How to invest in the AI energy boom

Some analysts have proclaimed that AI is more important and will be more transformative for human development than the creation of the railways. With the world’s largest technology companies now spending more than $1 trillion a year expanding their presence in the market, there’s no doubt this theme will dominate for the foreseeable future.

However, as investors focus on the so-called hyperscalers – Alphabet, Microsoft and Amazon – which are building their AI infrastructure at an alarming rate, as well as chip manufacturers such as Nvidia and Micron, which are supplying the industry, not much attention is being directed to the infrastructure that will power this revolution. This is where bottlenecks are now starting to throttle growth.

The energy grid needs an upgrade to power AI

The critical one is the power grid. In the US, for example, the grid is around 50 years old and was not designed to handle the current level of rapid growth in demand. The graphics processing units, or GPUs, that underpin AI data centres today are vastly more energy-intensive than their previous counterparts.

Research compiled by Goldman Sachs and JPMorgan estimates that by 2027, AI server racks will require 50 times more power than the equivalents that formed the backbone of cloud infrastructure five years ago.

The computing power of any facility consumes only around 60% of the total energy requirement. The rest is taken up by cooling systems and other infrastructure.

As the hyperscalers expand, they are learning that Silicon Valley moves much faster than the rest of the world. GPUs have become 50 times more energy-intensive over the past five years, but global energy output has risen by just 1%-3% per year. In the real world, it can take five to seven years just to secure permits and sign initial contracts to build the power infrastructure. This has started to change in the past two years, but there’s still a long way to go. According to the International Energy Agency (IEA), global electricity consumption by data centres will double to 945 terawatt-hours (TWh) by 2030, representing roughly 3% of global demand for electricity. That’s roughly the same as adding 34 Hinkley Point C-scale nuclear-power plants to the global grid. Between 2025 and 2030, data-centre electricity consumption is expected to grow by 15% per year, four times the growth rate of total electricity consumption across all other sectors.

Electricity consumption from accelerated AI data centres, the most intensive units that train AI models, will rise by 30% annually. In the worst-case scenario, the IEA estimates that global data-centre demand for electricity could exceed 1,700 TWh by 2035, nearly 5% of global demand for electricity. If the industry becomes more efficient at utilising power, that figure could fall to 970 TWh. If the electricity industry fails to rise to the challenge, demand could be limited to 700 TWh by 2030, nearly 25% below the base-case scenario.

This bottleneck is most apparent in the US and China, where the most time and energy are being spent on AI development. China and the US will account for nearly 80% of global data-centre electricity consumption growth to 2030, according to the IEA. The US, in particular, is facing a projected power access shortfall ranging from 10.4 gigawatts (GW) up to 49GW by 2028, even though projections from the US Energy Information Administration (EIA) show the grid adding 86GW of new utility-scale electricity-generation capacity in 2026, the largest single-year rise since 2002.

Rise of the ‘Bring Your Own Power’ model for data centres

To get around some of these issues, data-centre providers are increasingly seeking to innovate. The “Bring Your Own Power” (B-Y-O-P) movement, for example, is bypassing grid-connection bottlenecks by building on-site microgrids, utilising utility-scale batteries, solar panels, fuel cells and wind and gas turbines. Elsewhere, data-centre operators and hyperscalers are working with utility providers to purchase and install natural-gas power stations.

Data-centre providers are also shifting their attention to gas-rich zones such as the Permian Basin in Texas and New Mexico, where natural-gas pipeline capacity is severely constrained (gas prices recently dropped below zero despite the war in the Middle East). Companies are building off-grid data centres directly at these extraction sites, monetising otherwise stranded gas that would have zero economic value.

For example, Microsoft (Nasdaq: MSFT) and Chevron (NYSE: CVX) are partnering to build a $7bn, 2.5GW off-grid natural-gas power complex in Pecos, Texas, specifically to supply Microsoft’s AI data centres under a 20-year agreement. Williams (NYSE: WMB), a pipeline company transporting a third of the natural gas moving across the US, is developing “Neo”, a $2.3 billion project utilising gas turbines paired with battery energy storage systems (BESS) for a major hyperscaler. This is the company’s fifth BYOP agreement.

Some providers are also turning to AI to help mitigate AI’s impact on power grids. A recent report from the World Economic Forum notes that “power-flexible” AI factories can dynamically modulate their electricity use, throttling energy-intensive tasks such as model training during periods of stress for the grid and routing more mundane tasks (such as answering simple questions on ChatGPT) to other locations. This flexibility enables data centres to capitalise on the volatile nature of renewable-energy generation.

(Image credit: Cheng Xin/Getty Images)

Energy prices take the strain

As the utility market has struggled to adapt to the surge in demand for electricity, prices have responded. Wholesale electricity prices have jumped across all global markets, and the impact is particularly acute in the US. In some eastern US states, prices have risen 76%. According to the Bureau of Labour Statistics, across the country, electricity prices are rising nearly 61% faster than general inflation. The entire supply chain is feeling the pain. Lead times for the production and delivery of grid equipment have skyrocketed. Standard electricity transformers now take 128 weeks to deliver, compared with just 16 weeks in 2019. In some cases, specialist transformers are being delayed for nearly three years.

The production of highly efficient combined-cycle gas turbines can take up to four years, more than double the length recorded in 2022, and across the entire supply chain analysts put the average price rise at 30% across all grid equipment. There’s also been a dramatic shortfall in the number of construction engineers and electricians, with the figure put at nearly 300,000 construction engineers and electricians in the US over the next decade. There are no quick solutions to any of these problems. While producers try to scale up output to meet rising demand, it looks as if they will continue to hold all the cards for the next five years at least.

There are three ways for investors to play this trend. There are the companies that generate power, those that make equipment for power stations, such as gas turbines, and those that manufacture cables and equipment to transmit electricity from A to B.

Tap into the AI energy boom with power players

(Image credit: Getty Images)

One of the hottest plays is GE Vernova (NYSE: GEV). Created as part of General Electric’s break-up, GE Vernova specialises in designing, manufacturing and maintaining equipment for the power-generation industry. Its technology provides roughly 25% of the world’s electricity and the group has an order backlog of $163 billion, or 3.5 times sales. Its order backlog for gas turbines sits at around 100GW – around 2.5 times the UK’s total daily electricity consumption. UBS has modelled 14% annual organic sales growth for the group through 2028 based on its current order backlog, with a 22.7% Ebitda margin by 2028, up from 8.4% in 2025.

Unlike the GPUs that power data centres, which have an estimated average life of around five to eight years, gas turbines can last up to three decades, locking in a multi-decade service contract for Vernova. The firms also offers kit for firms running older units (20 years and upwards) to help improve reliability and efficiency. Despite this growth and its key market position, there’s a lot priced into the stock at a mid-30s price-earnings (p/e) ratio, but UBS argues that the valuation is worth it given the revenues and potential for margin growth.

Rolls-Royce (LSE: RR), which came close to a government bailout in the pandemic, is now one of the world’s most sought-after power engineers. For the year to the end of 2025, the company reported a 12% jump in underlying revenue to £20 billion and underlying operating profit rose by 41% to £3,462 million, equating to a margin of 17.3%. Profit growth was driven primarily by the power-systems arm (25% of revenue), where the divisional margin expanded by 430 basis points to 17.4%. The firm put this down to “growth driven by data centres” and it’s hoping its “power-dense” next-generation diesel and gas engines will continue to drive growth. Its technology is in demand as data-centre providers seek alternatives to bypass ever increasing queues for power-grid connections. The FTSE 100 company recently noted that orders across gas and diesel engines in the first quarter was around 50% higher than last year and March was a record month. Power Systems’ order backlog was £7.3 billion at 31 March.

But Power Systems isn’t just about data centres. The business also produces battery energy-storage systems and engines for Leopard tanks. What’s more, last year Rolls-Royce conducted the world’s first successful test of a high-speed marine engine running on pure methanol. There’s also the company’s nuclear business. A long-time supplier of nuclear reactors to the Royal Navy, Rolls-Royce has begun moving into the civil market with its small modular reactors (SMRs). In June last year, Rolls-Royce’s SMR was chosen as the sole provider in the Great British Energy – Nuclear competition to build three SMR units in the UK.

Rolls-Royce SMR also received a strategic investment from CEZ Group, alongside a commitment for up to six units in the Czech Republic. In mid-June, the division was selected to deliver three SMRs on Sweden’s west coast in partnership with Videberg Kraft. Based on current estimates, Rolls-Royce is trading at a forward p/e of 38.2, falling to 32.7 in 2027, according to average analysts’ estimates. Those have ticked higher after the company’s latest upbeat trading update and Berenberg has pencilled in a forecast of £8 billion of share buybacks over 2026-2028, split by £2.5 billion in 2026, £2.7 billion in 2027 and £2.8 billion in 2028, with scope for more cash returns if cash flow beats projections over the coming months.

A US peer of Rolls-Royce is BWX Technologies (NYSE: BWXT). Like its UK counterpart, BWX has the backstop of a US Navy contract in its back pocket to support its general operations – it has been the sole nuclear-fuel provider to the US Navy for more than 70 years. It’s now seeking to grow in the civil market, where it provides specialised, complex, high-precision equipment used in nuclear reactors, including steam generators, reactor-pressure vessels and piping. It has an order backlog of $8.7 billion (around 2.2 years of revenue) bolstered by the recent $1.4 billion set of contracts through the US Naval Nuclear Propulsion Programme. However, at nearly 50 times forward earnings, there’s a lot baked into the current share price.

How to invest in the undersea cable kings

(Image credit: Getty Images)

The energy-transfer market is more concentrated than the other two potential investment segments. The renewable-energy transition has triggered an unprecedented global demand for ultra-high-voltage subsea cables to transport power from offshore wind and solar sites to urban centres, a market that did not exist 15 years ago. It’s currently dominated by a European oligopoly consisting of Prysmian (Milan: PRY), Nexans (Paris: NEX), and NKT (Copenhagen: NKT). These companies emerged as the winners in what was an incredibly competitive market, with lots of smaller players that couldn’t keep up with the capital-spending commitments required to manufacture vast undersea sea cables.

High-voltage direct-current (HVDC) cables can be thick, and they must be kept completely straight during manufacturing, which often requires companies to hang them inside skyscraper-high warehouses. The capital required to build this infrastructure runs into the billions. For example, the 500-kilometre Eastern Green Link 2 (EGL2) project in the UK, the single largest ever investment in electricity-transmission infrastructure in the country, has a price tag of £4.3 billion, with £2.7 billion of that for the cable itself.

The global high-voltage submarine cable market is expected to grow at a compound annual growth rate of 17.3% over the next decade. Production hit 7,000 kilometres in 2025, an all-time high, and the major players are rapidly ramping up production. Prysmian is drawing on its experience in this market to expand in the DC inside-building segment – essentially wiring up the power inside data centres. The company believes it will become a one-stop shop for data-centre construction contracts, building long-haul subsea connections and shore-based transmission infrastructure, and then for infrastructure throughout the building to power GPUs and air-conditioning units.

Management has estimated that overall global demand for DC power will expand at a compound annual growth rate of 33% over the next five years, with the bulk of this coming from AI-related data-centre growth. Analysts have pencilled in earnings growth of 25% for 2026, followed by 23% for 2027, with a net profit of €1.7 billion projected for 2027, up nearly ten times from 2020. Based on these projections, the shares are trading at a 2027 p/e of 24.9, which doesn’t seem too demanding for a high-growth business operating in an oligopoly.

Prysmian is around three times the size of its smaller peers, both of which are using their growing profitability and cash flow to expand into newmarkets. Of the two, Paris-listed Nexans is the cheapest, trading at a 2028 p/e of around 13 based on management’s growth targets. The group has laid out a road map to achieve an adjusted Ebitda of €1.2 billion by 2028 (up from about €750m) through growth in its three main businesses: PWR-Transmission, PWR-Grid and PWR-Connect. Sales are already locked in with a backlog of €7.9 billion by early 2026, enough to cover sales through to 2028.

Deals will also be a major part of the future growth plan. It recently added US-based Republic Wire to the stable to bulk out its US arm (about 15% of revenue). Republic reported sales of €52millionmn in its latest fiscal year and will be a key conduit for Nexans to enter the US data-centre market. Nexans plans to use Republic Wire’s established channels to sell its own comprehensive offering of medium-voltage and grid technology into premium US end markets. The acquired business is currently finalising a significant expansion programme, which will increase its production capacity by about 30% by the end of 2026.

How to play coal

Another FTSE 100 company that’s strategically well placed is National Grid (LSE: NG). Although still small compared with the US and Chinese markets, the UK data-centre market is the largest in Europe. National Grid believes demand for electricity in the UK will increase by 30% by 2035 to 290GW with a 90% increase in installed generation capacity to 370 TWh. To meet this demand, the company is investing £41 billion by 2031 to expand its regulated asset value by 60% to £60 billion. It is also going to invest £29 billion to expand its US business to a regulated asset value of £45 billion, with a focus on its key markets of New York and Massachusetts. The shares currently look cheap, selling at a forward p/e of 13.9.

One sector investors could also consider is coal. According to Global Energy Monitor, more than 2,200GW of coal-powered generation still operates worldwide, with another 710GW under development. China is scaling up its coal-output market to meet increased demand for energy and a total of 32 countries are proposing, or building, new coal plants to meet the growing need for power. At the beginning of June, Donald Trump announced plans to build two new coal plants in Alaska and West Virginia under the Defence Production Act, adding to the US coal fleet, which supplies 15% of the country’s demand for power. Alliance Resource Partners (Nasdaq: ARLP), Peabody Energy Corp (NYSE: BTU) and Warrior Met Coal Inc (NYSE: HCC) are three left-of-field plays worth considering here.

---

This article was first published in MoneyWeek’s magazine. Enjoy exclusive early access to news, opinion and analysis from our team of financial experts with a MoneyWeek subscription.