The market is still treating AI capex and data-centre growth mainly as a chips-and-servers story. But the harder constraint is increasingly power: getting enough electricity to the site, and then getting that power from the grid to the rack, and from the rack to the chip.
That shift matters for investors because it changes where value shows up. The winners are not only chipmakers and cloud platforms. They can also be utilities, grid equipment makers, electrical infrastructure suppliers, power-and-cooling vendors, and companies that help data centres secure reliable power faster.
This is the link worth focusing on:
AI capex → denser racks → higher power and cooling needs → grid and interconnection stress → more spending on power equipment, utilities, transmission, backup/on-site generation, and power-first data-centre design.
[The Economic Times (Stephen Schwarzman at Davos, Jan 2026); The Economic Times (Vertiv CEO interview, Feb 24, 2026); IEA Energy and AI; US DOE/LBNL data centre electricity report]
Why this is more than a temporary “AI boom” story
Stephen Schwarzman’s Davos comment matters because it frames the issue as a power shortage investment problem, not just a software trend. In his Economic Times interview, he said one of the biggest opportunities now is the shortage of electricity and power in the developed world, especially the US, tied to the scale of data-centre buildout for AI. He also described the capital need as running into the trillions. [The Economic Times (Stephen Schwarzman at Davos, Jan 2026)]
That thesis now has support from industry and grid data:
- The IEA projects global data-centre electricity use to more than double to around 945 TWh by 2030, with AI a key driver. [IEA Energy and AI, 2025]
- The US DOE/LBNL report says US data centres used about 4.4% of total US electricity in 2023 and could rise to 6.7% to 12% by 2028. [US DOE, Dec 2024 release on LBNL report]
- PJM said its 2027/28 auction demand forecast was about 5,250 MW higher than the prior auction, with nearly 5,100 MW of that increase tied to data-centre demand. [PJM / PJM release reporting, Dec 2025]
This is why the “AI capex” trade is spreading into sectors that looked unrelated a year ago.
The key connection: data-centre density is turning power infrastructure into the bottleneck
The recent Economic Times interview with Vertiv CEO Giordano Albertazzi is directly relevant because it explains, in simple operating terms, why this theme is real now.
He says the industry is moving beyond roughly 100 kW racks, heading toward ~600 kW, and that 1 MW racks are not far-fetched. He also describes the “power train” needed to move megawatts/gigawatts from the grid to the chip (transformers, switchgear, UPS, power distribution, busways), and the thermal shift from air cooling to liquid cooling. [The Economic Times (Vertiv CEO interview, Feb 24, 2026)]
That is the investment bridge between:
- AI compute demand
- data-centre capex
- electrical equipment orders
- grid upgrades
- local power shortages / delays
A separate industry datapoint points in the same direction: DatacenterDynamics reported Satya Nadella saying Microsoft had AI GPUs sitting in inventory because it lacked power to install them. Even if one treats that as anecdotal, it is a strong real-world signal that the bottleneck is moving from silicon supply toward power availability and delivery. [DatacenterDynamics, Nov 2025]
Why this can still work after the move in AI and infrastructure stocks
Many investors now accept that AI will drive data-centre spending. Fewer are underwriting the second-order capex cycle: the power side.
1) The constraint is physical, regulated, and slow
Power infrastructure cannot be scaled at software speed. Interconnection queues, permitting, transformer lead times, and transmission buildout take time. Berkeley Lab’s queue data shows ~10,300 projects seeking interconnection at end-2024, representing ~1,400 GW of generation and ~890 GW of storage, while completion timelines remain long. [Berkeley Lab “Queued Up: 2025 Edition”]
2) Grid stress is now visible in market signals
PJM’s auction results and large-load policy debate show data-centre demand is no longer a niche planning issue. It is showing up in reliability and capacity market discussions. Reuters also reported PJM plans/rules to speed large-load connections, including frameworks where new loads bring generation or operate under curtailment conditions. [PJM; Reuters, Jan/Feb 2026]
3) The supply chain for electrical gear is still tight
Reuters reported major grid equipment makers expanding US manufacturing, with surging demand in transformers and high-voltage equipment, and very long lead times (including GSUs). [Reuters, Dec 2025]
4) Big tech is adapting around the bottleneck, not waiting for it to disappear
Reporting now shows an “all-of-the-above” power strategy (renewables + gas + nuclear + storage + on-site options), and even examples of private or dedicated power approaches for data centres. That supports the idea that power access has become strategic, not just an operating cost line. [Reuters, Dec 2025 / Jan-Feb 2026; Washington Post, Feb 2026]
The correlation investors should care about (and how to think about it)
There is a correlation between power, data centres, AI capex, and electricity deficiency, but it is not a simple one-line relationship.
What is strongly correlated
- More AI capex generally means more accelerated compute
- More accelerated compute generally means higher rack density
- Higher rack density means more spending on power delivery and cooling per rack
- Faster data-centre buildout increases pressure on local grid capacity, interconnection, and equipment supply
[IEA; DOE/LBNL; PJM; Vertiv CEO interview]
What is not linear
If grid access is delayed, the spend does not always disappear. It can shift:
- to another geography
- to a different utility territory
- to on-site generation / storage
- to phased deployment
- to more spending on electrical/cooling systems to make denser sites work
This is why the theme creates multiple investable paths rather than one single “winner takes all” trade. [Reuters; Washington Post; CRU Ireland policy]
Where the demand is structurally strongest
1) United States: biggest AI load growth, biggest power bottlenecks, biggest power capex response
The US remains the main centre of the AI data-centre buildout. Schwarzman highlighted the concentration of data-centre buildout in the US in his Davos remarks. [The Economic Times (Stephen Schwarzman at Davos, Jan 2026)]
The system-level evidence supports that:
- IEA sees the US as the largest part of the increase in data-centre electricity demand. [IEA Energy and AI]
- DOE/LBNL shows rapid growth in US data-centre power use. [US DOE/LBNL]
- PJM data shows data centres driving a large jump in load forecasts. [PJM / PJM release reporting]
- NERC’s 2025 LTRA flags growing resource adequacy risks and specifically points to demand growth from data centres and other large loads. [NERC LTRA, Jan 2026]
- Southern Co raised its five-year capex plan to $81 billion and cited data-centre and industrial demand; Reuters reported ~10 GW contracted and ~75 GW of data-centre connection interest. [Reuters, Feb 19, 2026]
US takeaway: This is the deepest market for the theme, but also the most crowded. The cleaner setups are often in the “picks and shovels” (electrical gear, grid spend, power/cooling) rather than only in data-centre landlords.
2) Europe (including UK/Ireland): power policy and grid constraints are becoming part of the investment case
Europe’s setup is different. The issue is less about one giant AI cluster and more about grid readiness, policy rules, and who gets connected on what terms.
Useful signals:
- The European Commission has repeatedly highlighted the need for major grid investment and referenced an estimate of €584 billion in grid investments needed by decade-end to meet REPowerEU goals. [European Commission, May 2025 consultation note / grids package materials]
- The Commission’s own communications on data centres reference the IEA’s trajectory toward ~945 TWh by 2030. [European Commission, Nov 2025 data-centre note]
- Reuters (citing a McKinsey report) reported Europe’s data-centre power demand could nearly triple by 2030. [Reuters, Oct 2024]
- Ireland’s CRU introduced a stricter large energy user / data-centre connection policy, including a requirement that data centres meet at least 80% of annual demand with additional renewable power and a glide path. [CRU, Feb 2026 decision]
Europe takeaway: In Europe, the thesis is less “AI spending = immediate volume” and more “AI/data-centre growth + climate rules + grid bottlenecks = durable spending on grids, electrification equipment, efficiency, and connection-enabling infrastructure.”
3) India: AI ambition is rising fast, but power and transmission execution matter even more
India is compelling because the AI/data-centre buildout is now being discussed alongside very large domestic investment plans and local manufacturing opportunities.
Recent signals:
- Reuters reported major AI infrastructure commitments at an India AI summit, including large planned investments by Reliance, Adani, Microsoft, Yotta, and L&T-Nvidia partnerships. [Reuters, Feb 19, 2026]
- Reuters separately reported Adani’s plan to invest $100 billion by 2035 in renewable-powered, AI-ready data centres. [Reuters, Feb 17, 2026]
- Vertiv’s CEO told ET that India is one of its fastest-growing markets, with powertrain and thermal-chain components manufactured in India (Pune), supporting the idea that local manufacturing can benefit from AI infrastructure growth. [The Economic Times (Vertiv CEO interview, Feb 24, 2026)]
- Reuters reported Maharashtra is already using local solar and grid workarounds to manage transmission bottlenecks while planning large capacity additions to meet rising demand from data centres, EVs, and green hydrogen. [Reuters, Feb 19, 2026]
India takeaway: India may offer the clearest “power-first AI” setup. The upside is not just data centres. It is the full chain: generation, transmission, distribution upgrades, electrical equipment, cooling, and domestic manufacturing.
How to express the thesis in public equities (1–3 year horizon)
The cleanest way to build this theme is by role first, then by region.
Role 1: Grid buildout and transmission/distribution bottlenecks
These names benefit if the bottleneck is getting power to the site.
United States
- GE Vernova (GEV) – grid equipment, power systems, electrification exposure
- Eaton (ETN) – electrical systems, switchgear, power management
- Hubbell (HUBB) – utility/distribution components
- Quanta Services (PWR) – grid and electrical EPC / utility infrastructure
- MYR Group (MYRG) / MasTec (MTZ) – transmission/distribution buildout contractors
- Powell Industries (POWL) – electrical equipment / switchgear exposure
Europe / UK
- Schneider Electric (SU.PA) – data-centre electrical + energy management
- ABB (ABBN) – electrification, switchgear, automation
- Siemens (SIE.DE) – electrification and industrial infrastructure
- Siemens Energy (ENR.DE) – grid and power systems exposure
- Prysmian (PRY.MI) / Nexans (NEX.PA) – cable and grid buildout
- Legrand (LR.PA) – power distribution and data-centre components
- National Grid (NG.L) / SSE (SSE.L) – UK grid/power capex beneficiaries (regulated utility angle)
India
- Power Grid Corp (POWERGRID) – transmission backbone
- NTPC (NTPC) – generation + grid-linked capacity growth proxy
- Tata Power (TATAPOWER) / JSW Energy (JSWENERGY) – generation + infrastructure buildout
- Siemens India (SIEMENS) / ABB India (ABB) / Hitachi Energy India (POWERINDIA) – electrical equipment and grid technologies
- CG Power (CGPOWER) – transformers/switchgear/electrical equipment
- KEC International (KEC) / Kalpataru Projects (KPIL) – transmission EPC exposure
What confirms this leg: bigger utility capex plans, rising order books for transformers/switchgear/cables, faster execution on T&D projects. [Reuters; EU Commission; Southern Co]
Role 2: Data-centre power and thermal infrastructure
These names benefit if the bottleneck is inside the data centre (rack power, UPS, distribution, cooling).
United States
- Vertiv (VRT) – direct beneficiary of higher rack density, powertrain, liquid cooling
- Eaton (ETN) – overlaps with both grid and data-centre electrical systems
- nVent (NVT) – electrical enclosures/thermal/electrical infrastructure exposure
- Carrier (CARR) / Trane (TT) / Johnson Controls (JCI) – cooling and thermal systems (more indirect, but relevant)
Europe / UK
- Schneider Electric (SU.PA) – strong data-centre power/cooling ecosystem exposure
- ABB (ABBN) – electrical infrastructure and automation
- Legrand (LR.PA) – data-centre power distribution components
- Siemens (SIE.DE) – electrical and building infrastructure exposure
- Munters (MTRS.ST) – cooling / air treatment exposure (specialist angle)
India
- Blue Star (BLUESTARCO) / Voltas (VOLTAS) – cooling/HVAC exposure (indirect but relevant to thermal chain)
- Cummins India (CUMMINSIND) / Kirloskar Oil Engines (KIRLOSENG) – backup power genset ecosystem exposure
- Techno Electric & Engineering (TECHNOE) – electrical infrastructure + data-centre development exposure (hybrid proxy)
- ABB India / Siemens India / CG Power – also relevant here via electrical systems supply
What confirms this leg: more commentary like Vertiv’s on rack density and liquid cooling, rising order intake from data-centre projects, and improving delivery visibility. [The Economic Times (Vertiv CEO interview, Feb 24, 2026)]
Role 3: Utilities and power suppliers to large-load growth
These names benefit if the bottleneck drives regulated capex, rate-base growth, and contracted power demand.
United States
- Southern Co (SO) – clear large-load demand signal in recent reporting
- Dominion Energy (D) – data-centre-heavy territory exposure (Virginia)
- Constellation Energy (CEG) / Vistra (VST) / Talen Energy (TLN) – merchant/contracted power angle tied to large-load demand
- NextEra Energy (NEE) – broader utility/renewables/network exposure (less pure-play, but relevant)
Europe / UK
- National Grid (NG.L), SSE (SSE.L), Terna (TRN.MI), Redeia (RED.MC), Enel (ENEL.MI), Iberdrola (IBE.MC), RWE (RWE.DE) – grid and power-system investment beneficiaries, depending on country regulation and capex execution
India
- NTPC, Power Grid, Tata Power, Adani Power, JSW Energy, NHPC, SJVN – different mixes of generation, transmission, and regulated/contracted growth
What confirms this leg: utilities raising capex plans and load forecasts because of data-centre demand, especially in fast-growing regions. [Reuters on Southern/Dominion, Feb 2026]
Role 4: Data-centre owners/operators and “power access” winners
This is where AI capex meets real estate/infrastructure, but power access becomes the key differentiator.
United States
- Equinix (EQIX) – interconnection and premium colocation exposure
- Digital Realty (DLR) – hyperscale and colocation platform
- Iron Mountain (IRM) – data-centre growth alongside legacy business
- American Tower (AMT) – smaller relative exposure via edge/data-centre assets
Europe / UK
Pure public data-centre choices are fewer in Europe, so many investors use equipment + utilities + cables as cleaner expressions of the theme.
India
Direct listed pure-play exposure is limited, so the market often uses proxies:
- Adani Enterprises (ADANIENT) – data-centre and infrastructure platform exposure (through group initiatives)
- Reliance Industries (RELIANCE) – digital + infrastructure + AI ambition proxy
- Larsen & Toubro (LT) – AI factory / infra execution exposure
- Techno Electric & Engineering (TECHNOE) – more direct listed proxy among smaller names
What confirms this leg: project announcements turning into actual power allocations, utility agreements, and commissioning milestones rather than only MoUs/headlines. [Reuters; ET/industry interviews]
The alpha hooks: what to watch from here
This theme will be won on execution, not headlines. The best signals are simple and observable:
- Utility capex revisions linked to large-load demand (not just generic growth)
- Transformer/switchgear/cable lead times and order-book strength
- Interconnection and queue reform progress (PJM/FERC and local grid operators)
- Data-centre power allocation timelines (time-to-power becoming more important than land)
- Cooling and electrical system upgrades tied to high-density AI racks
- Local policy rules that shape who gets connected first and on what conditions (Europe/Ireland/India state-level execution)
[Reuters; PJM; FERC; CRU; Vertiv interview; Berkeley Lab]
Risks (so the thesis stays investable)
- Valuation risk: Many names in electrification, data-centre infra, and AI-linked industrials already re-rated.
- Timing risk: Projects can slip on permits, interconnection, or equipment delivery.
- Policy risk: Grid rules, tariffs, environmental approvals, and utility regulation can change returns.
- Demand overstatement risk: Some announced AI/data-centre plans may be delayed, resized, or relocated.
- Technology mix risk: Faster efficiency gains in chips and models may reduce power intensity per workload (even if total demand still rises).
[IEA scenarios; Reuters; NERC; PJM/FERC policy discussion]
Closing: the clean way to state the trade
This is not just an AI capex trade. It is a power bottleneck trade.
AI spending is pushing data centres into a new density regime. That is forcing a second wave of spending on electricity supply, grid upgrades, power delivery equipment, cooling systems, and faster connection pathways. The result is a broader investment universe than most AI narratives suggest.
The strongest setup over the next 1–3 years may be the middle of the chain: grid and electrical equipment, power-and-cooling infrastructure, and utilities with visible large-load demand, with selective exposure to data-centre platforms where power access is a real advantage.
Disclaimer: This article is for informational purposes only and does not constitute investment advice, a recommendation, or an offer to buy or sell any security. It is a thematic research note based on public sources and may be incomplete. Any examples of companies or tickers are illustrative only. Do your own research and consult a qualified financial adviser before making investment decisions.
