Compute where the power already is

In 2025, France produced more solar and wind electricity than its market wanted during some hours of the year.

Wholesale prices fell below zero. Generators capable of responding reduced their output. The French transmission operator RTE estimates that around 3 TWh of wind and solar generation was not produced during these periods, almost twice the volume recorded a year earlier.

It is tempting to describe all of that electricity as renewable power thrown away because the grid could not carry it.

That would be a better story than an explanation.

Most of the 3 TWh reflected generators responding to market prices. Only around 0.1 TWh of wind and solar output was reduced at RTE’s direct request to maintain system balance. Physical congestion, economic modulation, and operational curtailment can all reduce generation, but they do not describe the same event. RTE’s 2025 electricity review makes those distinctions unusually clear.

At the same time, data-center developers are struggling to secure electricity connections.

The International Energy Agency reported that global data-center electricity demand rose by 17% in 2025, much faster than total electricity demand. It expects consumption to roughly double by 2030, with AI-focused facilities growing faster still. Supply-chain constraints, slow planning processes, and overloaded connection queues are already delaying projects.

Europe can therefore have too much electricity during one hour and too little usable capacity for a new data center during another.

The issue is not simply how much electricity exists. It is whether power is available in the right place, at the right time, through infrastructure capable of delivering it.

Power can be abundant and unavailable at the same time

Electricity does not behave like a stockpile that can be produced anywhere and saved indefinitely for whoever needs it later.

Supply and demand have to remain balanced continuously. Transmission lines have physical limits. A region can produce more electricity than local demand can absorb while another region faces scarcity or a long connection queue.

Stronger grids help bridge that distance, but Europe is struggling to build and operate them quickly enough.

In 2024, European transmission operators used remedial actions involving 60 TWh to manage grid congestion, at a cost of €4.3 billion. That does not mean 60 TWh was wasted. It means operators intervened in generation and network flows because the power system could not carry electricity exactly as the market had scheduled it. ACER’s congestion report says delayed reinforcement is widening the gap between the network Europe has and the one its changing power system needs.

Data centers intensify the local problem because they create large, concentrated loads. Their projected global share of electricity use remains modest beside the whole power system, but a global percentage hides what several hundred megawatts requested from the same substations can mean for one region.

The IEA estimates that grid constraints could delay around 20% of the data-center capacity planned worldwide through 2030. Among its recommendations is a deceptively simple one: place more facilities where generation and grid capacity are already available.

Europe still needs more transmission, distribution capacity, transformers, interconnectors, storage, and generation. Changing where data centers are built cannot replace any of that.

It can reduce the habit of creating new loads first and asking the grid to solve their geography afterward.

A data center is not one unchangeable load

“Data-center power demand” makes a facility sound like one machine drawing a fixed amount of electricity. The reality is more complicated.

A data center contains computing equipment, storage, networking, cooling, power conversion, and other supporting systems. Its draw changes with installed hardware, utilization, workload, climate, and operating model. A headline megawatt figure can describe several different things, from grid connection capacity to actual annual use.

The work inside matters too. Some computing demand is difficult to move. A real-time service may have strict latency requirements. A transaction system may need to remain close to its users and data. A long-running job may be technically interruptible but too expensive or risky to pause.

Other work has more room to move. Rendering, model evaluation, data processing, simulation, backups, and some machine-learning jobs may tolerate another location, a later start, or a temporary reduction in power. How much flexibility exists depends on the application, service commitment, data location, architecture, and cost of leaving expensive hardware idle.

In March 2026, Google said it had incorporated 1 GW of data-center demand response into agreements with several U.S. utilities. This is first-party program evidence, not a claim that every workload or facility is flexible. The arrangements cover selected machine-learning demand that Google says can be limited or shifted during constrained periods. Google describes the contracts here.

European transmission operators are examining the same possibility. In May 2026, ENTSO-E described data centers as systemically relevant electricity users and published work on their effect on grid planning and possible participation in flexibility services. The important word is possible. Flexibility has to be designed, contracted, measured, and permitted by the workloads involved. ENTSO-E presents it as an opportunity to develop, not a capability the sector can take for granted.

Demand response has its own operational bargain: which workloads can move, what operators give up when they reduce load, and who carries the cost and risk.

Computing demand has two characteristics many industrial loads do not share to the same degree. Some work can move through time. Some can move through networks.

That gives data-center planners another variable besides waiting for more electricity to arrive at a fixed location.

Moving compute does not make the grid disappear

The phrase “move the load to the power” sounds simpler than the infrastructure behind it.

A data center beside a wind farm does not automatically run on that wind farm. It may be connected behind the same meter, through a private wire, or through the public grid under a contractual renewable-energy arrangement. Those structures have different physical, commercial, and regulatory meanings.

Proximity alone proves little. Nor does it show that the facility consumes electricity that would otherwise have been curtailed. That conclusion depends on the electrical connection, metering, market conditions, operating schedule, and what the generator would have done without the data center.

Without those details, “turning curtailed energy into compute” remains a plausible mechanism rather than a measured result. Curtailment itself has several causes, each demanding a different response.

Reliability adds another limit. Customers generally expect services to remain available when local renewable production falls. A site built around variable generation still needs an operating model for those hours. That may involve the public grid, storage, another source, workload movement, reduced demand, or an agreed combination.

Networking can become the next constraint. Electricity may be plentiful at a remote energy site while fiber is inadequate. Moving large datasets takes time and money. Latency-sensitive services may need to stay closer to users. Data-protection rules or contracts may limit where information can be processed.

Climate can become another power constraint because cooling uses electricity. A site with attractive energy availability but difficult heat-rejection conditions may produce a less attractive overall system. Cooling connects the workload to the climate.

The best data-center location is not simply the place with the cheapest electricity or the nearest generator. Power has to fit with networking, workload, cooling, land, regulation, ownership, and operations. Those conditions are brought together in data-center site selection.

Make location part of the infrastructure design

Traditional data-center regions exist for good reasons. They have extensive fiber, experienced operators, established suppliers, large customer bases, and utility infrastructure built around years of demand.

Concentration also creates its own constraint. When thousands of megawatts chase the same substations and network corridors, the place that once offered the best infrastructure can become difficult because so many facilities want it.

Modular infrastructure widens the range of sites worth considering without making location irrelevant.

Policloud develops and deploys physical, modular data-center infrastructure for defined sites. For an energy producer or industrial host, that creates a practical question: could the available power support useful compute and storage infrastructure here?

The word could carries a lot of work. A site with electricity still needs land, fiber, permits, a viable configuration, an operating model, and real workload demand. The economics and allocation of risk cannot be inferred from the presence of a transformer.

The physical infrastructure is one layer. Where a deployment uses Hivenet, Hivenet provides the software mesh and distributed services layer used by the deployment. Policloud remains the physical infrastructure; Hivenet provides the software and service layer when included.

Antimatter connects the wider energy, hardware, and software work. That relationship does not guarantee that every unit joins a shared network, that workloads will move between sites, or that an energy producer will achieve a particular level of utilization.

What the structure provides is the ability to ask a more useful question before construction begins: where can this workload run within its actual constraints?

The grid should not have to solve every location decision

Return to the French power system in 2025.

During some hours, market conditions encouraged wind and solar generators to reduce output. During others, network operators intervened to maintain balance. Meanwhile, new electrical loads were seeking connections that could take years to deliver.

Those conditions cannot be matched with a line on a diagram. The generator may be in the wrong region. The data center may need firmer power than the site can provide. The workload may be tied to another location. The network may be missing. The economics may fail.

But dismissing the relationship because it is complicated preserves an assumption that no longer deserves to go unexamined: computing demand chooses its location first, and the power system accommodates it later.

Europe needs to move more electricity. It also needs to become more deliberate about where new demand appears and how rigidly that demand behaves.

“Compute where the power already is” is not a promise that every spare electron can become useful processing. It is a discipline for choosing where physical infrastructure belongs.

The grid still has to carry electricity. We can stop asking it to correct every geography we choose for compute.