Today I read an interesting article about the data center project of $NBIS (-2,62%) in Vineland (New Jersey). The Nebius cops among us might be interested. But as I said, nothing confirmed yet, but rather a chain of circumstantial evidence.
The investors should know that Nebius is building a data center there. However, the article focused on an exciting additional hypothesis, namely that the site is possibly planned in an (unusually) large expansion stage and relies heavily on its own on-site power supply so as not to be dependent on the years-long grid connection process. There was talk of a size of around 400 megawatts, i.e. hyperscale level, where the plant could in principle function like its own energy supplier "behind the meter".
The article explained that the electricity issue is currently the central bottleneck in the Ki infrastructure. Buildings can be built, GPUs can be bought, but without sufficient power on site, everything remains theory. Grid connections often take several years and the procurement and integration of technology "behind the meter" is also an increasingly limiting factor. This is why strategies are becoming more and more relevant in which operators solve the power supply themselves in order to get projects up and running more quickly and not have scaling dictated by the grid operator.
For months, the CEO of DataONE has publicly referred to a "systemic innovation" in Vineland related to power, speed and efficiency, without giving details.
Technically, an industrial island microgrid has been described as a plausible concept. Many large, medium-speed natural gas engines generate continuous electricity directly at the site, supplemented by flywheels for stabilization. The background to this is that AI workloads generate very rapid load jumps, for example due to inference peaks. Flywheels can absorb such short-term peaks extremely quickly while the motors readjust. This would enable stable operation that can also meet the high availability requirements of a hyperscale AI data center. I found it interesting that the article also mentioned earlier speculation about Vineland, e.g. about fuel cells or alternative generator technologies, but categorized these as either too expensive, too maintenance-intensive or not sufficiently scalable. Instead, the current hypothesis clearly focused on the pragmatic approach of "own generation plus very fast stabilization" because this solves the timing problem most directly.
Other indications cited were industry reports and public comments pointing to an East Coast project with around 400 megawatts, phased expansion and a combination of gas engines, high-efficiency generators and flywheel stabilization. Specifically, there was talk of starting with around 12 engines and scaling up to 36 units. However, the article also emphasized that details on the final configuration and the actual full size have not been officially confirmed, but are derived from a chain of circumstantial evidence.
The bottom line was this:
The potential advantage for Nebius lies less in new technology and more in a structural time advantage. If Nebius makes large amounts of power available quickly in Vineland, this could circumvent grid connection bottlenecks and deliver AI capacity sooner - and in a market where "time to power" is increasingly determining competitive advantage, this would be a strategically relevant lever. Also for other projects.
