Around 2,500 GW of projects are sitting in connection queues, more than three years of record additions. The constraint has moved from the cost of generation to the right to deliver it, and the biggest industrial users have started buying their way out.
Global renewable capacity grew by a record 693 GW in 2025, taking the installed total to about 5.15 TW. The International Renewable Energy Agency now expects 2026 additions to fall, which would end five consecutive years of growth, and it is not because solar got more expensive.
The number that explains it is the queue. Roughly 2,500 GW of advanced-stage renewable, storage and large-load projects are waiting for grid connections worldwide. That is more than three years of record-breaking construction, already developed, waiting for permission to deliver.
The constraint has moved. For fifteen years the binding question was what a megawatt-hour cost to generate. It is now what it costs to be allowed to deliver one, and that is a different kind of problem with a different kind of answer.
Barzel's 1974 theory of rationing by waiting is the cleanest way to see what an interconnection queue actually is. When a scarce thing is not allocated by price, it gets allocated by waiting, and waiting is not free. It is a real cost, paid in capital sitting idle, options held open and staff retained. Barzel's central result is uncomfortable: that cost is largely dissipated rather than transferred. Nobody receives it. It is burned.
Two further implications follow, and both are visible in the data.
The first is that queues sort by who can afford to wait, not by who would use the resource best. A developer with cheap patient capital and low carrying costs outlasts one with expensive money and a construction crew on retainer, regardless of which project would deliver more electricity to more people.
The second is that when entering the queue is cheap, people enter speculatively, which lengthens it for everyone. The United States shows this exactly. Connection requests stood at 2,060 GW at the end of 2025, down from a 2,290 GW peak, while only 549 GW had draft or signed agreements. Roughly a quarter of what is in the queue has a real agreement behind it. The rest is occupying space in a line, and the median time from request to commercial operation now exceeds five years.
Europe is in the same position with different numbers: over 340 GW holding connection agreements and roughly 700 GW still in queues.
The waiting is only half the waste. At the far end of the same network, electricity that did get connected is being discarded because the wires cannot move it. Germany, France and the Netherlands curtailed a record 3.9 TWh in 2025, up 21 per cent. British wind curtailment reached 10 TWh in 2025, against about 3.8 TWh in 2020.
So the system now has projects that cannot connect and generation that connects and is then told to stop. Both are symptoms of one shortage, and the shortage is priced at about $902 billion a year of grid investment needed through 2030 against roughly $450 billion actually spent on transmission and distribution in 2025. Half.
IRENA's own estimate is that grid-enhancing technologies and flexible connection arrangements could unlock 1.2 to 1.6 TW of advanced-stage projects without laying proportionate new line. That is a very large number relative to the gap, and it is mostly software, sensors, contract terms and dynamic ratings rather than steel. It is also the kind of thing that no single developer can procure and no single utility is rewarded for installing.
Faced with a queue, a sufficiently large user stops queueing.
At the Kamoa-Kakula copper complex in the Democratic Republic of the Congo, CrossBoundary Energy has brought online 233 MWp of solar paired with 526 MWh of battery storage, delivering 30 MW of continuous baseload power under a long-term agreement with the mine. The facility is described as Africa's first solar-plus-storage baseload plant, and a second installation takes the combined programme to roughly 433 MWp of solar and 1,107 MWh of storage supplying around 60 MW continuously.
Read against the queue, this is not primarily a decarbonisation story. It is a procurement decision by a user whose cost of waiting is extremely high. A copper concentrator that stops because the grid dipped loses production that cannot be made up, so reliability is worth more to it than to almost any other customer. It therefore has both the motive and the balance sheet to exit the common network and build a private one, and the long-term power purchase agreement is the instrument that lets a third party finance it.
Barzel's framework says what to expect next. The users who leave a rationed queue first are those with the highest cost of waiting, which is also to say the ones with the most valuable use of the resource. The queue that remains is composed of the users who could least afford to leave.
The policy reflex of the past decade has been to procure more generation, because generation was the expensive, slow, politically difficult part. It is now the cheap, fast, divisible part, and the scarce, slow, lumpy part is the network. Continuing to optimise the element that is no longer binding produces exactly what the figures show: record additions, record queues and record curtailment at the same time.
Three consequences follow for anyone allocating capital.
Auction design should price delivery, not capacity. A procurement that awards contracts to projects without a connection path is manufacturing queue, and the queue is where the value burns.
Flexibility is the cheapest capacity in the system right now. Storage additions already reached about 112 GW and 307 GWh in 2025, 48 per cent above 2024, and grid-enhancing technology is cheaper still per unlocked megawatt. Neither has a constituency the way a new solar farm does.
Private networks will keep being built, and the question is who is left behind. Each large user that exits removes a payer from the common network without removing its fixed costs. That is a tariff problem for everyone remaining, and it arrives slowly enough that nobody has to decide about it until it is already done.
Three things would turn this from an argument into evidence, and all three are obtainable.
What the queue actually costs a developer per year of waiting. Carrying cost on secured land, grid studies, options and retained staff, for a representative mid-size project. Nobody publishes it and every developer knows it.
What share of a queue is speculative, measured rather than estimated. The US agreement-to-request ratio implies it. Interviews with queue managers in three jurisdictions would establish it.
What the smallest financeable private baseload offtaker looks like. Ask the lenders behind Kamoa-Kakula and comparable deals what covenant they needed, then compare that against the load profile of a mid-sized processor. That number defines who can leave the queue and who cannot.
For regulators, the conclusion is that queue design is now energy policy, and an unpriced queue is a policy that selects for patient capital. For developers, it is that a connection path is the asset and the panels are a commodity. For industrial users, it is that the calculation Kamoa-Kakula made is available further down the size curve than most assume, and the binding constraint on it is the credit agreement rather than the engineering.
If this reads close to a decision you are weighing, we would like to hear about it. A short note is enough to start.
The Lab is an independent research team, not an advocacy organisation. We do not campaign, endorse products, or accept commissions where a finding is expected in advance. Where a partner's own theory of change is what needs testing, that is a legitimate engagement, and the Lab tests it honestly.