The natural gas ‘bridge’ becomes a highway

Episode Summary

Executive Summary: The episode examines the sudden surge of gas-fired generation commitments tied to AI/data center growth, with Meta, Microsoft, Google, and Crusoe all pursuing large new gas builds. The hosts argue gas is being used as a fast, familiar solution, but warn that opaque planning, high equipment costs, and weak load forecasts could raise rates and strand assets unless paired with storage, flexibility, and better cost allocation.

Main Topics: Hyperscalers' dash to gas for AI power (Priority: 5/5): Meta, Microsoft, Google, and Crusoe are all backing major natural gas builds to serve data center demand, reflecting an industry-wide scramble for firm power. How the Meta-Louisiana project was structured (Priority: 5/5): The Meta deal is presented as a large, third-party-financed build that includes generation, transmission, and storage, but still leans heavily on gas for speed and reliability. Rate impacts, stranded costs, and utility planning risk (Priority: 5/5): Jigar Shah argues that utilities may overbuild gas based on uncertain load forecasts, leaving residential customers to absorb costs if utilization is lower than expected. Flexibility, batteries, and the future grid shape (Priority: 4/5): Caroline Golan argues the conversation should move from gas as a bridge fuel to flexibility markets, where storage, demand response, EVs, and virtual power plants matter more. Transparency problems in load queues and project counts (Priority: 4/5): The hosts criticize inflated or speculative load and project announcements, saying regulators need clearer data to separate real projects from hype. Equipment shortages and supply-chain constraints (Priority: 4/5): Turbine scarcity, long lead times, and rising EPC costs are making new gas projects harder and more expensive to execute, likely consolidating the market. Tech-company responsibility vs. utility responsibility (Priority: 4/5): Both hosts argue hyperscalers should not carry the full blame; utilities, suppliers, regulators, and DOE guidance also shape the outcome and should be held accountable.

Key Arguments: Natural gas is becoming the default fast-track solution for AI data centers because existing grid capacity is insufficient and developers want minimal friction and maximum security. Meta's Louisiana project is not unusual in mechanics; what is unusual is the scale of new generation required because the grid could not absorb the load. Jigar Shah contends the reported gas project costs look too low compared with current turbine pricing, suggesting a future repricing or disclosure gap. The real risk is cost socialization: if gas plants run less than planned due to VPPs, storage, and demand flexibility, utilities may spread costs to all ratepayers. Caroline Golan argues the grid is moving toward micro-markets where flexibility, not just firm capacity, will determine the right resource mix. Storage is increasingly relevant both as a transmission solution and because AI training loads need behind-the-meter buffering for highly erratic demand. The tech sector's planning horizon is too short for conventional utility resource planning, which can lead to overbuilding and stranded assets. Utilities in vertically integrated markets can absorb and finance build risk more easily than competitive suppliers, which makes data centers prefer IOU structures even if it shifts risk onto customers. The lack of transparent, credible queue data makes it harder for regulators to assess whether gas build-outs are real, justified, and affordable. DOE policy and federal signaling matter; if policymakers favor gas-only answers, tech companies will follow that guidance even if cheaper alternatives exist.

Data Points: Meta Louisiana gas commitment: 7.5 GW across 10 plants - Tripled from 2.3 GW in under a year for a single AI campus in Louisiana. Meta project cost: nearly $11 billion - Cost cited for the Louisiana build-out. South Dakota comparison: enough capacity to power the entire state - Used to illustrate the scale of Meta's Louisiana gas build. Microsoft Permian project: 2,500 MW initial, expandable to 5,000 MW - Exclusive talks with Chevron and Engine No. 1 for a gas plant in West Texas. Microsoft project cost: $7 billion projected - Estimated cost for the Permian Basin generation project. Google/Crusoe Louisiana-Texas project: 933 MW - Behind-the-meter natural gas plant permit referenced at the Goodnight campus in Texas. Crusoe gas equipment secured: about 4.5 GW - Referenced as Crusoe's growing commitment to gas generation equipment. U.S. operating gas fleet: about 500 GW - Jigar describes current operating natural gas capacity in the U.S. Historic gas share of U.S. mix: about 25% - Used in discussing how gas has grown from a bridge fuel position. Current gas share of U.S. mix: about 43% - Referenced as the present contribution of natural gas to U.S. electricity. Hyperscaler planning horizon: 2-5 years - Caroline says tech companies often plan on short cycles compared with power plant timelines. Typical power plant development timeline: 5-6 years - Contrasted with faster-moving data center load growth. Gas turbine price range mentioned: around $2,800/kW or lower - Jigar questions how Entergy could have secured gas turbines below prevailing market costs. Historical cheaper turbine price: $800/kW - Jigar cites 2021 as a period when Southern Company could buy turbines much more cheaply. Turbine lead times: 6 years - Citing Wood Mackenzie, the hosts note severe manufacturing bottlenecks. Turbine price inflation: 195% since 2019 - Used to show how expensive gas equipment has become. Global turbine order backlog: through 2027 - Wood Mackenzie report referenced in the discussion of shortages. PSE&G planned spending: $18B in 2023, $22B in 2024, $26B in 2025, then $30B expected - Used as an example of rising rate-base needs amid load growth. Georgia Power battery build-out: 600 MW to 6.6 GW - Cited as a counterexample showing a major utility battery expansion. ERCOT peak load current level: about 70-80 GW - Jigar uses this as the starting point for Texas load growth discussion. ERCOT peak load outlook cited: around 120 GW by 2030/2031 - Presented as an ambitious but uncertain forecast. Jigar's bet on ERCOT peak growth: not more than 40 GW increase - He argues many projected loads will not materialize.

Pivotal Quotes: "This is leading to the least optimized grid you could possibly build." — Jigar Shah: He warns that piecemeal gas commitments plus later additions of solar and other resources could produce a costly, inefficient system. "The conversation needs to pivot a little bit or expand a little bit to say, what is the role of natural gas vis-a-vis virtual power in a flexibility conversation?" — Caroline Golan: She argues the grid should be evaluated through flexibility and micro-market signals, not just gas-as-bridge thinking. "What I don't want to see is build out, get pushed completely into the IOU territories where it does get sort of lazily absolved." — Caroline Golan: She cautions against shifting too much development risk and cost into utility rate bases.

Implications: Gas is reasserting itself as the near-term power source for AI growth, but without better transparency, flexibility planning, and cost allocation, utilities and regulators risk higher rates, stranded assets, and a grid built for yesterday's assumptions.