And the first thing to understand about that market is that it already has two prices: the one published in earnings calls, and the one that clears between counterparties who hold positions. If you only know the first price, you are the retail side of this trade.
The market that already trades
The public numbers first, because they frame the board. GE Vernova's gas power equipment backlog grew from 33 GW at the end of 2024 to 40 GW at the end of 2025, and slot reservation agreements, where customers pay upfront to hold future production positions, grew from 29 GW to 43 GW over the same period, with management guiding to roughly 100 GW under contract during 2026 and signaling reservations effectively sold out toward the end of the decade. Siemens Energy and Mitsubishi report the same shape: record backlogs, order books dominated by data-center demand, capacity expansions measured in years.
Now the prices, because the prices tell you what the backlog charts cannot. The figures that follow are Oculus market data, drawn from our own channel coverage and transaction flow rather than from any published source. New equipment traded at roughly $1.2 million per MW at the end of 2025. Halfway through 2026, if you are able to get blessed by one of the major power equipment manufacturers for new allocation at all, the price is over $2 million per MW. Used and zeroed-out rebuilt equipment clears at roughly $1.4 million per MW. Everyone is moving to behind-the-meter power at once, and the market reflects it: a two-thirds price move on new equipment in six months, with access itself becoming the gating term before price is even discussed.
Source: Oculus market data.
Sit with the spread for a moment, because it contains this article's forward call. Used and rebuilt equipment at $1.4 million per MW delivers years before new equipment at $2 million-plus, and yet it trades at a thirty percent discount. That spread says the market has repriced scarcity but has not yet fully priced time. When it does, and the delay mathematics laid out in The Queue Is the Product guarantee that it will, expect existing-equipment prices to converge toward new-equipment parity and, for immediately deployable configurations, through it.
The cheapest thing in this market right now is the thing that already exists, and that is a mispricing with a countdown on it.
The coverage has not caught up to the rest either. A slot reservation is a forward instrument: paid time-priority on future delivery. Forward instruments with time value trade, and these already do. Reservations and positions change hands today through assignments, novations, project sales, and structures built for exactly this purpose, in a market that is bilateral, relationship-gated, and invisible to anyone reading only OEM disclosures. More pointed still: the accumulation phase is well underway. Buyers are amassing reservations by the hundred, not the handful, across OEMs and frame classes, and the intent is not deployment. It is control of downstream supply. When delivery slots are the scarcest input in the largest infrastructure buildout in history, a large enough reservation book is market power over other people's deployment schedules. The OEMs, focused on margin recovery after two brutal decades, have effectively become printing presses for other parties' option books, and not all of them have noticed.
Three consequences follow for anyone planning capacity. First, published lead times are simultaneously too pessimistic and too optimistic: too pessimistic because positions can be acquired from holders at a price, too optimistic because the uncommitted slots the guidance implies are thinner than they look once accumulated books are netted out. The tape you can see is not the market. Second, expect the OEMs to respond the way every manufacturer facing a cornered order book eventually does, with assignment restrictions, use-it-or-lose-it milestones, and buyer qualification, the same tools aircraft makers deployed against speculative skyline positions. When that tightening lands, and 12 to 24 months is the reasonable window, grandfathered transferable positions become the premium instrument in the entire market, which is one more instance of the regulatory-vintage logic laid out in Behind the Meter Goes Mainstream: the paper written before the rules settle is the paper worth holding. Third, the endgame has a precedent with a ticker symbol. Aviation produced the leasing giants because airframes were capital-intensive, mobile, long-lived, and chronically misallocated between those who held delivery positions and those who needed lift. Every one of those conditions now describes gas turbines. The AerCap of turbines is being assembled right now, position by position, and in five years the industry will talk about turbine lessors as if they were always inevitable.
Lead times, costs, and why the shortage outlasts the headlines
From the buyer's side, the story reads in years and percentages. BloombergNEF puts new combined-cycle plant lead times at five years, up from three and a half in 2023, with costs up 49% over the same window. Waits on large frames like Mitsubishi's M501JAC have run as long as seven years against a historical norm of one to three. Wood Mackenzie has described turbine manufacturing as among the most acute bottlenecks in meeting data center demand, with production falling short of the capacity committed by major US utilities alone, before a single hyperscaler order is counted.
The structural point most analyses miss: this shortage is policy, not accident. The manufacturers carry institutional scar tissue from the early-2000s turbine bust, when they built into a demand spike and spent a decade eating the overhang. They are expanding deliberately, prioritizing margin over volume, and saying so on earnings calls. That discipline means the gap does not close on the demand side's preferred schedule under any scenario short of an AI investment collapse. Plan on the shortage as a market condition, not a market moment.
The heat-rate fallacy
Every internal debate about fast-deployment generation runs into the same objection: aeroderivative and smaller machines burn more fuel per megawatt-hour than a modern combined-cycle frame. True, and it is the correct objection for the wrong decade. When the megawatt feeds a grid selling commodity electricity, heat rate is the business. When the megawatt feeds GPUs, the fuel penalty is a rounding error against the compute revenue and the cost of delay quantified in The Queue Is the Product, where a single month of delay prices out at roughly the cost of the machine itself. Run the number for a facility you are actually planning: the efficiency spread between machine classes, priced at delivered gas, against the value of energizing eight quarters earlier. Organizations still optimizing heat rate in 2026 are optimizing the metric of the market they used to operate in.
What actually clears inside a deployment window
Three sources of capacity clear in months rather than years, and they overlap: aeroderivative units, secondary-market fleets, and relocation of existing machines. Aeroderivatives win the window on physics and logistics: modular, road-transportable, fast-starting, installable in a fraction of the civil-works footprint a heavy frame demands, with an aviation design lineage that means they were built from the first drawing to be moved, serviced, and redeployed. That mobility is not a convenience feature. As Bridge to Permanent argues in full, it is the property that converts a bridge asset into an option rather than a monument.
Secondary-market and relocated units carry the most underpriced advantage in the market, and now you have the number: they exist, and they clear at $1.4 million per MW while new allocation, for those who can get it, runs past $2 million. A machine with a serial number, an inspection record, and a transport plan is a categorically different asset from a reservation, and in a market where reservations themselves now trade at a premium, the existing machine is the scarcest instrument of all. The competence that unlocks it is unglamorous and decisive: logistics, certification, serialization, service-agreement transferability. In this cycle, freight forwarding and OEM paperwork are alpha.
The window has a closing date
Manufacturers have announced production ramps that begin to matter in the back half of the decade. Take them at their word directionally, and the acute arbitrage between machines-that-exist and machines-on-order is roughly a 36-month window. Inside it, the correct posture is a physical trader's: secure inventory, manage the basis between what you hold and what you need, treat logistics as the edge, and remember that positions, not promises, are what the tightening will grandfather. Outside the window, this becomes a normal market again, and normal markets do not pay for courage.
The exception, stated without hedging: if your horizon is genuinely 2031 and beyond, your site is fixed, and your load is permanent baseload, the heavy frame is the right machine and the reservation is the right instrument. Reserve it, hold it, and treat it as the tradeable position it is. Then solve the six years in between, because as the energization gap data shows, your competitors are solving them right now.