At the 9 April 2026 US House of Representatives Committee Public Hearing on data centres, the Electricity Reliability Council of Texas (ERCOT) shared three important data points:

• ERCOT is tracking approximately 410 GW of large loads seeking interconnection, of which ~87% are data centres, with most of the interconnection requests at 750+ MW.
• Solar and energy storage account for more than 76% (341 GW of 454 GW) of the generation seeking interconnection in the ERCOT grid. 
• At a growth rate of 271%, some 61 GW of gas project capacity has entered the queue since the Texas Energy Fund (TEF) was passed in 2023.

These ERCOT statistics illustrate the impact that hyperscale entrants are having on the electricity system. Data centres now regularly request hundreds of megawatts to multiple gigawatts of power demand at a single location. This level of demand was historically associated with entire metropolitan areas and developed incrementally over decades. Hyperscale data centres, particularly those optimised for artificial intelligence (AI) workloads, now concentrate that demand in a small number of nodes and deploy on commercial timelines measured in months rather than years.

This shift is not simply about more electricity consumption. As outlined in the four findings below, the arrival of hyperscale data centres changes how loads interact with the power system, how risk is created and allocated, and how institutions govern the grid access function.

Different power system interactions 
Gigawatt-scale hyperscale data centre demand represents a structural change in power system development. A single customer now introduces city-scale demand in a few concentrated transmission nodes, creating non-linear requirements for substations, transformers and transmission reinforcement.

Supply chains face similar non-linear effects. The same long-lead equipment required for large generators is needed on the load side, creating direct competition for turbines, transformers and switchgear. When multiple projects cluster geographically and temporally, these stresses compound.

Gigawatt-scale demand creates operational challenges for grid operations. When such a facility abruptly disconnects or shifts to backup generation without coordinating with the system operator, generation continues while demand disappears, creating voltage fluctuations and local stability challenges and surplus conditions. When service is abruptly restored, demand returns in a step change that again creates voltage fluctuations that stress substations, transformers and transmission corridors not intended to handle such concentrated fluctuations.

The need to respect grid constraints  
The grid is responding to large-load growth with constraints that accommodate hyperscalers while ensuring system stability. In Texas, legislative and market reforms require large loads to declare minimum operating levels and accept curtailment obligations as a condition of interconnection. Most recently, Texas Senate Bill 6 established formal Public Utility Commission of Texas (PUCT) oversight of co-location agreements, further strengthening grid integrity requirements for large loads. In the Mid-Atlantic, market dynamics and policy responses are driving capacity price increases and prompting political intervention and the creation of dedicated procurement mechanisms. In the Southwest Power Pool (SPP), accreditation requirements increase the cost and complexity of serving large new loads.

Across regions, the signal is consistent: hyperscale demand must align with the realities of the power grid rather than expect a custom redesign of a shared utility. The system functions as a highway network designed for conventional traffic, where most routes are not engineered to accommodate the extreme performance demands associated with the AI era.

Curtailment obligations, controllable load resources (CLR) and bringing your own generation (BYOG)/behind-the-meter (BTM) arrangements are important mechanisms to ensure adequate electricity supply and grid reliability, particularly during system stress events. Large loads that can reduce demand during constrained periods and offset their system footprint with dispatchable generation are treated more favorably than inflexible demand that seeks unconditional service. Through these mechanisms, the grid signals what types of hyperscale load behaviour are acceptable.

The burden of demonstrating deliverability and the ability to operate within grid constraints rests increasingly with customers and their power counterparties. Stakeholders in ERCOT have largely accepted this direction, as reflected in strong support for the planned evolution of CLRs, with a clear majority indicating that CLR and BTM or BYOG capabilities should be declared at the application stage because of their potential to materially reduce net grid impact.

‘Google believes that co-location [of data centres and power generation] should be resource-agnostic provided it can cover its load obligations... entities may be able to operate complex, multi-resource co-location configurations – renewable, thermal, storage, demand response – that are able to manage 100% of a large load’s resource obligations.’ – Google, public comment to ERCOT Large Load Interconnection Workshop, February 2026

Grid access for large loads is evolving quickly 
Interconnection and grid access frameworks were built around incremental growth and sequential review. At gigawatt-scale, grid access cannot be efficiently granted on a project-by-project basis because it requires a system-wide evaluation of how much load the network can reliably absorb. ERCOT’s transition to batch studies reflects this shift.

When large-load interconnection requests grew rapidly, sequential studies could not keep pace. Batch studies allow the system operator to evaluate requests collectively, identify binding constraints, and determine how much load can be served within existing and planned infrastructure. Some projects receive full power allocation, others partial allocation, and some none at all. The outcome depends not only on project readiness but also on how each request interacts with others in the queue.

The high volume of gigawatt-scale projects also revealed a fundamental visibility gap. As interconnection queues grew, they became less reliable indicators of executable demand. Projects entered the queue before securing sites, capital or equipment, often to preserve optionality. From the system’s perspective, this made it difficult to distinguish which requests would ultimately materialise. This is not a physical problem, but an information and allocation problem: operators must plan infrastructure without a clear signal of committed demand.

In response, grid operators and regulators have introduced readiness criteria, financial security requirements and development milestones to improve queue quality. These mechanisms are designed to separate committed projects from speculative requests and to align queue position with execution intent. They also shift risk, requiring customers to demonstrate deliverability earlier in the process.

In one example, from another regional market, a major utility required developers to back their interconnection requests with financial commitments. The tariff required three things: a binding financial commitment upfront, coverage of at least 85% of stated minimum billing demand, and exit fees with liquidated damages for projects that walked away or under-delivered. The mechanism was straightforward, requiring developers to act like committed customers rather than option holders. The result was a roughly 80% reduction in projected demand in that service territory, from approximately 30 GW to approximately 5.5 GW. ERCOT's own Batch Zero process reflects a similar commitment to queue integrity, ensuring that interconnection requests represent executable projects that can be reliably served.

‘Our peak demand is around 85 GW built over decades. Hyperscalers are proposing 4 GW data centres. That's bigger than Austin [Texas] and arriving in 18 to 36 months. That is different.’ – Independent System Operator

Misalignment between hyperscaler demand and grid supply is a limiting factor 
The dominant constraint across all findings is time. Hyperscale data centres deploy in roughly 18–24 months. Dispatchable, firm, flexible generation (for example, natural gas) typically requires 4–5 years, and transmission often takes longer. This mismatch is structural and does not change with market design or institutional reform.

BTM generation emerges as a direct response to this timing gap, with the ability to deliver solar (or wind) and battery solutions in 18–36 months. On-site or co-located generation offers a path to power delivery on a hyperscaler-compatible timeline while longer-duration grid and transmission assets are developed in parallel. The timing logic supports hybrid approaches in which hyperscalers initially rely on BTM resources and transition to greater grid reliance as system capacity becomes available. Loads can then ramp in alignment with grid build-out plans, with BTM solutions bridging near-term shortfalls in generation availability.

At the gigawatt scale, however, self-supply introduces its own limitations. The same turbines, transformers, switchgear and engineering capacity required for utility-scale generation are also required for BTM projects of this size. Equipment lead times remain long, permitting processes are complex, and operating large power assets requires specialised and limited expertise in fuel procurement, dispatch, maintenance and compliance.

For power providers, the timeline and offtake mismatch creates a distinct and elevated risk profile. Capital must often be committed ahead of firm, long-duration grid service certainty. Assets may need to be sized for load that materialises unevenly or later than forecast. Stranded asset risk, customer concentration risk and execution risk increase materially at gigawatt scale under these conditions. These risks are not hypothetical; they are inherent in attempting to accelerate large-scale power development beyond the system’s typical investment cycle. At gigawatt scale, a customer that walks back a commitment leaves the provider holding generation capacity with no obvious secondary market.

Consider for example the contract arrangements of a specialist digital infrastructure developer that leases a multi-building campus, paired with long-term power arrangements from a utility-scale power producer. The campus lease is multi-decade and phased. Rent and ramp-up are tied to energised capacity milestones. Additional megawatts, substations and mechanical-electrical infrastructure must be delivered by specific dates and to defined reliability standards. If power-ready capacity is late, the contractual responses resemble project finance, with pre-agreed delay payments and step-in rights if the developer cannot perform. In parallel, the operator structures its power contracts so that campus energisation milestones and commercial operation dates are closely coordinated, and consequences of delay are coordinated across both sides of the arrangement.

For the hyperscaler, the effect is to ‘lease’ both space and megawatts: long-term control over powered land plus a matching stream of contracted power. For the real estate and power counterparties, the structure consolidates risk around a single investment-grade anchor tenant and supports large-scale capital deployment. What reads as a lease and set of power contracts functions in practice as an integrated hyperscale power solution.

In conclusion, hyperscale demand cannot be served effectively through traditional utility or generation models alone. Success depends on coordinating across misaligned timelines, structuring around conditional grid access, operating in ways that support the grid, particularly during periods of scarcity or system stress, and explicitly allocating risk through commercial and contractual design. The future of hyperscale power supply will be shaped by system level constraints and coordination. Power providers that can deliver near-term capacity, operate within evolving institutional frameworks and build long-duration infrastructure in parallel will define how gigawatt-scale demand is integrated into the power system.

• Further reading: ‘How to avoid future black-outs: UK and Texas grid operators compare and contrast’. Two leading system operators – the Electricity Reliability Council of Texas (ERCOT) and the UK’s National System Operator (NESO) – are compared to better understand the challenges faced by electricity system operators as renewable energy systems are integrated with grids.
• ‘Will Texas still steer towards renewables despite many grid challenges?’ Discover why renewables are being deployed at scale in Texas, a US state rich in fossil fuels, and how they are impacting a strained grid.