xAI Powers Mississippi Data Center with Eight Modified Jet Engines

A New Approach to Data Center Power: Modified Jet Engines

While most AI companies are still grappling with grid access and clean energy quotas, xAI has chosen a more direct path. According to discussions in the Hacker News community, xAI has installed eight modified jet engines at its Mississippi data center to serve as on-site power generation equipment. Although this information is brief, it points to an industry trend worth watching: AI compute expansion is pushing the power supply issue to the forefront, and leading companies are beginning to address it through unconventional means.

Using jet engines for ground-based power generation is not an entirely new concept. Aeroderivative gas turbines have mature applications in the power generation sector, with advantages including high power density, fast start-stop capability, and short deployment cycles. xAI's choice to use "modified" engines rather than purchasing standard generator sets suggests its approach is closer to rapidly building its own power source and bypassing lengthy grid approval processes, rather than waiting for traditional power infrastructure to expand.

Aeroderivative gas turbine refers to ground-based power generation equipment converted from aircraft engines (such as the CF6 and LM2500 series, derived from military turbofan core engines). Its core engine shares the high-pressure compressor, combustor, and turbine design with the aviation version, but the low-pressure section is replaced with a power turbine to drive the generator. Compared to heavy-duty industrial gas turbines, its unit power typically ranges from 20-60MW, with lighter weight, smaller footprint, and startup times that can be compressed to under 10 minutes, making it suitable for peak shaving and distributed power supply. Drawbacks include slightly lower fuel efficiency, shorter overhaul intervals, and greater sensitivity to fuel quality (such as natural gas purity). In data center scenarios, these units are often used to bridge the transition period when grid capacity is insufficient.

Why Mississippi

Mississippi's role in this matter is noteworthy. Compared to data center-dense regions like California or Virginia, Mississippi's electricity costs, land availability, and regulatory environment are more favorable for large-scale compute facilities. The on-site power generation approach further reduces dependence on the public grid and also cuts the time costs associated with interconnection queuing.

For xAI, the delivery pace of training clusters is directly tied to how quickly power can be brought online. A power generation array composed of eight engines can form considerable power supply capacity in a relatively short period. This combination of "self-built power plant + self-built data center" essentially brings energy infrastructure within the scope of AI companies' construction responsibilities.

Interconnection queue is a key bottleneck in the U.S. power market. Before new large loads (such as data centers) or generation facilities can connect to the grid, they must go through interconnection studies by regional transmission organizations (such as MISO and PJM) to assess impacts on grid power flows, short-circuit capacity, and stability. In recent years, due to the surge in renewable energy and energy storage projects, the queue has exceeded 2000GW, with waiting times in some regions reaching 5-7 years. Behind-the-meter generation bypasses this process: as long as environmental emissions permits and local building codes are satisfied, generation equipment can operate directly on the customer side without involving transmission-level interconnection studies. This is precisely the institutional motivation behind xAI's choice of on-site gas turbines.

Controversies and Concerns

The Hacker News entry did not generate much discussion (Points: 7 | Comments: 1), but the controversy the topic points to is not insignificant. There is tension between the fuel consumption and emissions levels of jet engine power generation and the clean energy goals that tech companies commonly proclaim. Although on-site gas power generation is flexible, carbon intensity and noise issues are difficult to avoid. Against the backdrop of increasingly stringent environmental reviews, such solutions may face public opinion and regulatory pressure.

Another issue is operational complexity. Aviation engines are designed for different operating conditions than continuous ground-based power generation. The reliability, maintenance cycles, and fuel supply chain of modified engines all require dedicated team support. Incorporating power generation equipment into the data center operations system means xAI must simultaneously manage two sets of systems: compute facilities and energy facilities.

Carbon intensity is a metric measuring carbon emissions per unit of electricity generated, typically expressed in gCO2/kWh. The carbon intensity of natural gas combined cycle is approximately 350-400gCO2/kWh, while simple-cycle gas turbines are higher (about 500-600gCO2/kWh). The average grid carbon intensity in regions with high clean energy penetration can be as low as below 200gCO2/kWh. Tech companies commonly commit to 24/7 carbon-free energy matching, and on-site gas power generation directly increases their Scope 1 (direct emissions) carbon footprint. Additionally, the U.S. Environmental Protection Agency has strict limits on NOx emissions from gas turbines. Although Mississippi's regulations are relatively lenient, permits are still required, and large-scale deployment could trigger air quality reviews.

Implications for AI Infrastructure Competition

The signal this move sends is more important than the equipment itself: compute competition is extending upstream to the energy sector. When power becomes the bottleneck for training clusters, whoever can obtain stable power faster can bring models to market faster. Self-built power generation, long-term power purchase agreements, and even siting near power generation resources may all become standard practices for AI companies.

Modified jet engines are just one solution. They may not be the most economical or the cleanest, but they reflect a realistic trade-off at the current stage—when compute demand grows faster than grid expansion, engineering teams tend to fill the gap with rapidly deployable solutions. What is worth watching going forward is whether such on-site power generation facilities will become standard configuration for large data centers, and how regulators will respond.