SPEAKER_1: Alright, so last episode the key insight was that Vertiv and Eaton showed the AI trade migrating out of semiconductors entirely—into the physical infrastructure keeping data centers alive. Now I want to widen the lens further. Because once you follow the power chain out of the data center building itself, you encounter the broader implications of grid infrastructure on AI expansion. SPEAKER_2: Right, and that's where GE Vernova enters the picture. The claim from analysts at S&P Global Market Intelligence is that the main bottleneck for AI scaling has shifted from compute capacity to the availability of reliable electricity and grid infrastructure. That's a significant reframe. SPEAKER_1: So not just 'data centers need power'—the grid itself becomes the constraint. SPEAKER_2: Exactly. And GE Vernova sits at that layer. It's not one product—it's a strategic initiative to expand grid capacity and modernize infrastructure, including gas turbines for baseload generation and transformers for grid delivery, as reported by CNBC in July 2026. SPEAKER_1: Wait—fastest-growing? Because most people think of GE Vernova as a legacy industrial name. Gas turbines, wind. Not a growth story. SPEAKER_2: [short pause] That's exactly the counterintuitive point. When electricity demand expectations change structurally, an old-line industrial power business gets repriced as a growth asset. The numbers back it up. Analysts project GE Vernova's Electrification segment revenue surging roughly 44% to about $13.9 billion in 2026—overtaking its wind business entirely. SPEAKER_1: And the order flow confirms that's not just a forecast. What did the actual order data show? SPEAKER_2: Full-year 2025 orders grew 34% organically to $59.3 billion. Then Q1 2026 alone came in at about $18.3 billion—up 71% year-on-year. And here's the number that really lands: GE Vernova's Electrification segment booked approximately $2.4 billion in data-center equipment orders in Q1 2026 alone, exceeding its total data-center equipment orders for all of 2025. SPEAKER_1: So one quarter exceeded the entire prior year. That's not a gradual ramp. SPEAKER_2: Not at all. And Reuters reported in April 2026 that GE Vernova raised its full-year revenue and margin forecasts after that sharp rise in orders for gas turbines and grid equipment—explicitly linked to power-hungry AI data centers. Total remaining performance obligations are projected to reach about $174 billion in 2026, up roughly 16% from the prior year. SPEAKER_1: For someone tracking this as an investment signal—what's the backlog picture? Because backlog is the metric that confirms multi-year visibility, not just a good quarter. SPEAKER_2: The Electrification segment backlog alone is forecast to rise about 31% to roughly $45.3 billion in 2026. On the generation side, the gas turbine backlog grew from 62 gigawatts to 83 gigawatts by 2025, and combined with slot reservations it reached 100 gigawatts in Q1 2026—expected to hit at least 110 gigawatts by year-end. Customers are locking in manufacturing capacity years ahead. SPEAKER_1: Mm-hmm. Think of it like the CoWoS dynamic we covered with TSMC—customers reserving scarce capacity before they actually need it, because the lead times are too long to wait. SPEAKER_2: That's the right analogy. And GE Vernova has partnered with Chevron to deliver about 4 gigawatts of gas-turbine power specifically for data centers by 2027. Oil-and-gas firms and grid-equipment providers collaborating to meet near-term baseload needs—that's a new kind of supply-chain relationship. SPEAKER_1: Now, how large is the underlying demand signal? Because the investment case only holds if electricity demand is genuinely structural, not a one-cycle pull-forward. SPEAKER_2: The IEA data is striking here. Global data centers consumed about 415 terawatt-hours in 2024—roughly 1.5% of worldwide electricity use. The IEA projects that demand will roughly double to around 950 terawatt-hours by 2030, highlighting the challenges and opportunities in integrating renewable energy sources with traditional power generation. U.S. data centers alone consumed about 180 terawatt-hours in 2024, with roughly 240 terawatt-hours of additional consumption expected through 2030. SPEAKER_1: And ICF, cited by CNBC, put a number on the broader U.S. picture. SPEAKER_2: They forecast U.S. electricity demand jumping nearly 39% by 2035—driven by AI data centers, reshoring of manufacturing, and grid modernization together. The IEA's own electricity outlook has global demand growing about 3.7% in 2026, the fastest pace in years, with data centers as a primary driver. SPEAKER_1: So what's the pressure point? Because the thesis sounds compelling, but there have to be risks that could break it. SPEAKER_2: [emphasis] Several. Permitting and grid connection delays are real challenges—GE Vernova's Electrification CEO highlighted the need for policy and regulatory frameworks to facilitate faster grid connections to meet AI-driven electricity demand. That timing mismatch between AI buildout cycles and traditional grid planning cycles is a genuine friction. Add permitting delays, potential policy reversals on clean energy incentives, and the possibility that AI power demand grows more slowly than projected. SPEAKER_1: Right—but that grid connection delay is actually a two-sided risk. It slows deployments, but it also extends the duration of the bottleneck. SPEAKER_2: Exactly. If utilities can't connect fast enough, the scarcity persists longer. That's the same logic we applied to HBM—time-to-supply determines whether a constraint is a short squeeze or a multi-year theme. Here, the IEA projects renewables meeting nearly half of incremental data-center power demand through 2030, with natural gas and nuclear filling much of the rest. That supply mix question is what sets up the next episode. SPEAKER_1: The key idea for someone tracking this chain: GE Vernova isn't just a power company that benefits from AI. It's a grid infrastructure company that becomes strategically necessary when electricity supply itself is the binding constraint. SPEAKER_2: And that realization—that reliable baseload power is the next scarce input—is exactly what points toward nuclear and uranium. Once the market accepts that data-center electricity demand is doubling by 2030, the question becomes: which generation sources can deliver firm, around-the-clock power at scale? That's where the rerating moves next. SPEAKER_1: And that nuclear question is worth sitting with for a moment before we move on. Because the IEA's own supply breakdown is telling—renewables cover nearly half of incremental data-center power demand through 2030, but natural gas and nuclear fill most of the rest. That's not a clean-energy-only story. SPEAKER_2: It's not. And the nuclear piece is what sets up the next episode directly. But before we get there, the key idea from this episode is worth landing clearly: GE Vernova isn't just a power company that happens to benefit from AI. It's a grid infrastructure company that becomes strategically necessary when electricity supply itself is the binding constraint. SPEAKER_1: Right—and that distinction matters for how someone tracking this series should think about it. It's not 'AI is good for energy stocks.' It's 'the grid is now a bottleneck in the same way CoWoS packaging was a bottleneck for TSMC.' SPEAKER_2: [emphasis] Exactly that. Think of it this way: when Nvidia's H100 demand exceeded TSMC's CoWoS capacity, the constraint migrated from chip design to manufacturing. Now the constraint has migrated again—from the data center building to the transmission line connecting it to the grid. GE Vernova sits at that transmission layer. SPEAKER_1: And the grid connection delay point from GE Vernova's Electrification CEO is the clearest evidence of that. Hyperscale cloud providers demanding connections faster than utilities can deliver—that's not a temporary friction. SPEAKER_2: It's structural. Traditional grid planning cycles run on decade-long timelines. AI buildout cycles run on eighteen-month timelines. Those two rhythms don't match, and the gap between them is where the bottleneck lives. That mismatch is also why backlog is the right metric here—not quarterly revenue. SPEAKER_1: Mm-hmm. So for someone tracking this, the $45.3 billion Electrification backlog forecast for 2026 isn't just a big number. It's evidence that customers are contracting years ahead because they can't afford to wait. SPEAKER_2: Right. And the gas turbine side tells the same story. Combined backlog and slot reservations reached 100 gigawatts in Q1 2026, expected to hit at least 110 gigawatts by year-end. GE Vernova partnered with Chevron to deliver about 4 gigawatts of gas-turbine power specifically for data centers by 2027. Customers are reserving capacity before they need it. SPEAKER_1: Wait—Chevron. So an oil-and-gas company is now part of the AI power supply chain. SPEAKER_2: [short pause] That's one of the more striking supply-chain relationships to emerge from this cycle. It signals that near-term baseload needs are pulling in partners from outside the traditional tech ecosystem. The constraint is real enough that it's reorganizing industry relationships. SPEAKER_1: Now, the risks. Because the order book looks strong, but there are genuine pressure points that could slow this. SPEAKER_2: Several worth naming. Permitting delays are the most immediate—grid connection timelines are already a friction point. Policy reversals on clean energy incentives could shift the economics of certain projects. Supply-chain constraints on specialized equipment like large transformers, which have their own long lead times. And the scenario where AI power demand grows more slowly than projected—if hyperscaler capex pulls back, order books thin. SPEAKER_1: But here's the counterintuitive piece on the delay risk. If utilities can't connect fast enough, the scarcity actually persists longer. The bottleneck extends its own duration. SPEAKER_2: That's the same logic we applied to HBM. Time-to-supply determines whether a constraint is a short squeeze or a multi-year theme. Here, building new transmission infrastructure takes years—sometimes a decade with permitting. So the delay risk and the investment thesis are almost the same fact viewed from two angles. SPEAKER_1: The IEA numbers make that duration argument concrete. Global data-center electricity demand doubling from roughly 485 terawatt-hours in 2025 to around 950 by 2030. U.S. demand adding about 240 terawatt-hours on top of 2024 levels. ICF forecasting nearly 39% growth in total U.S. electricity demand by 2035. SPEAKER_2: And global electricity consumption growing at close to 4% annually through 2025 to 2027—the IEA describes that as the fastest pace in years. Data centers are a primary driver. Now, the takeaway for someone following this chain: the rerating at the grid layer isn't a peripheral utility story. It's a direct extension of the same bottleneck logic that started with Nvidia. SPEAKER_1: Chips, foundries, memory, data-center infrastructure, and now the grid itself. Each layer rerated when capital recognized it as the next binding constraint. SPEAKER_2: And the next question follows naturally. Once the market accepts that data-center electricity demand is doubling by 2030, investors start asking: which generation sources can deliver firm, around-the-clock power at scale? Renewables are intermittent. Gas helps near-term. But the source that's drawing the most attention for reliable baseload is nuclear—and that's where the uranium rerating begins. That's the next episode.