
In March 2026, Google announced that it had incorporated 1 GW of data-centre demand response into long-term electricity contracts with five US utilities. The arrangements allow Google to reduce or shift selected machine-learning workloads when the electricity grid needs relief, potentially helping utilities connect new data centres sooner. A gigawatt is a substantial amount of contracted flexibility, especially when concentrated in professionally managed facilities with sophisticated control systems.
Less than a month earlier, Google had submitted a more cautious assessment to the Midcontinent Independent System Operator, MISO. In its February 23 response to the operator’s consultation on connecting large electricity consumers, Google described flexibility as a finite resource that should be used sparingly, principally during pre-emergency and emergency conditions. Its computing services must remain dependable, and some workloads cannot be interrupted or delayed without consequences for customers.
The two statements are consistent. Google has identified electricity demand that can be managed, entered commercial arrangements to make that capability useful, and acknowledged the limits of what it can promise. Its primary business is supplying computing services, not grid flexibility. The same distinction applies to factories, refrigerated warehouses, water utilities and electric vehicle fleets. They contain equipment capable of adjusting electricity consumption, but that equipment exists to produce something other than a grid service. The flexibility available to the electricity system is the operating freedom remaining after those obligations are protected.
Google’s March announcement describes arrangements with Indiana Michigan Power, the Tennessee Valley Authority, Entergy Arkansas, Minnesota Power and DTE Energy. These agreements can help accommodate large new electricity loads while longer-term infrastructure is developed. Google’s submission to MISO explains why that opportunity cannot be treated as unlimited. Interactive computing has demanding continuity and latency requirements, while some machine-learning training and other less time-sensitive workloads can be rescheduled or moved.
Deferring a computing task does not eliminate it. Training workloads have deadlines, computational requirements and resource dependencies. Shifting work to another data centre changes where electricity is consumed and may affect network constraints elsewhere. Pausing activity also interacts with cooling, equipment operation and the capacity available to complete deferred work afterward. These considerations determine the flexibility that can be offered at a particular location and time, rather than invalidating the concept of flexible computing.
The distinction matters because data-centre developers have strong incentives to demonstrate that their demand need not be continuously supplied at its maximum level. Earlier access to electricity can bring valuable computing capacity into operation sooner. Utilities and regulators must meanwhile protect existing customers from reliability risks and inappropriate infrastructure costs. The credible proposition is not that every data-centre megawatt is firm or interruptible, but that a defined portion can be managed under agreed conditions while the customer continues providing its essential services.
Commercial vehicle charging offers another way of creating operating room. In a six-month pilot, Ford Pro and Southern Company coordinated charging across more than 200 Ford F-150 Lightning trucks and over 150 Level 2 chargers. The system used vehicle and charging information to adjust consumption in response to electricity-system conditions while maintaining the fleet’s operational readiness. Ford reported a 500 kW reduction during a 30-minute demand-response event, with three demand-response tests during the pilot.
The significance lies in coordinating charging around actual vehicle requirements rather than merely demonstrating that chargers can be switched off. A vehicle needed for an early departure must receive sufficient energy beforehand. Another vehicle parked overnight may have several hours of scheduling freedom. The flexibility available from the same depot changes as vehicles arrive, charging progresses and departure times approach. Managing those windows can reduce peak electricity demand without compromising mobility, but the schedule must leave room to complete the necessary charging.
Water pumping and refrigeration create similar opportunities through physical storage rather than vehicle or computing schedules. Water can be pumped into reservoirs before it is needed. Refrigerated spaces and products can retain cold while compressors temporarily reduce electricity consumption. These buffers allow a facility to change when it uses electricity without immediately changing the service delivered to customers.
A California Energy Commission field study tested both types of equipment. At a water pumping station, operators reduced the electricity demand of an extraction pump by approximately 30% for periods of up to four hours without disrupting water service. At an industrial refrigeration facility, researchers demonstrated upward and downward adjustments averaging 25% of baseline compressor demand for periods exceeding 12 hours. These were results for particular equipment under tested conditions, not reductions in total facility electricity consumption or universal capabilities of water and refrigeration systems.
The refrigeration example also illustrates that useful flexibility is not limited to reducing demand. Compressors can sometimes consume more electricity when renewable generation is plentiful, storing additional cold within acceptable temperature limits. During a later constrained period, they may use less electricity while the stored cooling continues providing the required service. The same general approach applies to water storage, hot-water systems and some industrial processes. What moves is the electricity consumption needed to produce a service, while the service itself must remain available.
These forms of flexibility share an important characteristic: the underlying work still needs doing. A vehicle that pauses charging must eventually receive the energy required for its next trip. A deferred computing task must be completed. Water withdrawn from storage may have to be replenished, while a refrigeration system operating at reduced power may need additional compressor operation later. The timing of that recovery can determine whether a successful demand-response event delivers the expected electricity-system benefit.
Shifting electricity demand out of a constrained period can create additional consumption later as deferred work is completed.
Animation description: An illustrated industrial campus shows computing, electric fleet charging and refrigeration continuing their primary functions. A schematic demand profile falls below reference consumption during a grid event and rises during recovery. The graphic distinguishes reduced demand during the event from the work and energy requirements that remain afterward. The schematic is illustrative, not measured data.
Consider Ford’s reported 500 kW reduction during a 30-minute event. If that reduction were sustained throughout the interval, it would represent 250 kWh of electricity demand moved out of that period. It would not establish that 250 kWh of electricity consumption had been eliminated. Some or all of the charging could occur later, subject to the actual charging schedule and changes in efficiency. The immediate service is reduced demand during the requested window, not necessarily an annual energy saving.
This distinction is particularly important when many flexible resources respond to the same signal. A group of depots may suspend charging during an evening peak, then resume together when the event ends. Refrigerated facilities could likewise increase compressor consumption while restoring their thermal buffers. A poorly coordinated recovery can create another peak shortly after the original one, potentially on a network already facing constraints.
Recovery can also affect a different electrical location. A data centre that moves computing to another region may relieve one constrained network while increasing demand elsewhere. Shifting work or charging to a cheaper period may help the wider system but concentrate consumption on a particular feeder. The value of the original reduction depends on the timing, location and scale of the subsequent activity. Evaluating the response without its recovery period gives an incomplete picture of what the flexibility arrangement accomplishes.
The full cycle may use slightly more or less electricity as equipment efficiency, ambient conditions and operating schedules change. Those effects should be included when assessing energy and emissions outcomes, but they are separate from the MW of flexibility delivered during a specified period. A resource can provide valuable peak relief without reducing annual electricity consumption. Likewise, a reduction in annual consumption does not necessarily provide dependable capacity during the hours when the system needs it.
Operating conditions introduce another limitation. A refrigerated warehouse may have less flexibility during extreme temperatures or unusually heavy product movement. A water utility’s available response changes with storage levels and water demand. A commercial fleet may offer substantial flexibility on one evening and almost none before an intensive workday. Data-centre scheduling freedom changes with workload mix, deadlines and computing capacity. A resource that can respond effectively under one set of conditions is not necessarily equally available under another.
Professionally managed facilities have an advantage because these constraints are understood by operators who already monitor production schedules, equipment condition and service requirements. Automation can turn that information into dependable commitments. The opportunity is substantial, but the commitment must be based on the operating room actually available, including what happens afterward. A service promising a particular reduction for two hours cannot ignore the possibility that the customer will need the same equipment operating at full capacity during the third.
Customer economics determine whether that operating room can be offered repeatedly. Shifting activity may reduce electricity costs, avoid demand charges or allow a business to use an existing connection more effectively. It can also impose scheduling costs, equipment wear, additional labour, lost production or risks to service quality. A programme that is attractive for occasional events may become uneconomic if interruptions are frequent. Compensation and operating obligations must reflect what the customer is giving up, rather than assuming flexibility is free because the equipment has already been purchased.
The commercial arrangements also need to distinguish two different kinds of obligation. A customer accepting conditional grid access may agree to remain below a defined import limit during constrained periods. Compliance with that limit is part of its connection rights, whether or not it would otherwise have consumed more electricity. An existing customer selling demand response is offering a service that must be assessed against the relevant dispatch and performance requirements. The same facility can potentially participate in both arrangements, but respecting a connection limit is not automatically an additional demand-response service that deserves separate payment.
The detailed limits and operating controls are matters for the responsible network operators, customers and regulators. At the architectural level, the important distinction is whether the utility is purchasing an incremental response or granting access subject to operating conditions. Each requires clarity about the applicable constraint, the rights being granted, who may instruct the customer, how compliance is demonstrated and what happens when the arrangement cannot be maintained. Conditional access can enable useful loads to connect sooner, but does not make the underlying network constraint disappear.
Customer overrides are another part of that agreement. A utility fleet may need a vehicle urgently, a refrigeration facility may approach a critical temperature, and a data centre may encounter computing requirements that cannot be deferred. These conditions should be reflected in the service offered and the capacity the buyer can rely upon. An override explicitly permitted by the agreement is different from failing to deliver an unconditional commitment. Conversely, an arrangement with broad, poorly defined exceptions may protect the customer without giving the electricity system a dependable resource.
The Flexibility Mesh treats these relationships as part of the architecture rather than peripheral commercial details. The physical resource, customer’s primary mission, operating authority, network permission, contracted service, verification and payment all have to align. A fleet’s charging capacity or a data centre’s electricity connection does not by itself establish the amount of flexibility available for purchase. The relevant resource is the portion that can deliver a named service under defined conditions, including duration, recovery and compatible commitments.
For utilities and investors, a credible proposal should identify what activity is being protected, how much electricity demand can change, when and for how long, how often the response can be repeated, and what recovery requires. It should establish who controls the response, which conditions permit an override, how performance is assessed and why continued participation makes sense for the customer. These questions do not replace engineering, contract development or operational testing. They establish the evidence and responsibilities those activities must resolve.
Google’s agreements, Ford’s managed-charging pilot and the California industrial trials show that useful flexibility can emerge from equipment already serving other purposes. Better controls, scheduling and commercial arrangements can make more of that flexibility available without requiring customers to abandon their primary activities. The opportunity is not to treat the entire electricity demand of these facilities as dispatchable, but to identify and coordinate the operating freedom that remains after their obligations are met.
That leads to the next question in the series. Some constraints can be managed through temporary reductions, conditional connections and better use of existing infrastructure. Others persist despite effective demand management, particularly as electrification increases the underlying need for network capacity. The next article, “Build the Grid and Use It Better,” will examine how permanent reinforcement and operational flexibility should be considered together, and when additional infrastructure remains necessary.
This is the fourth article in the Flexibility Mesh series, examining how physical resources, operating requirements, institutions and customer economics combine into dependable electricity-system services. Subscribe to the TFIE Strategy Briefing for the remaining articles and further assessments. For an independent assessment of a large customer’s flexibility, a commercial portfolio or an electricity-system strategy, contact michael@tfie.io.

