Short-Term Energy Outlook, November 10, 2026, Natural gas; US liquefied natural gas exports, production and demand outlook. 9 International Energy Agency, Powering Artificial Intelligence, October 14, 2026, Executive summary; uncertainty in electricity demand and allocation of infrastructure costs and risks. 10 Comfort Systems USA, Form 10-Q for the quarter ended September 30, 2026, Note 2, Revenue Recognition; revenue by customer type and business activities. 11 EMCOR Group, Form 10-Q for the quarter ended September 30, 2026, Note 3, Revenue from Contracts with Customers; remaining performance obligations and cancellation terms. 12 Quanta Services, Form 10-K for fiscal 2004, March 16, 2005; revenue, operating income and goodwill impairment, 2000 to 2004. 13 Quanta Services, Form 10-K for fiscal 2001, April 1, 2002; 2001 revenue. 14 DPR Construction, The Supply Chain Insights Shaping Tomorrow’s Data Centers, April 6, 2026; early equipment procurement before trade-partner mobilization.
Wealthion is a dba of Greylock Peak Investments, a registered investment adviser. This article is general information only. It is not investment advice or an offer or solicitation, and it is not tailored to your circumstances. Guests’ views are their own and may not reflect those of Wealthion. Wealthion has not verified guest statements, including any performance figures. Investing involves risk, including loss of principal, and past performance does not guarantee future results. Wealthion refers readers to other registered investment advisers and shares in the fees they receive from referred clients, which creates a conflict of interest. Please see our Form ADV Part 2A and consult your own financial advisor before investing.
What Serious Investors Are Watching
Dive into expert interviews, market analysis, and long-form content built to help serious investors think long-term.
Macro Bites – Metals Supply Tightens as Oil Shipping Costs Fall
At the Silver Institute’s Annual Industry Supply Dinner in NY, Metals Focus’ Managing Director Philip...
What an AI Spending Slowdown Could Mean for Copper, Energy and Construction
The AI spending boom reaches well beyond technology stocks. For investors in copper, energy and...
When Does a Market Selloff Become a Financial Crisis? | Steve Hanke
Steve Hanke connects the artificial intelligence investment boom with growing credit risk. The article explains how losses could spread, what historical crisis research establishes and which financial buffers could contain the damage.
Enjoyed This? Get More Insights
Expert insights and curated opportunities, delivered to your inbox.
Ready to Position for What's Coming?
Whether you're still learning or ready to act, your next step starts here.
- Independent
- Macro-Informed
- Real Asset Focused
Network Discussion
Join the Wealthion Network to Comment