STRANDED GAS – PRACTICAL OPTION TO AI DATA CENTER UTILITY CONSTRAINS

June 10, 2025

The demand for electricity to power the AI (Artificial Intelligence) data center expansion is estimated to triple by 2030 to 600 TWh in the US. This demand is exacerbated by the unique power demand of the AI GPUs (Graphic Processing Units). Traditional data centers mainly use the conditioned floor space for storing massive amounts of data. The AI data centers draw on the GPUs in parallel to crunch real-time challenges to “deep learning solutions”. 

A traditional data center rack utilized for storing data is an average of 8 kW of power. However, a rack utilized by AI-based GPU problem solving may require an average of 55 kW of power. Legacy data centers utilized for storing data had relatively flat power requirement profiles for memory-based loads. However, AI data centers with GPUs can vary from idle activity to 8X for crunching real-time solutions when called upon. This results in very cyclical power demands. If it is not mitigated by downstream power conditioning systems servicing the data center, this can place unusual load demands on an electric utility grid. Huge AI data center transient power spikes can lead to voltage drops on the utility line servicing other clients. 

The SynSel solution is to de-couple from electric and gas grid utility constraints. This is done by working closely with an ally that develops an oil & gas field. Often there are 600 cubic feet of natural gas associated with a barrel of oil. Pending site demographics, this “stranded gas” has no place to go since the gas transmission pipelines are not in place. The traditional means to stranded gas was to flare it off until a permanent solution could be configured – such as localized gas-to-electric generators or cleaning and compressing the gas. 

A more disruptive and radical approach is to utilize the stranded gas at its source to provide the energy reserve to power AI data centers. This removes the constraints and burden that is otherwise put on the electric and gas grid utility and its customers. However, it can take 4 years to bring an AI data center online due to construction and equipment availability and supply chain realities. During those 4 years, the early “wellhead” gas can be utilized by the Bitcoin miners – they don’t need the refinements of the AI data centers. Bitcoin miners need raw and cheap but reliable power, and they don’t require any fiber, microwave or satellite links for telecommunications – AI data centers do require this connectivity. 

SynSel is working with a very capable oil & gas developer and complimentary engineering, construction and equipment stakeholders to bridge this 4-year transition from the Bitcoin miner to the full-service/capability AI data center user.