Can a data center agree to use less power?

On March 19, 2026, Google announced an unusual kind of energy resource.

It was not a new power station, transmission line, or battery. It was 1 GW of electricity demand that Google had agreed could be limited or shifted under contracts with several U.S. utilities.

A gigawatt is a large number. The more interesting fact is what it does not mean.

Google did not promise to switch off an entire gigawatt of data centers whenever the grid asked. Nor did the announcement represent a gigawatt-hour of energy saved. It referred to demand-response capacity incorporated into long-term utility agreements, using selected machine-learning workloads that Google says can be reduced or moved during certain periods.

This is first-party program evidence. It does not show that the full gigawatt has already been dispatched simultaneously or establish how often it will be available. Google presents the agreements as one way to connect facilities before every planned generation and grid investment is complete.

The bargain is attracting utilities, regulators, and operators.

The grid gains a defined right to ask for less demand when capacity is scarce. The data center may gain something in return: an earlier connection, a different tariff, direct compensation, or access to capacity that would otherwise remain unavailable.

But data centers sell availability.

Demand response asks which part of that promise is negotiable.

The grid is buying a change in behavior

Demand response is sometimes confused with energy efficiency. They do different jobs.

Efficiency reduces the electricity needed to provide the same service. Demand response changes when or how much a consumer uses because the grid, market, or price sends a signal.

Lawrence Berkeley National Laboratory divides demand flexibility into three broad actions: shedding, where demand falls for a period; shifting, where consumption moves to another time; and modulating, where demand changes repeatedly or gradually in response to system needs.

For a data center, the difference might look like this:

• A more efficient GPU completes the same work with less electricity. That is efficiency.
• A batch job waits until later, when the grid has more capacity. That is load shifting.
• Processors temporarily operate below their normal power limit. That is reduction or modulation.
• A battery supplies part of the facility for an hour. The computing demand may remain unchanged, but the grid sees a smaller load.

These mechanisms can be combined. They should not be treated as equivalent.

A job moved to 3 a.m. still has to run. A power cap may lower demand while making the job take longer. A battery reduces grid draw immediately, but it eventually has to recharge and may also be reserved for resilience.

The grid is not buying a vague promise that a data center will be helpful. It is buying a measurable change in electricity use under defined conditions.

A data center contains several kinds of flexibility

From outside the fence, a data center can look like one large load. Inside, it contains workloads with different deadlines, hardware with different operating limits, cooling and power systems, batteries, storage, networking, and spare capacity kept ready for failures or demand spikes.

The most valuable flexibility may sit in the computing work.

Some workloads have little room to move. Payments, communications, medical services, live inference, and other customer-facing systems may need to respond immediately. Slowing or moving them can break latency targets, service agreements, or essential operations.

Other work has slack. Video processing, model evaluation, rendering, data preparation, simulations, backups, and parts of machine-learning development may tolerate a delayed start or slower execution.

The useful resource is the time between when a job could run and when it must finish.

Google began using that margin in 2020, shifting non-urgent tasks toward hours when lower-carbon electricity was more available. It later moved suitable media-processing work between facilities where data, privacy, capacity, and operating conditions allowed. The examples involved processing for services such as YouTube, Photos, and Drive rather than the real-time functions users expect continuously.

Hardware creates another option. Processor power can be capped, clock speeds reduced, or a training job paused at a checkpoint where the software supports it. These actions are not free. They can delay later work, leave expensive hardware idle, or require coordinated restart.

Facility systems create a third layer. Cooling controls, batteries, and on-site power equipment can change what the grid sees even when the workload does not move. Berkeley Lab’s 2026 review identifies computational load management, facility adjustments, energy storage, and on-site generation as complementary routes.

Cooling has limited room for improvisation. Temperatures may be managed briefly through thermal inertia, control changes, or stored cooling, but equipment must remain within safe conditions. The heat still has to leave.

Batteries can respond much faster than most workloads, but a battery installed for outage ride-through may not have its full capacity available for regular grid services. Using it changes state of charge, cycling, degradation, and readiness for the event it was meant to cover.

The flexible resource is not “the data center.” It is a selected combination of workload slack, hardware control, thermal margin, stored energy, and operating permission.

The service promise sets the limit

A job scheduled to finish by 6 a.m. has flexibility only while that deadline remains safe.

Suppose it needs four hours of computation and enters the queue at 8 p.m. The operator can delay it for two hours without changing the result delivered to the customer. A demand-response request at 9 p.m. can use some of that margin. A second request at 1 a.m. cannot: the spare time has gone.

The same facility may therefore have substantial flexibility one day and almost none the next.

Geographical movement adds another set of limits. A workload can run elsewhere only if another site has suitable hardware, capacity, network access, the required data, compatible software, and permission to process the information there.

Moving a small stateless task is one thing. Moving a job whose working data spans petabytes is another. Data location and jurisdiction may prevent movement even when the technical system allows it. Network transfer consumes time, energy, and money.

This is where an electrical problem becomes a distributed-systems problem. Hivenet’s distributed-systems explainer shows why placement depends on state, data, networking, failure handling, capacity, and operating responsibility rather than available hardware alone.

The question is never simply whether compute can move. It is what must move with it.

A fast demonstration does not establish lasting flexibility

In 2026, EPRI Europe reported a UK demonstration involving National Grid, Emerald AI, Nebius, NVIDIA, and EPRI.

According to EPRI, the pilot reduced demand at a high-performance AI data center by as much as 30% to 40% within seconds without disrupting critical workloads. This is first-party program reporting. It shows that rapid response was technically possible in that operating arrangement; it does not mean every AI data center can shed 40% on demand. EPRI describes the pilot here.

A commercial flexibility service needs more than a peak demonstration. How long can the reduction last? How often can it happen? Which workloads were protected? What happens to delayed work afterward? Can the same response be delivered during peak customer demand? How much notice is required?

Google’s 1 GW announcement has a similar boundary. Contracted capacity shows that utilities and a major operator are willing to build flexibility into long-term agreements. It does not reveal how much energy will shift over a year.

ENTSO-E’s May 2026 work presents flexibility as a potential role requiring new connection practices and operational coordination, not as a capability the sector can assume.

The grid needs a resource it can rely on. The operator needs confidence that answering the call will not break the service sold to customers. The contract sits between those two forms of reliability.

Most of the product is in the fine print

A serious demand-response agreement has to define more than megawatts.

It needs a baseline: what the facility would have consumed without the event. It needs a response time, duration, and frequency limit. It needs to address the rebound when delayed work returns, and it needs measurement and verification.

It also needs priority rules. Safety, cooling, storage integrity, essential services, and customer commitments may be excluded. Cost and risk must be allocated: failed delivery, idle accelerators, delayed jobs, battery cycling, workload migration, and non-firm connection rights all have a price.

EPRI’s DCFlex program has reviewed a wide variety of flexible-load tariffs from U.S. utilities. The variety is evidence that “data-center flexibility” has not settled into one standard commercial product. Utilities are still deciding what they need, what they will pay for, and how delivery should be verified.

Technology is only part of the work. Flexibility becomes useful when it can be bought, dispatched, measured, and trusted.

An earlier connection comes with a different right to power

The commercial attraction becomes clearest where connection capacity is scarce.

A firm connection is intended to give the customer access to contracted capacity under agreed conditions. Building enough network infrastructure to support that right can take years.

A flexible or non-firm arrangement changes the bargain. The customer may connect earlier while accepting that part of its demand can be limited under specified network conditions.

The IEA recommends exploring non-firm connections and incentives for demand response as bridges while slower grid investments proceed, not substitutes for those investments.

Demand response does not create electrical capacity from nothing. It changes the reliability right attached to part of the load.

A facility serving delay-tolerant batch work may find the exchange attractive. A site dominated by strict latency and availability requirements may have much less to offer. A mixed facility may keep one part of its electrical envelope firm while allowing another to respond.

Two sites may each have a 100 MW connection. One expects firm access to almost all of it. The other has agreed that 20 MW can be limited. The nameplate number is identical. The right to consume it is not. That is another reason megawatt labels need explanation.

The work comes back

Demand response is often illustrated as a clean dip in a load curve. Computing makes the second half more complicated.

If a workload was canceled permanently, its energy may disappear. If it was made more efficient, less energy may be needed. If it was delayed, the work remains in the queue.

Restart everything immediately and the delayed demand can create a rebound peak. Restart gradually and customer deadlines may come under pressure. Geographical shifting can move the rebound elsewhere rather than remove it from the wider system.

This connects demand response with renewable-energy curtailment. Moving suitable work toward hours of strong wind or solar output can help align demand with generation, but a grid emergency is not automatically a renewable-energy strategy. Reliability, connection capacity, price, renewable integration, and emissions are related goals, not interchangeable outcomes.

Flexibility has to be designed into the deployment

Policloud develops and deploys physical, modular data-center infrastructure for defined sites. That does not make a Policloud unit flexible demand by default.

The electrical agreement would need to define which part of the connection is firm. Workload owners would need to identify services that can slow, stop, or move. The operating team would need controls and telemetry capable of delivering and verifying the response. Batteries would need rules protecting their primary role. Commercial agreements would need to allocate the cost of interruption, delay, migration, idle hardware, and non-delivery.

A modular deployment may make the conversation easier to bound. Capacity can be considered in increments, and each increment assessed against the power rights and operating constraints available at the site.

The credible proposition is narrower than “the data center becomes a battery.” A deployment can be designed so that electricity availability and workload behavior are considered together before operating promises are fixed.

The grid is buying the margin around the workload

A data center does not become flexible because someone installs a switch beside the meter.

The resource is the margin inside its operations: the minutes before a job must start, the processor power that can be reduced without missing a deadline, the battery capacity left after resilience needs are protected, the capacity available at another site after data and jurisdiction are considered, and the difference between maximum connection capacity and the part that needs an uninterrupted right to power.

That margin changes from hour to hour. It also has a price.

A flexible data center is not one that can simply be switched off.

It is one that knows, before the grid calls, exactly what can move, for how long, under whose authority, and at whose cost.