AI Energy Stocks: Europe's Power Play Behind AI

AI energy stocks became the fallback trade of the week. On Thursday 8 October, a Financial Times report that OpenAI's annualised revenue was about $20 billion below an earlier estimate pushed the Nasdaq down 1.25%. CoreWeave fell nearly 8% and Oracle 5.5%. We covered that day in detail in AI bubble fears and your portfolio.

After a day like that, many investors look for a "safer" way to own AI: the companies that sell electricity, cables, transformers and switchgear to data centres. In Europe that means names like Siemens Energy, Schneider Electric, Legrand and Prysmian.

Here is the problem this post is about. If you bought those stocks to diversify away from AI, you probably did the opposite. Below is the data, a worked euro example, and a simple way to measure your own AI dependency.

What are AI energy stocks?

AI energy stocks are listed companies whose growth depends on the electricity and power equipment that AI data centres need, without selling AI chips or software themselves. They fall into four groups:

• Power generation equipment. Gas turbines and generators (Siemens Energy).
• Grid equipment. Transformers, high-voltage lines and substations (Siemens Energy, ABB, Prysmian, Nexans).
• Electrical and cooling gear inside the building. Switchgear, busbars, UPS systems, racks, cooling (Schneider Electric, Legrand, Siemens).
• Utilities. The companies that build and run the networks and sell the power (Iberdrola, E.ON, Enel).

It is a second-order AI trade. You do not bet on which AI model wins. You bet that someone keeps building data centres, whoever wins.

Why AI power demand is a European story

The demand is real and measurable. In its Energy and AI report (April 2025), the International Energy Agency estimates that data centres used about 415 TWh of electricity worldwide in 2024. In its base case that doubles to about 945 TWh by 2030, just under 3% of all electricity used on Earth.

Europe grows slower than the US, but it grows. The IEA expects European data centre electricity use to rise by more than 45 TWh by 2030, up 70%. In the US the rise is about 240 TWh, up 130%.

The think tank Ember counts a wider European footprint. Its Grids for data centres report (June 2025) projects European data centre electricity demand going from 96 TWh in 2024 to 168 TWh by 2030 and 236 TWh by 2035. The two sources measure differently, but they agree on direction.

The local numbers are more striking. Ember found that in 2023 data centres already used 33% to 42% of the electricity in Amsterdam, London and Frankfurt, and nearly 80% in Dublin. Connecting a new data centre to the grid in those hubs takes an average of 7 to 10 years.

That last figure is why investors care about grid companies. In Europe, the shortage is not chips. It is the cable and the transformer between the power station and the server.

What a data centre buys besides chips

A rough shopping list for a new AI data centre, outside the servers:

How much do European grid stocks depend on data centres?

More than their sector label suggests. Here is what the companies themselves reported this year.

Schneider Electric. In February 2026, Reuters reported that data centres and networks made up about 30% of Schneider's orders. In its half-year results on 30 July 2026, Schneider said second-quarter revenue grew 17% organically, with demand "led by Data Center". Its Systems business, where data centre projects sit, grew 28% organically in the quarter.

Legrand. On 29 July 2026, Legrand said data centres made up 26% of its 2025 sales and almost half of its US turnover. It raised its 2026 sales growth target to 16% to 19%, from 10% to 15%.

Siemens Energy. On 12 May 2026, Siemens Energy raised its fiscal 2026 revenue growth outlook to 14% to 16%, from 11% to 13%. It tied gas turbine demand to US data centres, and Grid Technologies orders rose more than 41% to nearly €7 billion. In August it reported record orders of €17.9 billion and a €162 billion backlog.

Prysmian. On 20 July 2026, Prysmian signed a deal with Molex worth up to €5.5 billion over as many as 10 years, for optical cables used inside data centres. It plans to invest €1.25 billion through 2031 to expand fibre production.

Siemens. In February 2026, Siemens raised its earnings outlook and CEO Roland Busch said data centre demand had "considerably exceeded our expectations". The shares rose more than 6% that morning.

None of this is bad news. It is the opposite. But it means that a large part of these companies' growth, and of the price you pay for their shares, now depends on the same thing your Nvidia shares depend on.

The diversification trap: why boring industrials doubled your AI bet

Most portfolio views sort holdings by sector. Schneider, Legrand, Siemens Energy and Prysmian sit in Industrials. Iberdrola and E.ON sit in Utilities. On a pie chart they look like the opposite of a tech bet.

The sector label describes what a company makes. It does not describe who buys it. A transformer maker whose order book grows because of data centres moves with AI spending, whatever its label says.

You can see this in the guidance changes above. Legrand, Schneider and Siemens Energy all raised their 2026 outlooks, and data centres were a main reason each time. If the large cloud companies cut capital spending, those same outlooks are what gets cut.

There is a valuation point too. When an industrial company is re-rated as an "AI winner", its price stops being priced like a normal industrial. Siemens Energy traded around €147 on 12 October, about 23% below its 52-week high of €191.66 and still at about 47 times trailing earnings, according to StockAnalysis data. A cable maker or a turbine maker at that multiple is priced for growth, and growth is exactly what an AI slowdown would hit.

The European AI stocks post covered the chip and network side of the rally, ASML, Nokia and Aixtron. This is the same rally, one layer further out.

A worked example: a €40,000 portfolio

Take an investor with a €40,000 portfolio. They own a world ETF, one chip stock, and a group of European industrials and a utility they bought "to balance the tech".

First, the view most people see: holdings grouped by sector. The ETF is split using the MSCI World sector weights at the end of September 2026 (Information Technology 31.78%, Industrials 10.8%, Utilities 2.3%, from the MSCI index factsheet).

On this view, tech is 24% of the portfolio. Industrials and utilities together are 38%. It looks balanced.

Now the same portfolio by AI dependency. The shares below are assumptions for the example, explained in the next section. For the ETF we use the weight of the seven largest US tech companies, which was 24.76% of the MSCI World at the end of September (rounded to 25%).

The look-through method from our AI bubble post would stop at the ETF and ASML: €8,000, or 20%. The energy and grid holdings add another €5,500. That takes the real AI dependency to 34%, against a "tech" share of 24%.

The €10,000 this investor put into European industrials to balance tech carries €5,000 of AI dependency. In a 30% fall in AI-linked prices, the portfolio would lose about €4,050, or 10.1%. The investor who only counted the 20% would expect about €2,400.

These numbers are only as good as the 50% and 25% assumptions. Change them and the total moves. The point is not the exact figure. It is that the gap between the sector view and the AI view can be 10 percentage points or more.

If you hold your positions at DEGIRO, Zune.Money's DEGIRO portfolio tracker imports your transactions CSV and lists every holding with its euro value and weight, which gives you the first two columns of this table in a few seconds.

How to measure your AI dependency share in 6 steps

You need your positions, each ETF's factsheet, and the latest results summary of each single stock. Set aside 30 minutes.

1. List every position with its value in euros. Include all brokers and cash.
2. Look through each ETF. Use the factsheet's top holdings to find the weight of AI-linked companies. For a world ETF, the seven largest US tech companies are a fair start. Our guide to ETF overlap and Magnificent 7 exposure shows how to do this for several funds at once.
3. Put each single stock in one of four tiers. This is the step most people skip.
   • Core, 100%. It sells AI chips, chip-making machines or AI cloud capacity. Nvidia, ASML, TSMC.
   • Heavy, 50%. Data centres are 20% or more of sales or orders, or they were the main reason guidance went up in the last year. Schneider Electric, Legrand, and in our view Siemens Energy and Prysmian.
   • Partial, 25%. Data centre demand is a named growth driver, but most profit comes from elsewhere or from regulated assets. Most utilities, and broad conglomerates like Siemens.
   • None, 0%. No visible link. Insurers, food, consumer staples, most banks.
4. Multiply and add up. Value times tier, for every position. Divide the total by your portfolio value.
5. Compare it with your tech share. The gap between the two numbers is your hidden AI bet.
6. Stress-test it. Multiply the AI-linked total by 30%. Chip stocks as a group fell more than that in 2022, so it is a fair figure for a sharp correction.

Read the results press release before you assign a tier. Companies say plainly when data centres drove a quarter, because investors reward them for it. If a company does not disclose a split, choose the lower tier and note why.

The risks of AI energy stocks

The data centre story is strong. That does not make these stocks safe. Five risks matter.

Prices already assume years of demand. Schneider shares hit a record high of €275 in February after results. Siemens Energy traded at about 47 times trailing earnings in October. When a stock is priced for growth, a good quarter that is a little less good than expected can still send it down.

AI capex cuts hit orders first. Equipment makers live on order books. If the large cloud companies slow their data centre building, new orders slow within a quarter or two, even if revenue from the existing backlog keeps flowing for a while.

Grid bottlenecks cut both ways. The 7 to 10 year connection queue Ember describes is a reason to buy grid equipment. It also means some data centre projects get delayed or move to other countries, and the equipment orders move with them. The IEA estimates that about 20% of planned data centre projects worldwide could face delays if grid risks are not addressed.

Utilities are regulated, not growth stocks. For a network operator, a national regulator sets the return it may earn on its grid investment. More data centres can mean a bigger asset base, which helps over years. But the return per euro does not jump because AI is popular, and regulators can lower it. Buy a utility for steady income, not to ride AI.

Politics and energy prices. High European electricity prices, local limits on new data centres, and changes in how grid costs are shared between households and data centres can all slow the build-out here, regardless of global AI demand.

What to do with the number

None of this is advice to buy or sell any stock. A 34% AI dependency can be a deliberate choice. The risk is holding it by accident while you think you are diversified.

Decide your ceiling before the next sell-off. If you want AI-linked holdings below 30% of your portfolio, write that down now. When the number drifts above it, you trim or redirect new money. Our guide to portfolio rebalancing covers how to do it without paying more fees and tax than needed.

Real diversifiers have no data centre story. If you want parts of the portfolio that do not move with AI, look at sectors that do not sell to data centres at all: consumer staples, health care, insurers, government bonds. The diversification guide explains the options.

Watch the order books. Siemens Energy reports full-year results on 11 November. Schneider, Legrand and Prysmian publish third-quarter figures in the coming weeks. Their order growth tells you more about real AI demand than any private-company revenue leak.

Tax rules for selling differ by country, so check with a tax advisor before you rebalance a large position.

Keep the AI dependency in view

The six steps work in a spreadsheet, but prices move every day and your tiers drift with them. Import your DEGIRO CSV into Zune.Money and your holdings, weights and euro values stay current on their own, so you can redo the AI dependency sum in minutes after the next headline. The free plan covers the import and the full holdings list.