Behind the Meter: Why AI Data Centres Are Becoming Power Projects

For the last two years, AI infrastructure has been framed as a race for chips, then a race for power. Both remain true, but the next constraint is now becoming more specific: time to power.

The issue is not simply whether enough electricity exists somewhere in the system. The issue is whether a specific site can be energised at the scale, reliability and timeline required by AI workloads.

That is why behind the meter generation is moving from the edge of the data centre conversation to the centre of it.

For most of the cloud era, the standard assumption was simple. Secure the land, secure the grid connection, build the facility, install backup generation, then scale. That model is now under pressure. AI campuses are larger, denser and more urgent than the infrastructure planning cycles around them. A data centre can often be designed and built faster than the grid can be reinforced, and that mismatch is reshaping how projects are being delivered.

But there is a more positive way to look at this.

Behind the meter generation can be more than a workaround for a slow grid connection. Done properly, it can bring new firm power to a site, reduce pressure on local electricity infrastructure, accelerate investment, create jobs and give communities a stronger basis for accepting major digital infrastructure.

That last point is critical. Communities are right to ask what they receive in return for hosting large infrastructure. If a data centre places new pressure on local roads, land, planning systems, water resources and power networks, the community benefit has to be real. It should not be a vague promise, a voluntary donation or a line in a planning statement. It should be contracted, financial and enforceable.

The future of AI data centre deployment will not be won by the projects that simply demand grid capacity. It will be won by the projects that bring power, investment and local benefit with them.

The grid clock and the AI clock are moving at different speeds

The International Energy Agency estimates that data centres consumed around 415 TWh of electricity in 2024, around 1.5% of global electricity consumption, with demand projected to reach around 945 TWh by 2030. It also notes that while data centres can be operational in two to three years, the wider energy system often requires longer lead times for planning, infrastructure build and investment.

That timing gap is now one of the biggest constraints in the market.

Grid infrastructure is not designed to move at software speed. New substations, transmission upgrades, transformers, wayleaves, connection studies, planning processes and reinforcement works take time. In some locations, the grid is not just delayed, it is already heavily committed.

For communities, this creates a legitimate concern. If a large data centre connects to an already constrained local network, will households, businesses, housing schemes and local employers be pushed further back in the queue. Will the area be asked to host a major project while the wider electricity system struggles to keep pace. Will grid reinforcement costs ultimately feed through to bill payers.

Behind the meter generation changes that discussion. Instead of arriving as a pure load, the campus can arrive with part of its own energy system. It can bring firm generation, battery storage, grid support capability and long term investment in local infrastructure. It can reduce its dependence on scarce local import capacity during peak periods. It can be designed to work with the grid rather than simply wait for it.

This is the new delivery reality. The commercial clock of AI is running faster than the grid clock, so the most credible projects will bring power strategy to the site, not treat it as someone else’s problem.

Behind the meter is not just backup

Behind the meter power used to be mostly a resilience conversation. Diesel generators, UPS systems and backup arrangements were there for outages, testing and business continuity. They were essential, but they were not the primary route to energisation.

Behind the meter now means something much broader. It means power generated, stored or managed on the customer side of the utility meter. It can include gas engines, gas turbines, fuel cells, solar, wind, batteries, private wire arrangements, microgrid controls, heat recovery and flexible load management. In some cases it may be temporary bridge power while the grid connection catches up. In others it may become a permanent part of the campus operating model.

The point is not to abandon the grid. The grid still matters enormously, it provides resilience, balancing, import and export capability, access to decarbonising system power and long term integration with the wider energy system.

But the grid alone is no longer always enough to get large AI campuses live on the timelines the market now wants.

JLL’s 2026 Global Data Centre Outlook says operators are expected to increase behind the meter power arrangements and explore colocated battery storage, as average grid connection wait times in primary markets now exceed four years. It also notes that natural gas is becoming more prominent in the US for bridge power and permanent on site generation, while renewables and private wire arrangements are more prominent in EMEA.

The data centre is no longer just a building that consumes power, it is becoming a power project with compute attached.

Gas generation can be a positive local infrastructure choice

Gas is often viewed as a carbon problem. That is understandable, because gas combustion creates emissions and those emissions have to be addressed. But in the context of constrained grids and delayed connections, behind the meter gas generation can also solve an important local problem. It can bring firm, controllable, additional power to the site.

That matters because communities are increasingly being asked to host data centres in places where the grid is already under pressure. A project that simply takes capacity from the local network can look like a burden. A project that brings its own firm generation, pays for its own infrastructure and supports the grid at constrained times can look very different.

Gas generation is not intermittent. It can run when required. It can support phased energisation. It can be scaled in modules. It can sit alongside batteries. It can reduce reliance on diesel backup. It can provide a bridge while the wider grid catches up. In some locations, it can help make a project deliverable years earlier than a grid only route.

That earlier delivery has consequences beyond the data centre operator. It can bring construction jobs forward. It can bring permanent operational roles forward. It can support local contractors, civil engineering firms, electricians, security providers, facilities management companies and technical maintenance suppliers. It can increase business rates or local tax receipts. It can justify new training programmes and apprenticeships. It can help anchor wider digital infrastructure, enterprise zones and industrial regeneration.

A powered AI campus is not just a server hall, it is a major infrastructure investment. If it brings its own power, it may also reduce the risk that local communities are asked to wait for benefits while the project waits for the grid.

That is the positive case for behind the meter gas. It can turn a data centre from a passive demand problem into an active delivery project.

The old model is too linear

The old model treated power as a dependency. The new model treats power as part of the product.

That changes the development sequence. Site selection can no longer start with land and fibre, then ask the energy question later. At 100MW, 300MW or 1GW scale, the energy question is the project.

A credible AI campus now has to answer a much harder set of questions from day one. Where will the power come from? How quickly can it be delivered? What is firm, what is intermittent, what is imported, what is generated locally, what is stored, what is flexible, what is reported, and what happens when the local grid is constrained?

This is why the phrase “powered land” is becoming so important. Land without a credible energisation path is not really development land for AI, it is a speculative option on future infrastructure.

In the next phase of the market, a smaller amount of power that can be delivered quickly may be more valuable than a larger theoretical allocation that sits behind years of reinforcement. The market is moving from announced capacity to energised capacity.

Behind the meter gas generation fits that shift because it is firm, financeable, technically understood and deliverable on a shorter timeline than many large grid reinforcements. That does not make it perfect, but it does make it practical.

The AI campus power stack

The serious projects will not think in binary terms. It is not grid or generation. It is not renewables or gas. It is not batteries or backup.

The better question is: what power stack gets this campus live fastest, most credibly and with the greatest local benefit?

That stack will usually include several layers:

• The grid remains the foundation where it is available. It gives the site access to system power, import capacity, backup options and future decarbonisation.
• Behind the meter gas generation provides firm controllable capacity. It can reduce dependence on constrained import capacity, support early energisation and help the project operate reliably while grid infrastructure catches up.
• Battery storage becomes the control layer. It can smooth rapid swings in AI load, reduce peak import, improve power quality, support ride through and allow the site to behave more intelligently during constrained periods.
• Renewables reduce emissions intensity and provide price hedging, especially where they are genuinely additional and physically linked through private wire or clearly contracted new supply.
• Microgrid controls orchestrate the whole system. They decide when to import, generate, store, discharge, curtail or export.
• Flexible workloads create another lever. Not every AI task has the same urgency. Training, batch inference, model optimisation and some enterprise workloads can be shifted more easily than latency sensitive inference.
• Transparent reporting is the final layer. Without it, communities and regulators will not trust the claim that the campus is helping rather than simply consuming.

This is what serious behind the meter strategy looks like. Not a generator hidden behind a fence, but an integrated energy architecture.

The community benefit has to be contractual

If behind the meter gas generation is going to be presented as a genuine benefit to the local community. That benefit cannot be soft, it has to be contracted.

A major AI campus with its own gas generation should be prepared to make binding commitments in return for planning support and local acceptance. Those commitments should be financial, measurable and enforceable.

That could include an annual community benefit payment linked to installed power capacity. It could include a local infrastructure fund for roads, skills, green space, broadband, community buildings or energy efficiency upgrades. It could include guaranteed apprenticeship numbers, local procurement targets and funding for further education partnerships. It could include discounted or free waste heat for local heat networks where technically viable. It could include funding for local grid resilience, emergency power support or community energy projects.

The principle is simple. If a data centre benefits from local consent, local land and local infrastructure, the local area should share directly in the value created.

This is especially important for gas backed projects. Communities will reasonably ask why they should accept a power plant attached to a data centre. The answer cannot be only that the project is good for national digital infrastructure. It also has to be good for the place that hosts it.

A credible community benefits package should therefore be built into the project from the start, not added late in the planning process. It should be legally documented. It should have clear payment triggers. It should survive ownership changes. It should be independently monitored. It should be easy for local people to understand.

The better projects will treat this as part of the capital stack, not as public relations.

Gas is the near term reality, and it needs discipline

For large AI campuses that need firm power quickly, gas is often the most realistic near term option. It is dispatchable, scalable, familiar to power engineers and available on a timeline that nuclear, large transmission upgrades and many renewable projects cannot match.

Pretending otherwise is not useful. Many AI projects looking at behind the meter power are looking at gas because they have run into the reality of grid queues, transformer lead times, permitting delays and the operational needs of high density compute.

The IEA says data centre developers are advancing a large number of projects with onsite natural gas based power generation, largely in the United States. It also warns that many of these projects remain at an early stage and face technical and financial hurdles. One issue is that AI data centres can have rapid and large swings in demand, which can stretch the technical capability of onsite gas plants, making onsite battery storage increasingly important.

Gas can solve the time to power problem. It can also reduce strain on local electricity networks if it is genuinely additional and properly integrated. But it needs discipline.

The plant should be high efficiency. It should be designed to minimise local pollutants. It should sit alongside batteries, not operate as a blunt standalone solution. It should reduce diesel dependence. It should have transparent emissions reporting. It should be compatible with a longer term decarbonisation pathway. It should not be sold as temporary unless there is a real plan for what replaces it.

Carbon capture may help in specific locations, especially where there is existing or planned CO₂ transport and storage infrastructure. But it should not be treated as a universal fix. A project that says “gas now, carbon capture later” without a funded route to capture, transport, storage, measurement and long term liability has not solved the problem. It has deferred it.

The stronger position is more honest. Gas can be a good thing where it brings new firm power, protects local grid capacity, accelerates job creating investment and is tied to a contractual community benefit package. It is not a free pass, it is a practical infrastructure choice that must be earned through delivery, transparency and local value.

Batteries make gas more community compatible

Batteries do not replace generation. They store energy, they do not create it.

But batteries are central to making behind the meter gas generation more useful and more acceptable.

AI workloads can move quickly. Demand can ramp and swing in ways that are awkward for both grids and gas generators. Batteries provide the fast response layer that makes the wider system more manageable.

Reuters reported that behind the meter batteries can help data centres manage demand spikes, reduce consumption when the grid is strained, cover temporary outages and reduce reliance on backup diesel generators. It also cited experts saying that while data centres can be built in 18 to 24 months, grid connection can take three to seven years in parts of the US.

When paired with gas generation, batteries can reduce ramping, improve efficiency, smooth campus demand and support flexible grid interaction. They can help the site reduce import during constrained periods. They can provide short duration resilience without immediately starting backup engines. They can also create optionality for future low carbon supply.

This is why batteries should not be treated as a sustainability add on. They are part of the core delivery architecture.

For AI campuses, the battery is becoming less like an accessory and more like the control system between compute, generation and the grid.

Reducing local strain should be a planning argument

The public conversation around data centres often starts from scarcity. Scarce power. Scarce land. Scarce water. Scarce grid capacity.

Behind the meter gas generation gives developers a better answer on power.

The argument should not be that the data centre is avoiding the grid entirely. Very few large campuses will or should do that. The stronger argument is that the campus is reducing its claim on constrained local import capacity, especially during peak periods, by bringing new firm capacity to the site.

That can matter for local communities. If a campus can energise without waiting for the same reinforcement needed by homes, hospitals, schools, housing developments and local employers, the project is easier to justify. If the campus can fund its own connection assets and related infrastructure, the case improves further. If the site can also provide grid support services, battery response or emergency resilience under agreed conditions, the community value becomes more tangible.

This should become part of the planning narrative. A data centre should be able to show how much local grid capacity it would otherwise require, how much behind the meter generation reduces that requirement, how storage reduces peak pressure, how the project will avoid worsening local constraints, and what financial benefits will flow back to the community.

That is a stronger argument than saying the project is important for AI. It shows how the project fits into the local energy system.

The regional picture is different

The US, UK and Europe are all facing the same broad issue, but the answer will not look the same in each market.

In the US, behind the meter gas generation is moving fastest because the ingredients are more available. There is more gas, more land, more regional variation, more private power procurement and more urgency in markets where the grid is struggling to absorb new AI load. The US is therefore likely to see more gas led bridge power, more permanent onsite generation and more battery backed campus microgrids.

The opportunity in the US is significant. Gas backed campuses can bring new firm power to constrained locations, accelerate major investment and reduce dependence on overstretched interconnection queues. But the social licence requirement is just as significant. If communities believe AI campuses are bringing emissions without meaningful local value, they will resist. The projects that move fastest will be the ones that pair power self sufficiency with local financial benefit.

In the UK, the issue is more constrained. The grid queue is heavily oversubscribed, and data centres are now a central part of the policy debate. Ofgem has identified around 140 data centre projects in the demand queue, representing around 50GW, against GB peak electricity demand of 45GW on 11 February 2026. It also expects a significant number of projects in the demand queue to be non viable.

The UK answer is therefore unlikely to be full independence from the grid. It is more likely to be grid backed self supply, better site selection, phased connections, battery storage, flexible operation, private wire renewables where possible and strategic alignment with AI Growth Zones. The UK Government is already consulting on reforms to address speculative demand, reserve or reallocate capacity for strategic projects, and enable viable options for AI Growth Zone developers to build their own high voltage grid infrastructure.

Behind the meter gas generation could have a role in the UK, but only if it is framed properly. It should be presented as additional firm capacity that reduces local grid strain, supports strategic infrastructure and brings contracted local benefit. A project that simply adds emissions without clear community value will struggle.

In Europe, the answer is more fragmented. France, the Nordics, Spain, Germany, Ireland and the Netherlands each have different energy systems, planning constraints and public attitudes. Nuclear, hydro, renewables, private wire, industrial site reuse and heat recovery will matter more in some markets than others.

Europe is also moving towards greater scrutiny. The European Commission is preparing a Data Centre Energy Efficiency Package, including assessment of reporting data, a rating scheme for data centres and work on minimum performance standards.

That means behind the meter power in Europe will need to be especially well justified. It will have to show additionality, efficiency, emissions discipline, grid benefit and contracted community return.

The social licence test is also a financial test

Behind the meter generation can make a project more credible, but only if it is done properly.

It helps when the project brings new capacity rather than simply consuming scarce local grid capacity. It helps when storage reduces peak pressure. It helps when the campus can support the local network during constrained periods. It helps when diesel reliance is reduced. It helps when the developer pays a fair share of reinforcement costs. It helps when emissions, water use, grid imports and community benefits are reported clearly.

It also helps when the community benefit is written into the economics of the project.

A voluntary community fund is better than nothing, but it is not enough for infrastructure of this scale. The better model is contracted community benefits linked to project size, energisation milestones and operating life.

For example, a data centre campus could commit to a fixed annual community payment per MW of installed generation or IT load. It could make additional payments when new phases are energised. It could fund local grid upgrades, training centres or energy efficiency schemes. It could support local apprenticeships and guarantee procurement opportunities for regional suppliers. It could provide ring fenced funding for community assets, local transport or green infrastructure.

The structure matters because communities need certainty. They should not have to rely on goodwill after consent has been granted.

If behind the meter gas generation is being used to accelerate a project, then some of the value created by that acceleration should be shared locally. Faster energisation creates commercial value for the operator. A fair portion of that value should be contracted back into the place that hosts the infrastructure.

From passive load to active energy asset

The strongest argument for behind the meter gas generation is not that it allows data centres to bypass the system, it is that it allows them to take more responsibility for the system impact they create.

A passive load asks the grid to solve the problem. An active energy asset brings generation, storage, controls, flexibility and money to the table.

That is the future model for AI infrastructure. The campus should not simply ask how much power it can take. It should ask what power it can bring, what strain it can reduce, what infrastructure it can fund and what benefits it can contract locally.

This is where behind the meter generation can play a constructive role. It is firm. It is deliverable. It is scalable. It can be paired with batteries. It can support phased deployment. It can reduce pressure on local grid capacity. It can make job creating projects happen sooner. It can give communities a direct financial stake in hosting infrastructure.

But it has to be done openly. The emissions have to be counted. The local impacts have to be mitigated. The benefits have to be contractual. The long term transition pathway has to be real.

The winners will build power projects, not just data centres

The next generation of AI infrastructure will not be won only by the companies that can buy GPUs or announce the largest campuses. It will be won by the companies that can assemble credible energy systems around compute demand.

That means power ready land. It means grid relationships. It means behind the meter generation. It means batteries. It means cooling and water strategy. It means flexible workloads. It means planning credibility. It means contracted community benefit. It means a decarbonisation pathway that survives contact with public scrutiny.

The data centre sector is entering a new phase. The old model was a building connected to the grid. The new model is a campus integrated with an energy system.

Behind the meter gas generation is becoming a major part of accelerating deployment because it gives developers more control over time to power. It can also be a positive local infrastructure choice where it brings new firm capacity, reduces pressure on the local grid and unlocks investment that might otherwise be delayed for years.

But the bargain has to be clear. Communities should not be asked to accept major infrastructure on trust. They should receive binding financial benefit, local jobs, local procurement, infrastructure investment and transparent reporting.

Used badly, behind the meter gas will intensify the backlash. Used well, it can help turn data centres from passive loads into active energy projects.

The future belongs to projects that can prove they are not just taking power from the system, but helping to build, finance, store, shape and optimise the power system around them, while sharing the value with the communities that make deployment possible.