← All insights

United States / Data centres · Grid access · Flexible demand

The Data Centre Power Race: What Is Flexibility Actually Worth?

For a data-centre investor, the value of flexibility may begin before the first electricity trade: in the connection terms that determine when the campus can operate. But every promise to reduce load carries a computing obligation.

The scarce input can be deliverable power

A campus can secure land, equipment and a renewable supply agreement without securing the firm network access needed to operate on schedule. That makes electricity a development constraint as well as an operating cost. A flexibility arrangement deserves evaluation against both dimensions, with the connection benefit kept separate from the revenues earned after energisation.

The scale of demand is material, but precision matters. LBNL’s December 2024 study estimated US data-centre consumption at 176 TWh in 2023 and modelled a 2028 range of 325–580 TWh. Those are historical estimates and dated scenarios, not a current queue of financeable projects or a forecast of how much load can be interrupted.[1]

THE SCALE / THE UNCERTAINTY

Power demand is a range, not a single number.

US data-centre electricity use · TWh per year

2023 estimate
176
2028 scenarios
325–580

December 2024 LBNL study. Shaded extension shows the scenario range, not a confidence interval. Electricity consumption is not a measure of flexible capacity. [1]

Flexible load has entered major power agreements

In March 2026, Google said it had incorporated 1 GW of demand-response capacity into long-term contracts with US utilities. It described shifting or limiting a portion of machine-learning workloads, with arrangements intended to support faster grid connections. This is evidence that a major buyer is contracting for flexibility. It is not independent evidence of a gigawatt delivered simultaneously, or proof that every workload can respond.[2]

The regulatory direction is also consequential. FERC’s June 2026 large-load orders called on six regional operators and their transmission owners to justify or reform relevant tariffs, including consideration of new transmission services for flexible large loads. The orders establish a reform direction; the applicable tariff and site-specific connection agreement determine the rights available.[3]

PJM’s published 2026 Connect and Manage work makes the trade-off concrete: the framework for new large loads without their own new generation envisages curtailment before pre-emergency demand response, subject to development of the operational arrangements. October workshop materials include draft tariff revisions. Curtailment priority can be as important to computing economics as the connection date.[6]

Texas has its own framework. ERCOT’s approved Batch Zero approach groups qualified large-load requests of 75 MW and above for system study. An investor needs to establish how the relevant site is treated within its actual connection process; “the US power market” is too broad a basis for an energisation assumption.[5]

Define the flexibility before assigning its price

There are several different products hidden inside the phrase “flexible data centre”. Work can be rescheduled at the same location. Suitable work can be moved elsewhere. Grid imports can be reduced while storage supports operations. Each changes the energy portfolio and the computing service in a different way.

The IEA 4E EDNA’s July 2026 report distinguishes market, network and system-serving flexibility and identifies operational and economic barriers across data-centre types. A dependable proposition must therefore specify the response time, duration, notice, maximum event frequency, recovery needs and computing services that remain protected.[4]

A battery used for grid response cannot simultaneously promise its full inventory as an untouched resilience reserve. Moving workloads to another campus may shift the constraint or create a new power cost elsewhere. Deferring work can produce a rebound in consumption. The financial model needs to follow these consequences through the whole operating portfolio.

Separate four sources of value

First is development value: an earlier or more economical connection, where the relevant utility or network actually offers it. Second is power-system value: a recognised service under a tariff or agreement. Third is energy-portfolio value: avoiding expensive supply or using contracted renewable output more effectively. Fourth is the cost of flexibility: lost or delayed computing contribution, additional infrastructure, recovery and performance exposure.

These values cannot simply be added. The same interruption may be required under the connection agreement and therefore unavailable for sale as a separate discretionary service. A demand-response payment may compensate for a commercial restriction rather than represent a free additional margin. The contracts determine the available combination.

A simple illustrative calculation shows why energy revenue alone is an incomplete valuation. Reducing a 100 MW campus by 15% for four hours shifts 60 MWh. At an assumed US$900/MWh difference between the avoided and replacement energy prices, the gross energy benefit is US$54,000. That excludes computing costs, rebound, tariff effects and payments. It is a worked example, not a forecast or project return.

The same 15 MW capability might also influence connection timing, but only through a specific agreement that recognises it. That benefit could be strategically significant; it must be valued from the actual development schedule and binding terms, rather than inferred from the spot-market calculation.

The diligence question is what the promise commits

Investors should ask which loads can respond under adverse conditions, which services cannot be interrupted, and who authorises the action. They should examine the availability model, reserve policy, event history or test evidence, and the consequences of a missed response. Contracted megawatts are a starting point for diligence, not the final measure of dependable capability.

There is also a credible limit to the opportunity: new generation, networks and storage remain necessary. Flexibility can improve how existing capacity is used; it cannot make every constrained site viable or support unlimited computing growth. A business case that relies on broad assumptions about interruptibility should be discounted until the specific operating and contractual evidence exists.

ESKALA’s view is that electricity strategy and computing strategy increasingly need a shared decision framework. The value comes from comparing feasible choices across power supply, flexibility, service commitments and risk. The question is not simply what a megawatt of response earns, but which development and operating outcomes that response makes possible.

Sources & methodology

Primary sources checked on 29 October 2026. The commercial interpretation is ESKALA’s analysis. Illustrative calculations are identified in the text; no modelled result is presented as an observed project outcome.

  1. Lawrence Berkeley National Laboratory · US Data Center Energy Usage Report ↗

    December 2024. 2023 estimate and 2028 scenario range; not a new 2026 forecast.

  2. Google · 1 GW demand-response contracting milestone ↗

    19 March 2026. Company-reported contracted capacity; not independent evidence of delivered response.

  3. FERC · Large-load integration orders ↗

    18 June 2026. Includes flexible transmission-service reform. The orders do not establish a universal flexible-connection entitlement.

  4. IEA 4E EDNA · Data Centres and Flexibility ↗

    1 July 2026. Types of flexibility and operational/economic barriers.

  5. ERCOT · PUCT approval of Batch Zero ↗

    18 June 2026. Batch study framework for qualified large-load connection requests of 75 MW and above.

  6. PJM · Connect and Manage framework for large loads ↗

    2026 implementation work, read alongside PJM’s October 2026 Co-Located Load Order workshop materials. Framework development and draft tariffs do not establish a site-specific connection right.

THE PORTFOLIO IMPLICATION

What does this change
for your business?

Discuss the commercial question with ESKALA ↗
Discover ESKALADiscover ESKALA