More Data Centers, Same Grid

Flexible grid connections and onsite power could unlock new AI infrastructure without waiting years for every network upgrade.

Israel’s Electricity Authority has paused the processing of new data-center connection requests for 140 days, after applications reached 27,000 megawatts, roughly three times the country’s average electricity consumption. Denmark temporarily paused new large-load grid connection requests, while Texas introduced a batch-screening process after a surge in large-load applications strained project-by-project review. The scale of the queue justifies tougher screening, but treating every application in the same way risks delaying mature projects alongside speculative ones. Israel wants to become a regional hub for artificial intelligence, and the Authority should reconsider the blanket nature of the pause by creating a narrow, regulated track for projects that can meet stricter connection conditions.

Today, the choice is often treated as binary. Either the electricity system guarantees a data center its full requested capacity, including during the most constrained hours, or the project waits until new generation, substations and transmission lines are built. That model made sense when very large loads were relatively rare. For data-center developers, speed-to-power is becoming as important as access to land or fiber. As AI infrastructure arrives in clusters and on a much larger scale, a flexible connection model offers a third path.

A data center’s annual electricity consumption and the maximum capacity the grid must reserve for it at every moment are different questions. A facility may require substantial power throughout the year, but the grid’s inability to serve its full load can be concentrated in peak periods, local network constraints or contingency events. Designing every connection around the assumption that the grid alone must cover the theoretical maximum at all times can leave potential capacity unused.

There is another option. Under a flexible connection, a data center could receive most of the grid capacity it needs while accepting a clear obligation to reduce its draw during defined constrained periods. It could do so by shifting workloads where operationally possible, using storage for short-duration response or activating verified onsite power for longer events. The system operator would control when the obligation is triggered, while the data-center operator could connect years earlier than under a fully firm connection.

This changes the role of backup power. Data centers have traditionally installed onsite generation for rare outages, leaving expensive assets idle for most of their lives. The same assets can also help bridge limited grid constraints, provided they meet strict requirements for availability, duration, response time and safe operation. In that model, reserve power becomes part of the connection architecture rather than an emergency system that matters only after the grid fails.

International energy research increasingly points in this direction. The International Energy Agency has argued that regulators should examine incentives for data centers to use spare computing capacity, backup generation and storage more flexibly. It also notes that grid connection queues are lengthening while new transmission lines in advanced economies can take four to eight years to complete. Long-term grid expansion remains essential, but flexible connections can make better use of existing capacity while those investments advance.

The Electricity Authority should reconsider the across-the-board nature of the pause and create a narrow, regulated track for projects that offer grid flexibility. Eligibility should depend on clear evidence that a project is financed, advancing and able to limit its reliance on the grid during constrained periods through a combination of workload flexibility and onsite generation capacity. The rules should define how much firm grid power is reserved, when demand must be reduced, how performance is tested and what penalties apply if the data center operator fails to deliver.

The standard for onsite generation capacity must also be higher than simply owning a generator. The system must start reliably after long periods of inactivity, operate for the required duration and remain available when local fuel supply or the wider grid is under stress. Emissions, noise and safety matter as well, especially when data centers are built close to populated areas. Flexible connections will earn public and regulatory trust only if their backup layer is treated as critical infrastructure, with a commitment to community safety and environmental responsibility.

Israel will still need more generation, stronger transmission and distribution networks, and faster planning. Flexible connections distinguish between infrastructure required for long-term growth and capacity that can be made available sooner through better operating rules. Building every project around its worst possible hour is an expensive way to manage a system whose constraints change over time and by location.

The Electricity Authority has reportedly been developing a framework that would require data-center developers to make annual payments for reserved grid capacity, helping screen out projects that are not advancing. Such a requirement could support a more selective process during the pause. Screening can identify who is ready, while a tightly controlled flexible-connection track can test whether credible projects can proceed without weakening reliability.

Israel does not have to choose between AI infrastructure and a stable electricity system. It can require the companies building the largest new loads to become part of the solution, taking responsibility for the periods when the grid is under pressure. The Electricity Authority should preserve the pause where it is needed, while opening a disciplined exception for projects that can prove they are ready, reduce their peak dependence on the grid and meet enforceable obligations.

By Emmanuel Levy, CEO of Phinergy

Photo: Phinergy