Edison explains: The picks and shovels of AI buildout – unlocking value in the upstream supply chain

Why are investors talking about ‘picks and shovels’ when it comes to AI?

The analogy comes from the California gold rush where the merchants selling tools to miners often earned more reliably than the prospectors. The same logic applies to AI. Nvidia, Broadcom, TSMC and ASML have absorbed the lion’s share of AI flows on the semiconductor side, while electrical equipment names such as Schneider Electric and Eaton have re-rated on grid and transformer demand. The picks-and-shovels trade is no longer a fresh idea. What remains underappreciated is the long tail of upstream suppliers, whose products are mechanically required by the AI buildout but which are not classified as AI names by index providers, sell-side analysts or ETF constructors. These are typically small- or mid-cap companies that sit one or two steps upstream of the visible end-product and still report financials dominated by weak industrial cycles rather than AI demand. As a result, many continue to trade on industrials multiples despite carrying meaningful technology-adjacent exposure.

How large is the AI energy demand and where does the investment opportunity sit?

AI workloads consume roughly 10 times more power than conventional computing. Goldman Sachs estimates data centre power demand could grow 160% by 2030, while McKinsey forecasts cumulative data centre investment of c $6.7tn through the decade. Meeting that demand requires far more than chips. A modern AI data centre needs land with secured grid access, steel and concrete, transformers and cabling, cooling systems, flame-retardant materials and dedicated power infrastructure. The opportunity therefore extends well beyond semiconductors into the industrial supply chains that physically enable AI deployment.

What does the physical bill of materials for an AI data centre look like?

A modern AI data centre is effectively a steel-framed industrial facility packed with high-density racks, each drawing well above 100kW and connected by kilometres of specialised cabling. HPE has reported that its Nvidia GB200 NVL72 rack draws 132kW, while Schneider Electric expects some AI rack densities to approach 1MW. The demand cascades across three broad areas: structural materials and construction; electrical safety materials such as flame-retardant compounds and low-smoke cabling; and power-banked land, which is increasingly the scarcest input of all. CBRE has identified power availability as the most important global site-selection criterion, with grid connection lead times now stretching beyond three years in many European markets. New grid connections around London face delays into the next decade, Frankfurt operators are now searching for new regional power access and Amsterdam has restricted very large IT loads.

This means anyone who banked land and secured grid connections a decade ago now holds an unusually durable competitive position. SEGRO, typically categorised as a logistics real estate investment trust, is a case in point. SERGO has a data centre pipeline of more than 2.5GW, located in or adjacent to established European Availability Zones. Its Slough Trading Estate sits inside the London Availability Zone, one of Europe’s largest data centre clusters. In 2025, SEGRO announced a joint venture with Pure Data Centres Group to develop a 56MW facility at Park Royal, with c £1bn of potential investment. Yet the market continues to value the business largely on logistics metrics rather than strategic data centre infrastructure exposure.

Who profits from Europe’s data centre buildout?

Steel fabrication, construction services and industrial metals form the skeleton of every data centre. In the UK, Severfield* has supplied projects linked to Microsoft, Telehouse and Google, with management identifying data centres as a key long-term growth area. Current earnings remain depressed by broader UK industrial weakness, but the company’s order-book mix is increasingly exposed to AI-related infrastructure demand. This pattern repeats across continental Europe, where a parallel buildout is drawing on a regional supply chain that runs through Southeastern Europe.

Viohalco, listed in Brussels and Athens, owns businesses spanning steel, aluminium, copper and high-voltage cables. Its subsidiaries include Sidenor, a major steel producer across Southeastern Europe, and Hellenic Cables, one of the region’s largest cable manufacturers. As hyperscalers increasingly move east in search of land and power capacity, suppliers positioned across multiple parts of the infrastructure stack have become strategically important. However, the market still tends to value these companies primarily on cyclical industrial earnings rather than on future infrastructure demand.

What is the grid infrastructure and energy opportunity, and which companies are best positioned?

AI data centres require not only large volumes of electricity, but continuous, uninterrupted power supply. Europe’s transmission infrastructure was not designed for this scale of concentrated demand, creating investment needs across substations, transmission networks and private power systems. Regulators are increasingly responding by placing pressure on hyperscalers to secure their own power supply rather than drawing from already-strained public networks.

This shift is creating demand for private grid infrastructure and independent transmission operators, and is opening up opportunities in markets that have historically been overlooked. ADMIE (IPTO), Greece’s Independent Power Transmission Operator, is executing a 10-year, €4.1bn upgrade to the Hellenic Electricity Transmission System, positioning itself at the intersection of grid modernisation and AI infrastructure build. Businesses supplying this ecosystem also stand to benefit.

On the generation side, nuclear energy is gaining ground as hyperscalers seek long-term, low-carbon energy supply. Microsoft, Google and Amazon have all signed power purchase agreements with nuclear operators, and small modular reactors are in active development across North America and Europe. Distributed solar is also gaining significant traction and is placing companies like Solar A/S directly in the supply chain for the energy transition that AI demand is accelerating. Solar A/S operates across electrical, heating, ventilation and energy solutions, positioning it inside the broader electrification and energy-efficiency buildout. The buildout of solar capacity also creates sustained demand for specialist cabling. Oman Cables manufactures specialised cables for photovoltaic systems, as well as the flame-retardant, low-smoke halogen-free cables mandated by tightening fire codes in data centre environments.

Monitoring the performance and safety of all of this infrastructure creates further opportunities. Vaisala provides dissolved gas analysis for critical power transformers, which is a real-time monitoring tool that helps utilities prevent failures. It also supplies the weather data services that allow solar and wind operators to optimise and forecast generation output. Furthermore, its measurement instruments underpin the environmental monitoring systems that data centre operators use to manage cooling loads and reduce power consumption.

Why does thermal management matter and what makes the cable materials story interesting?

AI chips run significantly hotter than conventional server hardware, and packing them into racks at densities approaching 1MW creates a thermal challenge that traditional air cooling cannot fully address. Cooling infrastructure accounts for a substantial share of total data centre energy consumption and inadequate temperature and humidity control can cause equipment failure at considerable cost. Companies like Lu-ve, an Italian industrial cooling specialist, manufacture heat exchange systems for mission-critical environments, including immersion cooling technologies designed for high-density AI workloads.

Higher rack densities also increase fire-risk requirements, which creates a less obvious but potentially important materials opportunity. Aluminium hydroxide is the dominant flame-retardant filler used in halogen-free cable compounds, while halogenated alternatives are progressively being designed out of building regulations due to toxic emissions during combustion. As data centre cable volumes rise and fire standards tighten, demand for these compounds increases mechanically. Nabaltec*, one of the world’s largest specialist suppliers of aluminium hydroxide, has explicitly identified AI data centres as a structural growth driver for its flame-retardant product lines. Yet the company’s reported earnings remain dominated by weak European industrial demand, leaving the AI tailwind only partially reflected in its current valuation.

What about semiconductors – is there still an underpriced angle?

The well-known semiconductor names, including Nvidia, Broadcom, TSMC and ASML, have already absorbed the bulk of AI-related flows. The more interesting opportunities may now sit further upstream. Compound semiconductors such as gallium nitride and gallium arsenide are increasingly important in power electronics and advanced communications systems, and their production requires highly specialised manufacturing equipment. Riber manufactures the molecular beam epitaxy equipment used to produce these advanced semiconductor materials. It also supplies deposition equipment for solar thin-film technologies and infrared detectors. Riber is not a chip manufacturing company; it supplies the equipment needed to manufacture the materials behind the chips. This positioning leaves it largely absent from AI-related investor flows despite its indirect exposure to several AI-linked technology cycles.

How can investors gain picks and shovels exposure at industrial valuations?

The common thread running through each of these companies is a timing mismatch. Their reported financials reflect industrial cycle conditions; their order books, pipelines and management commentary point towards AI-linked structural demand. The AI mix in orders tends to lead the AI mix in revenue by 12 to 24 months, which means the tailwind is real but not yet visible in current market prices. The opportunity lies in this gap. However, if hyperscaler capex slows down, grid delays and materials oversupply could hit upstream suppliers harder than the visible plays. But for investors willing to look one layer further down the bill of materials, many of these companies offer technology-adjacent exposure at industrial valuations. The trade is not about whether the buildout happens; it is whether these names are still priced as cyclicals when the first-derivative plays have long been priced as growth.

Edison Insight

The AI picks-and-shovels narrative is now mature enough that the obvious beneficiaries trade at AI multiples. But the market has not yet pushed the same logic one layer further upstream, into the materials, fabrication, cooling and land-banking businesses, whose products are mechanically required by every new hyperscale facility, but which do not appear in any AI index. These companies’ reported financials still reflect industrial-cycle weakness, but their order books, pipelines and management commentary tell a different story. The trade is not about whether the AI buildout happens; it is about who gets re-rated next.

Megatrends: energy transition, cities of the future, disruptive technologies, automation & industrial innovation

*Severfield and Nabaltec are clients of Edison Investment Research.