Buying Power Isn't The Same As Creating Power For AI Data Centres

When a technology company announces that it's secured hundreds of megawatts of electricity for its AI data centres, it's easy to assume that another major infrastructure problem has been solved. After all, the company has found a power supplier, signed an agreement and secured what sounds like an enormous amount of electricity. Surely that means it's ready to start building?

Well, not necessarily. AI data centre power agreements are becoming more complicated, and the numbers attached to them don't always mean what we might think. Buying electricity from an existing power station, paying to increase its output and building a data centre beside it are three very different arrangements.

Google's latest agreement with Constellation Energy is a good example. Announced in October 2026, it includes plans to increase nuclear power generation alongside a separate agreement for electricity from existing facilities. Both involve substantial commitments, but only one is intended to create additional generating capacity.

And for anyone planning, building or relying on AI infrastructure, understanding the difference is becoming just as important as knowing how much electricity a facility needs.

What Does Securing Power Actually Mean?

When a company says it's secured power for a data centre, it usually means that it's made arrangements to obtain electricity. That might involve signing a contract with an energy supplier, investing in a new power station or arranging to receive electricity directly from an existing one.

But there's an important distinction between having an agreement to buy electricity and having electricity that's physically available to use. A contract can promise future supply without creating any new generating capacity. Even when additional generation is planned, the equipment still needs to be built, connected and brought into operation before it can produce anything.

It also helps to understand what the numbers in these announcements are measuring. Megawatts (MW) describe how much power something can produce or consume at a particular moment, while megawatt-hours (MWh) measure how much electricity is produced or used over time. For example, a power station generating 100 MW continuously for one hour would produce 100 MWh of electricity.

Data centre capacity adds another complication. The advertised figure often refers to the maximum power needed by its IT equipment, rather than the entire facility. Servers aren't the only things using electricity. Cooling, lighting and other supporting systems need power too, while actual consumption changes as equipment is installed and workloads increase.

The Electric Power Research Institute (EPRI) explains these differences in its February 2026 Powering Intelligence research, which separates advertised IT capacity, total facility requirements and actual electricity consumption.

So a company announcing a 500 MW data centre isn't necessarily saying it has 500 MW of electricity available today, or that its facility will continuously consume that amount. To understand what the announcement means, we need to look at how the electricity is being secured.

How AI Data Centres Secure Electricity

There are several ways developers can arrange electricity for AI infrastructure, and they aren't mutually exclusive. A company might buy power from an existing supplier while investing in additional generation for future expansion. Another might connect directly to a power station but retain a grid connection for backup.

The arrangements themselves aren't especially new. What's changing is how they're being used to support increasingly large computing facilities, and how easily their differences can get lost behind a headline about another enormous power deal.

Buying electricity from existing generation

The most familiar approach is to buy electricity through a power purchase agreement (PPA). This is a contract between an electricity buyer and a supplier, usually setting out how much electricity will be purchased, the price and the length of the agreement.

A PPA can provide some certainty about future electricity costs and support a power station's continued operation. However, if that station already exists and its generating capacity hasn't changed, signing another agreement doesn't automatically mean there's more electricity available across the network.

There are also physical and virtual PPAs, which work differently. A physical PPA involves arrangements for supplying electricity to the buyer, either directly or through the relevant electricity network. A virtual PPA is mainly a financial agreement, so the buyer can support a generating project without receiving its electricity directly.

Neither label tells us whether the contract is supporting new or existing generation. That's something we need to establish separately.

Supporting new or expanded generation

Rather than simply buying electricity that's already being produced, developers can help finance new power stations or upgrades that increase the output of existing facilities. Long-term purchasing agreements can make these investments easier to justify because suppliers have greater certainty that someone will buy the electricity they produce.

This is where the Google and Constellation agreement becomes particularly interesting. The companies announced a 20-year arrangement supporting more than $4.3 billion in upgrades across 11 existing nuclear units in the United States. These improvements, known as nuclear uprates, are expected to increase the stations' combined generating capacity by 890 MW, with the first upgrade planned for 2028.

The same announcement includes a separate 15-year agreement covering 2,700 MW from Constellation's existing generating fleet. That second arrangement is intended to support the continued operation of those facilities, rather than increase their generating capacity.

Both agreements have a purpose, but their contribution to electricity supply is different. One is intended to produce additional power, while the other supports generation that's already available.

Of course, even an agreement to create new capacity doesn't mean the electricity is available immediately. Construction, equipment upgrades, approvals and grid connections still have to be completed before the promised supply becomes operational.

Connecting directly to power generation

Another approach is to put the data centre beside a power station so it can receive electricity directly from the generating facility. This is often called a behind-the-meter arrangement, because some or all of the electricity reaches the data centre without passing through the usual grid supply connection.

Japan's JERA is exploring this approach through a proposed AI infrastructure development at its Chiba thermal power station. Announced with Dell Technologies and RHAELM in October 2026, the project would support up to 400 MW of computing infrastructure using power from JERA's operating generation site.

The companies believe this could allow the development to begin operating years earlier than a conventional grid-connected facility, although operations aren't expected to start until around 2028. The project is still being developed, so those advantages remain proposed rather than demonstrated.

Co-location can reduce some of the difficulties involved in connecting a large new electricity consumer to the grid. But using power from an existing station doesn't automatically increase its output, and a direct connection doesn't necessarily eliminate the need for wider electricity network services.

Why The Type Of Power Agreement Changes Infrastructure Planning

It's one thing to have a commercial agreement for electricity. It's another to have a facility that can actually use that electricity whenever it needs it. For infrastructure leaders, the difference affects everything from when a project can launch to how reliably it can operate once the servers are running.

Time to power and grid access

Time to power is essentially how long it takes to get enough electricity to a facility so it can begin operating. Signing a supply agreement might be one step in that process, but it doesn't mean the connection, generating equipment or supporting infrastructure is ready.

Lawrence Berkeley National Laboratory examined these challenges in its June 2026 Speed to Power report. The researchers identified more than 40 potential ways to accelerate large electricity connections, covering everything from connection processes and electricity planning to procurement and grid operations.

This helps explain why different power arrangements can have such different timelines. Connecting to an operating power station might avoid some conventional grid connection delays, while expanding an existing facility may be quicker than building an entirely new one. But both still depend on the specific equipment, approvals and construction work required.

For anyone planning an AI deployment, the important date isn't necessarily when the electricity agreement was signed. It's when enough power will actually be available to run the infrastructure.

Reliability and grid dependency

Electricity availability also isn't quite the same as electricity reliability. A power station might have enough generating capacity to support a data centre, but that doesn't guarantee it can supply the full amount every hour of the year.

Equipment needs maintenance, unexpected failures happen and some energy sources can't generate electricity continuously. Even dedicated generation may need backup equipment or access to the wider grid to keep critical systems operating.

The International Energy Agency examined this in its April 2026 Key Questions on Energy and AI report. Its analysis found that reliably meeting critical, variable data-centre demand using onsite gas generation could require generating infrastructure sized 30 to 70 per cent above the demand being served.

Directly connecting a data centre to a power station doesn't automatically solve these problems either. In December 2025, the US Federal Energy Regulatory Commission addressed how large facilities connected beside generators should use transmission services from the PJM electricity network. The decision reflects the fact that these facilities can still depend on the grid, even when much of their electricity comes from nearby generation.

Emissions and infrastructure costs

There's also the question of what kind of electricity an agreement provides. Buying electricity associated with renewable or nuclear generation can support a company's emissions goals, but it doesn't necessarily mean that clean electricity is physically supplying its data centre at every moment.

The GHG Protocol recognises this distinction through different methods of reporting electricity-related emissions. One considers the electricity grid serving a facility, while another accounts for qualifying electricity purchasing arrangements. Both provide information, but they don't describe exactly the same thing.

And then there's the cost of making the arrangement work. A developer buying existing electricity may face different obligations from one financing a new generating facility or connecting directly to a power station. Someone still needs to pay for equipment, connections, backup services and any additional infrastructure required.

These aren't reasons to favour one model over another. They're reasons to examine the physical and commercial arrangements together, rather than assuming that securing electricity means every infrastructure problem has been addressed.

How To Evaluate An AI Data Centre Power Agreement

So what should you actually look for when a data-centre provider announces another major electricity deal? Whether you're evaluating a supplier, planning an AI deployment or considering where to host your infrastructure, the starting point is understanding what the agreement will deliver.

There are five questions worth asking before treating any power commitment as evidence that a project is ready to proceed:

1. Is the agreement buying existing electricity or creating additional supply?

Find out whether the power comes from facilities that are already operating, new generation that's being developed or upgrades intended to increase output. If the electricity already exists, consider whether the agreement changes how that capacity is allocated.

2. When will the electricity actually be available? 

There's a difference between signing a contract, announcing a development and having generating equipment that's ready to operate. Check which parts of the agreement are already delivering electricity and which still depend on construction or approvals.

3. How will the electricity reach the data centre? 

Establish whether the facility will use a conventional grid connection, receive power directly from a generating site or rely on both. The arrangement should make clear which connections and network services are still needed.

4. What happens if the primary supply becomes unavailable? 

Ask about maintenance, backup generation, grid support and the amount of power the facility can reliably access. Maximum generating capacity isn't the same as a guarantee of continuous electricity supply.

5. Who is responsible for the infrastructure that remains? 

Identify which parties must deliver generating equipment, connection upgrades and backup arrangements. These responsibilities can affect project costs, timelines and the consequences of delays.

The answers can reveal quite different levels of infrastructure readiness, even when two announcements promise similar amounts of electricity. They also help enterprise buyers assess a provider's claims without needing to become electricity market specialists. 

What they're really trying to establish is whether the promised computing capacity has a dependable route to becoming operational, and what could still prevent that from happening.

Final Thoughts: Secured Power Isn't Always Available Power

The growing number of AI electricity agreements makes it tempting to compare companies by how many megawatts they've managed to secure. But those figures don't tell us much on their own. 

A contract supporting an existing power station, an investment in additional generation and a direct connection to a generating facility can all produce impressive announcements without delivering the same result. For infrastructure leaders, the more useful question is what each agreement actually changes. 

Does it increase the electricity available, provide a more direct way to access existing supply or simply establish a commitment that still depends on future development? Understanding those differences makes it easier to judge whether a proposed AI facility has a realistic path to operation, rather than relying on the scale of its announced power commitments.

As electricity becomes more closely tied to AI infrastructure planning, EM360Tech will continue examining the decisions and dependencies shaping how these projects move from proposal to production.