The Power

Energy, Grid & Cooling

Layer 02 of seven · Energy, Grid & Cooling

Why it matters
A data centre without an energised connection is a shed. Power availability now shapes where AI capacity gets built and how quickly, and in several markets it has become the binding constraint on the whole build out.
The bottleneck
Grid connection queues and long lead time equipment. Transformers, switchgear and turbines are ordered years ahead, and no amount of capital shortens a queue that is physically constrained.
Who captures value?
Owners of existing energised sites and firm generation, and manufacturers of equipment with genuine lead time advantages. Commodity generation without a contract captures much less.
What could change?
Sustained efficiency improvements per unit of compute, political resistance to large industrial loads competing with households, or small modular reactors arriving sooner than expected.

AI's infrastructure problem is increasingly not just whether GPUs can be bought, but whether they can be powered and cooled. That runs from generation through the grid to the rack.

Compute is electricity in a more interesting shape. A large AI campus draws power at a scale previously associated with heavy industry, and it wants that power continuously. The constraint is rarely one thing. It might be generation, or a transmission connection, or a transformer with a multi-year lead time, or the ability to remove heat from a rack drawing more power than an entire row used to. Treating this as a story about power stations alone misses most of where the money and the delay actually are.

Nuclear / SMRBASELOADAlways on generation. Small modular reactors are being pitched as dedicated power plants sitting next to data centres.Gas and solarVARIABLECheap and fast to build, but output moves with weather and fuel prices. Useful, not sufficient on its own for a site that never sleeps.Fuel cellsON SITEBloom Energy style fuel cells generate power at the building itself, bypassing grid queues that can take years to clear.The gridTRANSMISSIONTransmission is now the bottleneck. Connecting a new gigawatt scale site to the grid often takes longer than building the site.Data centreGIGAWATT SCALE DEMANDA single large AI campus can draw as much electricity as a small city, constantly, all year.Compute is electricity in disguise

Tap or hover a box to see what it does

Generation feeds the grid, the grid feeds the data centre. On site fuel cells and small reactors sit alongside it for power the grid cannot guarantee.

Generation

Nuclear, gas, renewables, fuel cells and on site distributed generation. Buyers increasingly want power that is both firm and low carbon, which is a harder combination to source than either on its own.

publicCEG

Constellation Energy

What it does

Operates a large fleet of nuclear generation in the United States and sells power under long term contracts.

Why it matters

Nuclear output is carbon free and runs around the clock, which suits buyers who want firm supply and a clean energy claim in the same contract.

What could go wrong

Power purchase deals attract regulatory scrutiny over who pays for grid costs, and merchant power prices can move against the fleet.

Position in the ecosystem
publicVST

Vistra

What it does

Runs a mixed generation fleet including nuclear, gas and storage, and sells into wholesale power markets.

Why it matters

Existing sites with existing grid connections are more valuable than new projects that cannot get connected for years.

What could go wrong

Commodity exposure, plant outages, and the political sensitivity of large industrial loads competing with households for supply.

Position in the ecosystem
publicNEE

NextEra Energy

What it does

Develops and operates renewable generation and storage at scale alongside a regulated Florida utility.

Why it matters

Renewables are the fastest capacity to build in many markets, and pairing them with storage or gas is how a lot of new data centre load is actually being served.

What could go wrong

Interest rates matter enormously to project economics, and interconnection queues can delay revenue for years.

Position in the ecosystem
publicBE

Bloom Energy

What it does

Makes solid oxide fuel cells that generate electricity on site from natural gas, biogas or hydrogen.

Why it matters

On site generation can bring a site online while the grid connection is still queued, which is often the difference between building this year and building in three years.

What could go wrong

Fuel cost exposure, emissions questions when running on natural gas, and a history of lumpy order timing.

Position in the ecosystem
publicOKLO

Oklo

What it does

Designing small modular fast reactors intended to sell power directly to large users such as data centres.

Why it matters

If small reactors work commercially, they change the geography of compute by putting firm clean power next to the load.

What could go wrong

Pre-revenue, licensing dependent and years from meaningful deployment. This is a bet on a regulatory and engineering outcome, not on current cash flows.

Position in the ecosystem

Grid infrastructure

Transmission lines, substations, transformers and the connection agreements that determine whether a site can draw the power it has contracted for. This is where most of the waiting happens.

publicGEV

GE Vernova

What it does

Supplies gas turbines, grid equipment and wind generation, and services the installed base.

Why it matters

Turbine and grid equipment order books have lengthened considerably. When the constraint moves from chips to electrons, the people who build generating and switching equipment become part of the AI supply chain.

What could go wrong

Long lead time manufacturing is exposed to input costs and execution risk, and a slowdown in data centre commitments would show up in orders with a lag.

Position in the ecosystem
publicABBN.SW

ABB

What it does

Supplies electrification and automation equipment including transformers, switchgear and drives.

Why it matters

Transformers in particular have long lead times. A site without one is a building site, whatever else has been ordered.

What could go wrong

Cyclicality, and competition from lower cost equipment makers as supply catches up with demand.

Position in the ecosystem

Data centre electrical

Uninterruptible power supplies, switchgear, busway and backup systems sit between the substation and the rack. Availability is engineered here rather than assumed, and no design eliminates outage risk entirely.

publicETN

Eaton

What it does

Supplies electrical distribution equipment, switchgear, busway and uninterruptible power systems.

Why it matters

Between the substation and the rack sits a lot of unglamorous equipment. Lead times on that equipment have become a real constraint on how quickly capacity comes online.

What could go wrong

Order books reflect committed projects. If commitments soften, the backlog is a lagging comfort rather than a leading one.

Position in the ecosystem
publicSU.PA

Schneider Electric

What it does

Supplies power distribution, UPS, cooling and data centre management software, often as a packaged design.

Why it matters

Sells the electrical and thermal design as a system rather than as parts, which matters when rack densities keep rising faster than reference designs.

What could go wrong

Competitive pressure on packaged offerings and exposure to construction cycles well beyond data centres.

Position in the ecosystem

Cooling and heat rejection

Air cooling struggles as rack densities climb, so liquid cooling has moved from specialist to routine at the high end. Whatever the method, the heat still has to be rejected somewhere, which is why water and site selection keep appearing in the conversation.

publicVRT

Vertiv

What it does

Makes data centre power and thermal management equipment, including liquid cooling for high density racks.

Why it matters

Rack power densities have climbed to a point where air alone struggles. Liquid cooling has moved from specialist to mainstream, and Vertiv sells into that shift directly.

What could go wrong

Customer concentration among hyperscalers, and margin pressure as competitors move into liquid cooling.

Position in the ecosystem
publicMOD

Modine Manufacturing

What it does

Makes thermal management products including data centre cooling and heat rejection systems.

Why it matters

Heat has to go somewhere. Chillers, coolant distribution and heat rejection are the least discussed and most physically constraining part of a dense site.

What could go wrong

A large legacy vehicle thermal business dilutes the data centre exposure, and cooling is a competitive market.

Position in the ecosystem

How to think about the economics

  • GrowthVery high

    How quickly demand in this part of the ecosystem is expanding.

  • Capital intensityVery high

    How much money has to be spent up front before revenue arrives.

  • Competitive moatModerate

    How difficult it is for a credible new entrant to take the business.

  • Customer concentrationModerate

    How much revenue depends on a small number of buyers.

  • Disruption riskLow

    How exposed the layer is to a technical or commercial shift.

This is a framework for thinking about the economics of a layer, not a recommendation. An important AI company, a strategically advantaged company, an investable security and an attractively valued security are four different things.

Where value may accrue

Contracted cash flows, scarcity of energised land, and lead times. This is a layer where being early to secure a connection matters more than being clever, and where regulated returns cap the upside for some participants while protecting the downside.

Key terms in this layer

Fuel cell
A device that generates electricity from fuel through chemistry rather than combustion. Used on site at data centres to bypass years long grid connection queues.
ExampleBloom Energy installs fuel cells at the building itself.
Related termsThe gridSmall modular reactor (SMR)
The grid
The transmission network that moves electricity from power plants to users. For AI, connecting a new gigawatt scale site to the grid often takes longer than building the site.
ExampleUtilities quote multi year waits for new large connections.
Related termsFuel cellSmall modular reactor (SMR)Data centre
Small modular reactor (SMR)
A compact nuclear reactor designed to be factory built and installed near the demand. Pitched as dedicated power plants sitting next to data centres.
ExampleOklo is developing small reactors aimed at data centre power.
Related termsThe gridFuel cell
Liquid cooling
Removing heat by circulating fluid directly to or through the hardware rather than moving air. Once a rack draws enough power, air simply cannot carry the heat away fast enough.
ExampleVertiv and Modine sell liquid cooling systems for high density AI racks.
Related termsData centreThe grid

Related questions

Why does AI use so much energy and water?

Training a large model means running many thousands of accelerators at high utilisation for weeks. Those chips turn nearly all that electricity into heat, and the heat has to be removed, which is where cooling and in some designs water consumption come in. This is why energy, grid connections and cooling have become genuine parts of the AI investment conversation rather than background details. In several markets, securing power is now harder than securing chips.

Why are crypto miners relevant to AI?

Infrastructure overlap. Bitcoin mining, and historically Ethereum mining, required many of the same ingredients now sought by AI infrastructure developers: large power allocations, suitable sites and industrial cooling. Ethereum itself moved from proof of work to proof of stake in September 2022, so it is no longer mined. When mining economics weakened, several operators held energised sites that would take years to permit from scratch. Converting them for AI tenants is a substantial rebuild rather than a relabel, but the scarce ingredient, power, was already secured.

Sources and further reading (4)+
  1. 1Energy and AI, data centre electricity demand · International Energy Agency
  2. 2Powering intelligence: analysis of AI and data centre electricity use · EPRI
  3. 3Data centre cooling and density research · Uptime Institute
  4. 4Investor materials on data centre power and cooling · Vertiv
Last fact-checked: 29 August 2026

The AI ecosystem changes rapidly. Company positions, technologies and market data reflect information available at the date above.