Announced Gigawatts vs. Connected Megawatts: what RTE reveals about AI colocation providers?
TL;DR AI capacity announcements in France are now measured in gigawatts. Since its grid-connection reform, RTE has documented that only a minority of the capacity reserved for projects of this type is currently being effectively consumed. For a neocloud committing to several megawatts of IT capacity over several years, the gap between announced gigawatts and connected megawatts has become the dividing line in the AI colocation market, ahead of capacity announced on paper.
Several AI capacity projects in France have been announced in recent months with figures reaching several gigawatts. These announcements are dominating media coverage. The numbers are spectacular. They say nothing, however, about what will actually be physically connected to the power grid within the announced timelines. This is precisely the gap that RTE has been documenting since its reform of datacenter grid-connection procedures, and it is this gap that is now reshaping the AI colocation market.
For an infrastructure leader who has secured Series B or post-Series B funding and is committing 6 to 7 MW of IT capacity over several years, this gap is not an abstraction. It determines when their GPU clusters can go into production, in a market where gigawatt announcements and actual megawatt connections are following two distinct trajectories.
A Gigawatt announcement is not a delivery commitment
What RTE’s grid-connection reform actually measures?
A project of several gigawatts represents, roughly speaking, the electricity consumption of several million households. Such an announcement generally rests on a capacity reservation with the grid operator, an investment intention, and a multi-year construction schedule. None of these three elements constitutes a guarantee that the project will be commissioned within the stated timeframe.
Since 2025, RTE has been implementing a reform of grid-connection procedures specifically in response to the surge in capacity requests from datacenters and energy-intensive industrial sites. The reform introduces prioritization and conditionality criteria for power reservations, precisely because the volume of gigawatts requested from the grid far exceeds what the network can physically accommodate within the timelines sought by project developers.
The result of this tension can be summed up simply. Across the capacity reserved on the grid for projects of this kind, only a minority of the reserved power is currently being effectively consumed. The remainder corresponds to projects awaiting commissioning, delayed by grid-connection queues, sizing revisions, or internal decisions regarding deployment schedules.
This gap is not a market anomaly. It is structural and affects all players communicating in gigawatts without specifying their position in the grid-connection queue.
What this gap costs a neocloud during scaling
For an operator that has raised capital on the promise of large-scale inference deployment, a prolonged gap between announced capacity and actual commissioning has direct and measurable consequences.
GPU cluster deployment schedules depend on power that has actually been connected, which is distinct from announced capacity reservations. A grid-connection delay ties up capital already committed to hardware, postpones the commercial launch of inference capacity, and complicates the revenue trajectory presented to investors ahead of the next funding round.
This reality directly affects the case an infrastructure leader has to make internally, in front of the finance team and investment committee. This is the territory where execution matters more than announcements, and it is where Voltekko has built its model.
How does Voltekko structure its execution capabilities?
Voltekko designs and operates a European network of natively DLC-enabled colocation datacenters dedicated exclusively to high-density AI inference workloads.
Voltekko relies on EQUANS, a Bouygues subsidiary that has been designing, building, and operating datacenters for twenty years. This operational track record enables Voltekko to target deployment timelines of around 6 to 9 months, significantly shorter than the standards typically seen in the AI colocation market.
This approach is built around a level of capacity granularity designed for actual grid connections. Voltekko offers dedicated capacity blocks of approximately 6 to 7 MW IT per customer, sized according to the actual connection capacity available at each site. This granularity is designed to reduce exposure to the partial-connection risk that characterizes projects announced in gigawatts.
The network’s capital structure is backed by REED, a subsidiary of Société Générale Group and an investor in technology infrastructure.
On the hardware side, HYPERTEC handles the design, validation, and deployment of the latest generations of Nvidia GPUs as well as Groq architectures. Securing this supply chain upstream addresses a second risk factor, distinct from grid connectivity but equally critical to large-scale inference deployment: hardware availability at the moment the electrical capacity is delivered.
Frequently Asked Questions
What is the gap between announced gigawatts and connected megawatts?
It is the difference between the electrical capacity an AI datacenter project announces publicly, often in gigawatts, and the power actually connected to the grid and available for operations. RTE has been documenting this gap since its reform of grid-connection procedures began in 2025: only a minority of the power reserved for projects of this type is currently being effectively consumed.
Why did RTE reform datacenter grid-connection procedures?
The reform responds to the surge in capacity requests from datacenters and energy-intensive industrial sites. The volume of gigawatts requested far exceeds what the grid can physically accommodate within the timelines sought by project developers. The reform introduces prioritization and conditionality criteria for power reservations.
Preparing a large-scale inference deployment in Europe? Talk to the Voltekko infrastructure team.