Boeing Engines Are Now Data Center Infrastructure
The New York Times reported, and TechCrunch confirmed on November 3, that xAI has installed eight converted jet engine turbines in Mississippi to help power Elon Musk's Colossus 2 data center across the border in Memphis. The units come from ProEnergy, which rebuilds them from a GE engine commonly used on Boeing jets. Crusoe, the Denver startup building OpenAI's Stargate campus in Abilene, Texas, ordered 13 of the same converted units in April. OpenAI had already followed xAI's lead on gas turbines at that site back in 2025.
This is not a fringe experiment by one aggressive operator. Rystad Energy analyst Audun Martinsen told the Times the logic is straightforward: "It's all about getting power as fast as possible," and that operators using converted turbines "can have a data center running in 24 months." When three of the most aggressive AI infrastructure buyers in the market independently land on the same workaround, that is a signal about the state of grid capacity, not a quirky engineering choice specific to one company.
The Grid Timeline Is the Real Constraint
The International Energy Agency's reference point is stark: connecting just 100 megawatts to the grid in Northern Virginia, the densest data center market in the world, takes roughly seven years. That timeline has not meaningfully improved despite years of utility and regulatory attention, because transmission buildout, equipment lead times, and interconnection studies all move slower than compute demand. Mitsubishi Power, a major turbine manufacturer, told Bloomberg on October 27 that most of its new gas turbine capacity will not ship until 2031.
Microsoft gave the clearest real-world proof point of what this means in practice. In October, it withdrew as anchor customer of a planned Nscale data center in Loughton, England, after the local utility said it could not supply enough power until the early to mid 2030s. If Microsoft, with its scale and negotiating leverage, cannot secure a faster timeline from a utility, smaller enterprise buyers leasing colocation space should assume their own stated delivery dates carry real risk of slipping by years, not months.
Why Operators Are Choosing Worse Technology on Purpose
Converted jet engines are a known downgrade from purpose-built gas turbines: they are less efficient, burn more fuel per megawatt produced, and need more units to hit the same output. Operators are choosing them anyway because the alternative, waiting for a grid connection or a new-build turbine, can cost two to five years of delay in a market where every quarter of idle capacity has an immediate revenue and competitive cost. Crusoe's own path illustrates the pressure: in September it walked away from a $1.25 billion deal for 29 new gas turbines from Boom Supersonic, but it still plans to rely on turbines generally, just sourced faster.
The U.S. administration has also encouraged data centers to generate their own electricity rather than wait on utilities, which removes a regulatory brake that might otherwise have slowed this trend and makes self-generation an explicitly sanctioned path rather than a gray-area workaround. Analysts expect efficiency and emissions concerns to grow louder within a year or two as more of these installations become visible to regulators, air-quality agencies, and neighboring communities already contesting data center siting elsewhere, which means today's fast, dirty workaround could become tomorrow's permitting fight, retrofit mandate, or stranded asset.
What This Means for Your Capacity Contract
If you are an enterprise buyer with a multi-year colocation or dedicated capacity agreement, the operative question is whether your provider already holds a signed interconnection agreement or a turbine order with a confirmed delivery slot, not merely whether they say they will deliver power by a given date. A construction timeline without a secured power source is only a schedule, and the gap between a schedule and a commitment has widened enough that Microsoft itself got burned by it this fall when a utility partner's delivery date slipped by most of a decade.
Ask your provider directly which of the two paths they are on: utility interconnection, which the IEA pegs at five to seven years in constrained markets, or self-generation via turbines, which compresses the timeline to roughly 24 months but carries emissions, fuel supply, and potential local permitting risk that could resurface as a compliance or reputational issue later. Either answer changes your own downstream commitments to customers who are counting on your capacity roadmap.
The Decision This Forces on Site and Vendor Selection
For any technology leader evaluating where to place new AI workloads, power availability has become a harder constraint than chip availability, which is a reversal from eighteen months ago when GPU scarcity dominated every planning conversation. A provider in a region with an available grid connection or a self-generation plan already in motion is now a meaningfully safer bet than one quoting a lower price in a power-constrained market, even if the lower price looks attractive on paper today.
This also argues for building power contingency into vendor diversification strategy the same way you already diversify across cloud regions for resilience against outages. A single-region, single-provider AI infrastructure commitment now carries real power-delay risk that a multi-region or multi-vendor posture can partially hedge, because a delay affecting one utility territory or one turbine supplier is unlikely to hit every region you operate in simultaneously. Ask prospective vendors for their grid connection status and backup generation plan as a standard part of due diligence, the same way you already ask about uptime history and disaster recovery.
The Bottom Line for Technology Leaders
Three separate, well-funded AI infrastructure operators, xAI, OpenAI, and Crusoe, independently turned to retrofitted jet engines this year because the conventional path, utility grid connection or new turbine orders, is backed up for years. That is a clearer signal about the actual state of data center power than any capex announcement, because it shows what sophisticated buyers with essentially unlimited capital do when money is not the constraint and time is. Even Microsoft, with more negotiating leverage than almost any buyer on earth, could not talk its way around a utility's multi-year delay.
Before signing or renewing any capacity commitment that assumes a specific go-live date, get your provider's power source in writing, not just their construction schedule. The providers willing to show you a signed interconnection agreement or a confirmed turbine delivery slot are the ones whose timelines you can actually plan around, and the ones who cannot are telling you something important by their silence. Build that question into your next vendor review, and treat a vague answer as a red flag worth escalating before you sign.
