Post 082 documents ERCOT Batch Zero (PGRR145), approved June 2. The framework gates 238GW of large-load interconnection requests through a single system-wide batch study. Study results and capacity offers are expected around Q1 2027, with projects given 60 days to accept. The PUC rule on refundability (16 TAC section 25.194) is expected in August or September 2026. The queue is now a formal gated process with published deadlines and political cost attached.
Two Texas actions in this window reframed the data center buildout from a question of speed to a question of cost. ERCOT adopted the rule that gates large-load interconnection. The governor then ordered regulators to make data centers fully fund their own infrastructure and to cut residential transmission costs. The new constraint is the politics of who pays.
The signals around the crisis that started this story are shifting. A possible deal on the Strait of Hormuz has moved oil. But the AI buildout hasn't waited to see how it resolves. It has detached from the public systems that used to govern it and is assembling a power system of its own.
Wood Mackenzie and others now show that of ~12 GW of US data center capacity expected to come online in 2026, only about 5 GW is under active construction. Close to half of the planned 2026 pipeline is delayed or canceled - driven primarily by the unavailability of transformers, switchgear, and batteries.
The Big Four hyperscalers commit $630 billion to 2026 capex, a 62% surge, while signing a White House pledge to fund both new generation and all grid infrastructure upgrades required to connect their loads, eliminating the transmission bottleneck as political constraint.
Texas interconnection queue now tracks 410 GW of large-load requests, 87% from data centers, a 4.7x multiple of current peak demand. SB-6 rulemaking will determine whether projects connect to the grid or self-generate behind-the-meter.
The ceasefire moved in days, oil moved in hours, and hyperscaler money moved in commitments. The physical system barely moved at all. Turbines, transformers, LNG trains, and grid connections were already the binding constraint; the blockade and the $630B hyperscaler pledge simply made that constraint visible to everyone at once.
Microsoft and Chevron announced a $7 billion exclusive deal to build a dedicated 2,500 MW natural gas power plant near Pecos, Texas, marking the first time an oil supermajor is developing infrastructure directly for a tech hyperscaler's AI operations.
Texas's interconnection queue has quadrupled in a single year, forcing grid operators to overhaul planning processes while regulators push large loads toward behind-the-meter solutions.
Behind-the-meter gas generation for data centers accelerated to 56 GW across 46 projects in 2025, tripling new facility proposals and shortening project timelines to 3 years versus 5+ years for grid interconnection. The trend may break the 2002 record for annual gas power additions.
A two-week ceasefire between the US and Iran briefly halted oil's upward march, but Europe's fuel shortages have already materialized. Meanwhile, Chevron's $7 billion commitment to build Microsoft a dedicated gas power plant signals that oil's future lies in feeding data centers, not traditional grids.
The SPEED Act, passed by the House in December 2025, accelerates federal environmental reviews for energy infrastructure but faces a difficult Senate path requiring 60 votes. Permitting reform remains the binding constraint on gas turbine, nuclear, and transmission buildout timelines.
Permitting reform (SPEED Act, NEPA), Texas grid-cost rebalancing, NRC environmental-impact-statement pilots for AP1000 reactors, and the December 2025 federal AI executive order set the regulatory choreography that determines whether the AI buildout actually meets its capital pace.
Hitachi's HMAX cuts transformer failures 50 percent and repair costs 75 percent; NVIDIA / Emerald AI ramps GPU load to grid signals within seconds to unlock 100 GW of flexible US capacity; AI is moving from a passive load to an active grid asset.
CERAWeek 2026's hardest number was 226 GW of hyperscaler interconnection requests against ERCOT's 85.5 GW historical peak, with a projected 9-18 GW US power shortage by 2027 setting the binding constraint for AI deployment.
Yergin, Wright, Burgum, Wirth, Sawan, Lance, Pouyanne, Hollub, Zamarin, and ERCOT's Blevins together drew the through-line of CERAWeek 2026: energy security has displaced energy transition as the operative frame, and the AI race is fusing the energy and technology industries.
CERAWeek, the energy industry's flagship annual conference (Houston, by S&P Global), made one thing visible at its 2026 edition: AI is no longer only a technology-sector story. It has become a power generation story. The companies racing to scale AI are now running into the familiar constraints of the energy sector: dispatchable capacity, interconnection, permitting, fuel security, equipment lead times, transformers, turbines, and local project delivery.
Post 082 documents ERCOT Batch Zero (PGRR145), approved June 2. The framework gates 238GW of large-load interconnection requests through a single system-wide batch study. Study results and capacity offers are expected around Q1 2027, with projects given 60 days to accept. The PUC rule on refundability (16 TAC section 25.194) is expected in August or September 2026. The queue is now a formal gated process with published deadlines and political cost attached.
Two Texas actions in this window reframed the data center buildout from a question of speed to a question of cost. ERCOT adopted the rule that gates large-load interconnection. The governor then ordered regulators to make data centers fully fund their own infrastructure and to cut residential transmission costs. The new constraint is the politics of who pays.
The signals around the crisis that started this story are shifting. A possible deal on the Strait of Hormuz has moved oil. But the AI buildout hasn't waited to see how it resolves. It has detached from the public systems that used to govern it and is assembling a power system of its own.
Wood Mackenzie and others now show that of ~12 GW of US data center capacity expected to come online in 2026, only about 5 GW is under active construction. Close to half of the planned 2026 pipeline is delayed or canceled - driven primarily by the unavailability of transformers, switchgear, and batteries.
The Big Four hyperscalers commit $630 billion to 2026 capex, a 62% surge, while signing a White House pledge to fund both new generation and all grid infrastructure upgrades required to connect their loads, eliminating the transmission bottleneck as political constraint.
Texas interconnection queue now tracks 410 GW of large-load requests, 87% from data centers, a 4.7x multiple of current peak demand. SB-6 rulemaking will determine whether projects connect to the grid or self-generate behind-the-meter.
The ceasefire moved in days, oil moved in hours, and hyperscaler money moved in commitments. The physical system barely moved at all. Turbines, transformers, LNG trains, and grid connections were already the binding constraint; the blockade and the $630B hyperscaler pledge simply made that constraint visible to everyone at once.
Microsoft and Chevron announced a $7 billion exclusive deal to build a dedicated 2,500 MW natural gas power plant near Pecos, Texas, marking the first time an oil supermajor is developing infrastructure directly for a tech hyperscaler's AI operations.
Texas's interconnection queue has quadrupled in a single year, forcing grid operators to overhaul planning processes while regulators push large loads toward behind-the-meter solutions.
Behind-the-meter gas generation for data centers accelerated to 56 GW across 46 projects in 2025, tripling new facility proposals and shortening project timelines to 3 years versus 5+ years for grid interconnection. The trend may break the 2002 record for annual gas power additions.
A two-week ceasefire between the US and Iran briefly halted oil's upward march, but Europe's fuel shortages have already materialized. Meanwhile, Chevron's $7 billion commitment to build Microsoft a dedicated gas power plant signals that oil's future lies in feeding data centers, not traditional grids.
The SPEED Act, passed by the House in December 2025, accelerates federal environmental reviews for energy infrastructure but faces a difficult Senate path requiring 60 votes. Permitting reform remains the binding constraint on gas turbine, nuclear, and transmission buildout timelines.
Permitting reform (SPEED Act, NEPA), Texas grid-cost rebalancing, NRC environmental-impact-statement pilots for AP1000 reactors, and the December 2025 federal AI executive order set the regulatory choreography that determines whether the AI buildout actually meets its capital pace.
Hitachi's HMAX cuts transformer failures 50 percent and repair costs 75 percent; NVIDIA / Emerald AI ramps GPU load to grid signals within seconds to unlock 100 GW of flexible US capacity; AI is moving from a passive load to an active grid asset.
CERAWeek 2026's hardest number was 226 GW of hyperscaler interconnection requests against ERCOT's 85.5 GW historical peak, with a projected 9-18 GW US power shortage by 2027 setting the binding constraint for AI deployment.
Yergin, Wright, Burgum, Wirth, Sawan, Lance, Pouyanne, Hollub, Zamarin, and ERCOT's Blevins together drew the through-line of CERAWeek 2026: energy security has displaced energy transition as the operative frame, and the AI race is fusing the energy and technology industries.
CERAWeek, the energy industry's flagship annual conference (Houston, by S&P Global), made one thing visible at its 2026 edition: AI is no longer only a technology-sector story. It has become a power generation story. The companies racing to scale AI are now running into the familiar constraints of the energy sector: dispatchable capacity, interconnection, permitting, fuel security, equipment lead times, transformers, turbines, and local project delivery.