-
-
-
Pillar
Asset Class
Date
Date
Purview. This report covers the financing of gas-fired, on-site, and behind-the-meter power generation built specifically to serve AI data center load in the United States and the securitized and structured credit markets that have grown around it. It does not cover renewable PPA structures, nuclear SMR financing, or grid-scale transmission investment, each of which we treat as adjacent but distinct topics.
I. The bottleneck moved from grid to the factory floor
Grid interconnection queues were the first constraint to bind and in some regions the wait from application to energized service now stretches close to four years. Developers have responded by routing around the grid entirely which is where the term Bring-Your-Own-Power or BYOP comes around and is a way to relocate the bottleneck one link down the supply chain, from the utility queue to the gas turbine order book.
The scale of that order book is, in our view, the single most important physical constraint on the AI infrastructure buildout today. A unit ordered from any of the three major OEMs will not arrive before 2028 and for the largest firms, not before 2030 or 2031.
Figure 01: Heavy-duty gas turbine backlog by OEM, Q2 ‘26
Backlog composition differs by OEM — GE Vernova’s figure blends firm orders and paid slot reservations; Siemens Energy reports firm backlog only. Baker Hughes discloses backlog in dollar terms rather than GW, as its Gas Technology Equipment segment includes LNG equipment alongside turbines and is not broken out by capacity.
Purview. This report covers the financing of gas-fired, on-site, and behind-the-meter power generation built specifically to serve AI data center load in the United States and the securitized and structured credit markets that have grown around it. It does not cover renewable PPA structures, nuclear SMR financing, or grid-scale transmission investment, each of which we treat as adjacent but distinct topics.
I. The bottleneck moved from grid to the factory floor
Grid interconnection queues were the first constraint to bind and in some regions the wait from application to energized service now stretches close to four years. Developers have responded by routing around the grid entirely which is where the term Bring-Your-Own-Power or BYOP comes around and is a way to relocate the bottleneck one link down the supply chain, from the utility queue to the gas turbine order book.
The scale of that order book is, in our view, the single most important physical constraint on the AI infrastructure buildout today. A unit ordered from any of the three major OEMs will not arrive before 2028 and for the largest firms, not before 2030 or 2031.
Figure 01: Heavy-duty gas turbine backlog by OEM, Q2 ‘26
Backlog composition differs by OEM — GE Vernova’s figure blends firm orders and paid slot reservations; Siemens Energy reports firm backlog only. Baker Hughes discloses backlog in dollar terms rather than GW, as its Gas Technology Equipment segment includes LNG equipment alongside turbines and is not broken out by capacity.
Purview. This report covers the financing of gas-fired, on-site, and behind-the-meter power generation built specifically to serve AI data center load in the United States and the securitized and structured credit markets that have grown around it. It does not cover renewable PPA structures, nuclear SMR financing, or grid-scale transmission investment, each of which we treat as adjacent but distinct topics.
I. The bottleneck moved from grid to the factory floor
Grid interconnection queues were the first constraint to bind and in some regions the wait from application to energized service now stretches close to four years. Developers have responded by routing around the grid entirely which is where the term Bring-Your-Own-Power or BYOP comes around and is a way to relocate the bottleneck one link down the supply chain, from the utility queue to the gas turbine order book.
The scale of that order book is, in our view, the single most important physical constraint on the AI infrastructure buildout today. A unit ordered from any of the three major OEMs will not arrive before 2028 and for the largest firms, not before 2030 or 2031.
Figure 01: Heavy-duty gas turbine backlog by OEM, Q2 ‘26
Backlog composition differs by OEM — GE Vernova’s figure blends firm orders and paid slot reservations; Siemens Energy reports firm backlog only. Baker Hughes discloses backlog in dollar terms rather than GW, as its Gas Technology Equipment segment includes LNG equipment alongside turbines and is not broken out by capacity.
Purview. This report covers the financing of gas-fired, on-site, and behind-the-meter power generation built specifically to serve AI data center load in the United States and the securitized and structured credit markets that have grown around it. It does not cover renewable PPA structures, nuclear SMR financing, or grid-scale transmission investment, each of which we treat as adjacent but distinct topics.
I. The bottleneck moved from grid to the factory floor
Grid interconnection queues were the first constraint to bind and in some regions the wait from application to energized service now stretches close to four years. Developers have responded by routing around the grid entirely which is where the term Bring-Your-Own-Power or BYOP comes around and is a way to relocate the bottleneck one link down the supply chain, from the utility queue to the gas turbine order book.
The scale of that order book is, in our view, the single most important physical constraint on the AI infrastructure buildout today. A unit ordered from any of the three major OEMs will not arrive before 2028 and for the largest firms, not before 2030 or 2031.
Purview. This report covers the financing of gas-fired, on-site, and behind-the-meter power generation built specifically to serve AI data center load in the United States and the securitized and structured credit markets that have grown around it. It does not cover renewable PPA structures, nuclear SMR financing, or grid-scale transmission investment, each of which we treat as adjacent but distinct topics.
I. The bottleneck moved from grid to the factory floor
Grid interconnection queues were the first constraint to bind and in some regions the wait from application to energized service now stretches close to four years. Developers have responded by routing around the grid entirely which is where the term Bring-Your-Own-Power or BYOP comes around and is a way to relocate the bottleneck one link down the supply chain, from the utility queue to the gas turbine order book.
The scale of that order book is, in our view, the single most important physical constraint on the AI infrastructure buildout today. A unit ordered from any of the three major OEMs will not arrive before 2028 and for the largest firms, not before 2030 or 2031.
Purview. This report covers the financing of gas-fired, on-site, and behind-the-meter power generation built specifically to serve AI data center load in the United States and the securitized and structured credit markets that have grown around it. It does not cover renewable PPA structures, nuclear SMR financing, or grid-scale transmission investment, each of which we treat as adjacent but distinct topics.
I. The bottleneck moved from grid to the factory floor
Grid interconnection queues were the first constraint to bind and in some regions the wait from application to energized service now stretches close to four years. Developers have responded by routing around the grid entirely which is where the term Bring-Your-Own-Power or BYOP comes around and is a way to relocate the bottleneck one link down the supply chain, from the utility queue to the gas turbine order book.
The scale of that order book is, in our view, the single most important physical constraint on the AI infrastructure buildout today. A unit ordered from any of the three major OEMs will not arrive before 2028 and for the largest firms, not before 2030 or 2031.
