Google, Tapestry, and PJM take on the interconnection queue with AI

An Alphabet moonshot aims to compress the paperwork that keeps 2,600 GW of power capacity waiting to reach the grid.

The bottleneck is a queue, not a shortage of projects

Google announced a multi-year collaboration with PJM Interconnection—the largest grid operator in North America—and Tapestry, an Alphabet-incubated moonshot powered by Google Cloud and Google DeepMind. The stated aim is to speed up how new power generation connects to the PJM grid, which spans the District of Columbia and 13 states and serves 67 million people.

The framing matters. The problem Google is describing is not that there aren't enough energy projects. It's that projects are stuck waiting. Per Lawrence Berkeley National Laboratory, the U.S. interconnection backlog at the end of 2023 was roughly 2,600 gigawatts of potential capacity—more than double the total installed capacity of the country's existing power fleet.

Grid operators, the post notes, have seen interconnection requests rise 'from a few dozen annually to thousands.' That volume increase is the target. This is a data-processing problem dressed up as an infrastructure problem.

What Tapestry is actually building

The concrete deliverable is a unified data model. Tapestry plans to integrate the 'dozens of existing databases and tools' PJM uses to evaluate interconnection requests into a single unified model of the network, and to build a secure platform where grid planners and project developers can collaborate.

The AI work centers on automating and improving data verification—the checking step that today burdens both energy developers filing applications and PJM planners reviewing them. Google frames this as reducing the time to process new project applications so capacity comes online faster.

This is a narrower and more credible claim than 'AI will fix the grid.' The physical constraints—transmission lines, transformers, the actual electrons—don't change. What changes is the speed of the application, verification, and planning workflow that sits in front of construction.

The demand numbers behind the urgency

Google grounds the effort in a specific forecast shift: in 2024 the Federal Energy Regulatory Commission's five-year demand growth projection more than tripled compared to the prior year's forecast, with U.S. peak energy demand expected to grow by 128 GW before the end of the decade.

The subtext is Alphabet's own consumption. The post closes by noting Google is 'exploring new development and procurement approaches that can deliver new, firm electricity to help power Google's operations,' alongside investments in enhanced geothermal and advanced nuclear. A company building AI data centers has a direct interest in a faster interconnection process.

That alignment is worth stating plainly rather than reading as a conflict: the entity building the grid-planning AI is also one of the parties whose demand is driving the queue longer.

Why unifying the data model is the real deliverable

The most durable part of this announcement is not the AI framing—it's the database consolidation. PJM's interconnection evaluation reportedly runs across dozens of separate tools and databases. Any team that has worked on applied AI knows that fragmented, inconsistent source data is the thing that stalls automation, not the modeling.

If Tapestry succeeds at unifying those sources into one coherent model of the network, that structured foundation is what makes downstream automation of verification possible. The AI tooling is the visible layer; the data integration is the load-bearing work.

The specific implication: this is a bet that the interconnection queue is best attacked as a data engineering and workflow problem for a single large operator, not as a general-purpose grid AI product. Whether it moves the 2,600 GW backlog will depend on how much of PJM's approval time is administrative versus physical—and only PJM's own throughput numbers, over the multi-year horizon Google cites, will tell.

Find this story relevant to you?

Cerchi un partner di ingegneria AI capace di costruire davvero?

Aiutiamo le aziende a integrare l AI nei prodotti, automatizzare i flussi di lavoro ad alto valore e modernizzare i sistemi software che sostengono la crescita.

Get in touch

Letture correlate

Altre analisi su delivery di prodotto, IA operativa e il lavoro sui sistemi che fa reggere il deployment alla prova dei fatti.

Oct, 94 min to read

AI Products

Meta adds Hindi and Portuguese to its Reels translation and lip-sync feature

Feb, 114 min to read