AI Needs Firm Power in Months, Not Years. Powerships Promise It in 30 Days.

The binding constraint on artificial intelligence is no longer chips or capital. It is the lead time to put firm, dispatchable electricity next to a data center. In the largest United States grid, the wait from interconnection application to commercial operation has risen from under two years in 2008 to more than eight years in 2025, and the three companies that build most large gas turbines are sold out to 2028 and beyond. Against that backdrop, a fifteen-year-old maritime technology can dock at a coastline and deliver power in under 30 days. The AI industry is chasing nuclear restarts and onsite gas. It has so far overlooked the floating option.

Core facts

The demand curve is no longer in dispute. The International Energy Agency's Energy and AI report, published in April 2025, projects global data center electricity consumption roughly doubling to around 945 TWh by 2030, just under 3% of world electricity, with consumption from AI-optimized facilities tripling over the same period. DataCenterDynamics reported the same 945 TWh figure. The IEA puts data center electricity growth at about 15% per year from 2024 to 2030, more than four times the growth rate of all other electricity demand, with the United States adding around 240 TWh (up 130%) and China around 175 TWh (up 170%).

The supply side is where the model breaks. PJM Interconnection's December 2025 base capacity auction, for the 2027/28 delivery year, carried USD 16.4 billion in total capacity costs, of which USD 6.5 billion, or 40%, was attributable to data center demand, according to the grid's independent market monitor, Monitoring Analytics. Roughly USD 6.2 billion of that came from data centers not yet built. The auction cleared at the USD 333.44 per MW-day price cap, a record for the third consecutive auction, as Utility Dive reported. PJM projects peak demand rising 32 GW from 2024 to 2030, with all but 2 GW of that growth coming from data centers.

Generation cannot follow at that speed. Grid-queue data tracked by Lawrence Berkeley National Laboratory shows more than 2,600 GW of proposed generation and storage waiting in United States interconnection queues. In Texas, the Electric Reliability Council of Texas (ERCOT) is processing a large-load queue of roughly 410 GW, of which data centers are about 87%. The equipment is constrained too. Utility Dive reported that GE Vernova expected to end 2025 with an 80 GW gas turbine backlog stretching into 2029, and Climate Tech VC reported that the lead time for a new combined-cycle gas turbine (CCGT) plant rose from 3.5 years in 2023 to five years in 2025, with costs up 49%.

The prior equilibrium, and how it broke

For two decades the data center power model was simple: sign a utility interconnection, draw from the grid, and let the system operator worry about generation. That arrangement assumed load that grew predictably and slowly. AI broke both assumptions at once, and the hyperscalers have responded by reaching for firm power outside the normal procurement path.

Two deals mark the shift. In September 2024, Constellation Energy agreed to restart the 835 MW Unit 1 reactor at Three Mile Island in Pennsylvania, rebranded the Crane Clean Energy Center, under a 20-year power purchase agreement (PPA) selling the full output to Microsoft for its AI data centers, as CNBC and Utility Dive reported. A restart of a shuttered reactor for a single corporate offtaker had no real precedent. Then in October 2025, POWER magazine and DataCenterDynamics reported that VoltaGrid would deploy 2.3 GW of modular natural gas generation for Oracle's AI data centers in Texas, fuel supplied by Energy Transfer pipelines, much of it behind the meter and off the public grid. That deployment sits inside the Stargate program, which Data Center Frontier reported as targeting nearly 7 GW of capacity against more than USD 400 billion of planned investment.

The pattern is consistent. When grid timelines run to eight years and turbine slots to 2030, AI developers will pay for speed and firmness over grid economics. That is the exact value a maritime power plant sells.

The maritime option

Floating generation is not new, and its logic maps onto the AI constraint with unusual precision. A powership is a ship or barge-mounted, fully integrated power plant that connects to onshore substations. An FSRU (floating storage and regasification unit) paired with gas-to-power turns imported liquefied natural gas (LNG) into electricity at the quayside. Both bypass the slowest steps in onshore development: land acquisition, transmission buildout, and the turbine manufacturing queue.

The advantages are concrete. Deployment runs in weeks rather than years. The power is dispatchable, not weather-dependent. The asset is mobile, so it can be unplugged and redeployed where it is most needed. Units are modular, from tens of megawatts to roughly 500 MW, which matches a single data center campus rather than a national grid.

The Karpowership record

Karpowership, the floating-generation arm of Turkey's Karadeniz Holding, is the clearest test of whether this profile is real or marketing. The case for it rests on an operating history, not adjectives.

The record. Karpowership launched its first powership in Iraq in 2010, giving roughly fifteen years of continuous powership operation, according to company and encyclopedic records. (The "30 years" the group sometimes cites refers to the wider holding company, not the powership business.) On installed capacity, the figures diverge by source: Karpowership states on its website that its global fleet exceeds 8,000 MW, with a stated expansion target of 21,000 MW, while independent references including Wikipedia put installed capacity above 7,000 MW across more than 40 vessels. The company says individual units range from 30 MW to 500 MW and can be commissioned in under 30 days, a timeline corroborated by multiple deployments covered in the press. Anadolu Agency reported the fleet powering eight African nations, and the documented country list across company and independent sources spans roughly 20 countries on four continents, including Iraq, Lebanon, Ghana, Senegal, The Gambia, Sierra Leone, Guinea-Bissau, Mozambique, Indonesia, Brazil, the Dominican Republic and Ecuador.

Specific deployments anchor the scale. POWER magazine documented a 470 MW powership serving Ghana, among the largest single floating units on record. In 2024 the company launched what it described as Africa's first LNG-to-power project in Senegal, a move toward a lower-emission fuel than HFO. France 24 reported in 2018 that powerships supplied up to a quarter of Lebanon's electricity, a figure that shows the scale a single operator can reach.

Why the profile fits AI. The match is structural. AI needs firm, dispatchable power on a sub-year timeline in capacity-constrained markets, often near coasts where data center clusters and cable landings already sit. Karpowership's documented strengths, deployment in weeks and dispatchable output in 30 to 500 MW blocks, are the same attributes the Three Mile Island and VoltaGrid deals were bought for. That is the basis for treating it as a credible option.

The public-interest case

A powership does not only serve the AI buyer. When a hyperscaler draws its load from the public grid, the capacity cost is socialized onto everyone: the USD 6.5 billion data center share of PJM's December 2025 auction, identified by Monitoring Analytics, lands on ordinary ratepayers' bills. Self-supplied power at the quayside does the opposite. By taking AI's enormous and fast-growing demand off the public grid, a powership spares ratepayers that surcharge and leaves shared grid capacity for everyone else. The same deployment that gives a hyperscaler firm power in months also keeps the public from absorbing the bill. For the powership sector, which holds exactly this asset yet has no announced AI contract, that alignment of private speed and public interest is the opening.

Why supply is not the bottleneck

Unlike a new grid-scale plant, a powership does not wait in the turbine queue. The fleet already exists, and the marine and modular engines it uses sit outside the grid-scale gas-turbine backlog that stretches to 2030. That is precisely why floating power can deploy in weeks while onshore gas waits years. The fuel path is improving in parallel: LNG and dual-fuel engines are the AI-grade direction, the route Karpowership is already taking, as its Senegal LNG-to-power project and announced fleet conversion work signal.

What to watch

The decisive signal is the first publicly announced contract for a powership or quayside FSRU-to-power plant supplying a named data center operator. The conditions point toward it. If gas-turbine backlogs stay sold out through 2030, as GE Vernova's chief executive told investors he expects by the end of 2026, the premium on fast, firm, floating power rises. The IEA's central projection, that natural gas covers much of AI's near-term supply gap before nuclear scales late in the decade, leaves an open lane for dispatchable maritime gas. The operating record shows the asset is ready to fill it.