Data Centers Connect to the Grid Faster Through Flexible Power Consumption Agreements

Building a data center is expensive. Waiting for a grid connection is even more so. The interconnection queue can run 5–10 years. MIT Technology Review describes an approach that changes this logic: data centers agree to "flexible" consumption and get access to the grid much faster. For the AI industry, which is expanding capacity faster than ever, this could be the key to speeding up infrastructure deployment.

Kettles and overloaded grids

At the end of the first half of a goalless England–Germany match, millions of Britons switched on their electric kettles at the same time. The grid felt a sharp spike in demand — the classic "TV pickup" phenomenon that British energy engineers have known about for decades. Moments like this show how fragile the link between consumption and grid capacity can be.

Data centers create a similar challenge, but a constant one. A large facility consumes as much electricity as a small city. Building new transmission lines and substations takes years. Grid operators line up connection requests in queues — and as a result, finished buildings sit idle without power. In the UK, a number of new facilities waited more than seven years for a connection. In the US, the interconnection queue grew from 300 gigawatts in 2017 to more than 2500 gigawatts in 2024. Most of it consists of data centers and renewable energy projects.

What is a flex agreement?

Flex (short for flexible connection agreement) is a contract under which a data center gets immediate access to the grid, on one condition: at peak moments, at the operator's request, it cuts consumption to an agreed minimum.

The scheme works like this:

• The data center applies for a connection with a flexible load corridor
• The operator approves quickly: it does not need to build extra infrastructure for peak capacity
• When capacity runs short, the data center receives a signal and reduces its load
• Critical workloads keep running; non-critical ones are postponed
• The company receives compensation for its "flexibility" or a discounted tariff

Such agreements are already in place in the UK, Ireland, the Netherlands and Singapore. A number of large tech companies have moved to the flex model voluntarily — not because they are obliged to, but because otherwise they would have to wait.

Why this works for AI

Demand for compute has soared since 2022. OpenAI, Google, Microsoft, Amazon and dozens of startups are expanding infrastructure at unprecedented speed: in 2024 alone, new data centers with a combined capacity of over 50 gigawatts were announced. But power grids were built over decades — they are simply not ready for this pace.

The key observation: a significant share of AI workloads can be shifted in time. Training a 100-billion-parameter model can be started at 3 a.m. rather than at 6 p.m. Batch computing, data indexing and rendering can all be postponed by several hours without any loss in service quality.

"Load flexibility is not a constraint for a data center, it is a new asset,"

MIT Technology Review experts note.

This fundamentally changes the conversation with utilities. Instead of asking for a rigid maximum, the data center offers a flexible partnership. And grid operators value that more.

What this means

The race for AI infrastructure has run into a wall that is neither chips nor money — it is electricity. Flex agreements are one realistic way to synchronize the pace of data center construction with what the power system can deliver. Companies that build load management into their systems' architecture now will gain a competitive advantage: they will be able to deploy infrastructure where and when others are still standing in line.