Gas turbine prices have surged 300% in three years. GE Vernova is sold out through 2031. Microsoft, Meta, Google, and OpenAI are all fighting for the same finite supply — and the power supercycle is just getting started.
For two years, the AI narrative has been dominated by one thing: chips. NVIDIA's H100s, B200s, the endless scramble for compute. But while everyone was obsessing over GPU supply chains, a far more prosaic bottleneck was quietly tightening — and it makes the chip shortage look like a minor inconvenience.
The real constraint on the AI revolution isn't silicon. It's steel.
Specifically, it's the 30-foot-tall, 400-ton gas turbines manufactured in a handful of factories around the world — mostly in Greenville, South Carolina, at GE Vernova's sprawling 400-acre facility. These machines, each capable of powering 500,000 homes, have become the single most sought-after piece of industrial equipment on the planet. And the numbers tell a story that should make every tech executive, every investor, and every AI optimist sit up and pay attention.
Gas turbine prices have risen roughly 300% over the past three years. According to Wood Mackenzie, prices will hit $600 per kilowatt by the end of 2027 — a 195% increase since 2019 — and there is absolutely no indication the trend will reverse.
Global orders sat at 110 gigawatts at the end of 2025. Global manufacturing capacity? Somewhere between 60 and 70 GW. That's not a supply-demand imbalance. That's a supply-demand chasm.
Let's cut through the noise and look at what's actually happening at GE Vernova, the company sitting at the center of this vortex.
In Q1 2026, GE Vernova booked $18.3 billion in orders, up 71% organically year-over-year. Revenue hit $9.30 billion. The company's Electrification segment — the division that makes transformers, switchgear, and the grid infrastructure connecting data centers to power sources — captured $2.4 billion in data center equipment orders in a single quarter. That's more than the entirety of 2025. In three months, they beat an entire year.
The Gas Power division's combined backlog and slot reservation agreements grew from 83 GW to 100 GW sequentially. Management now targets at least 110 GW by year-end 2026. CEO Scott Strazik told analysts that new orders are "priced 10 to 20 points higher than our Q4 2025 orders on a dollar per kW basis." Translation: the price keeps climbing, and customers keep paying.
Here's what makes this genuinely astonishing: GE Vernova's gas turbine slots are effectively sold out through 2031. At the start of Q1 2026, the company had about 10 GW of available 2029 production capacity. After a single quarter of aggressive customer pull, only about 10 GW remains — across 2029 and 2030 combined. If you want a heavy-duty gas turbine today, you're looking at a three-year lead time, minimum. And that's if you can even get a slot.
The customer list reads like the NASDAQ 100. Microsoft has ordered seven turbines for Project Kilby — a 2.67-gigawatt data center complex in West Texas, backed by a 20-year power purchase agreement with Chevron. That's 2.67 GW for a single campus. To put that in perspective: the entire data center industry consumed about 17 GW across the US in 2022. One Microsoft project is consuming roughly 16% of what the entire industry used just four years ago.
Meta is reportedly in the queue. So is Google. OpenAI and Anthropic — the two companies most responsible for the AI boom itself — are both competing for turbines. When the companies that created the demand can't get the supply they need to sustain it, you know the market is broken.
But here's the twist that most analysis misses: only about 20% of GE Vernova's 100 GW under contract is explicitly tied to data center load. The other 80% is traditional utility, independent power producer, and industrial customers. These are power plants being built for general grid consumption — and they're competing for the exact same manufacturing slots.
The data center industry isn't just buying turbines. It's competing with every utility and industrial customer on the planet for a product that only three major manufacturers can produce at scale.
The gas turbine market is, for all practical purposes, a triopoly. GE Vernova, Siemens Energy, and Mitsubishi Heavy Industries account for more than 70% of global production capacity. Baker Hughes and a few others bring up the rear, but the heavy-duty large-frame turbine market — the kind data centers need — is overwhelmingly concentrated in three hands.
All three are scrambling. GE Vernova is spending $160 million to expand from roughly 50 large-frame turbines per year to 70-80 units by late 2026. The company has installed more than 280 new machines across its factories in 15 months and expects to reach 20 GW of annualized output by Q3 2026, with a target of 24 GW by 2028. It has hired roughly 1,800 US production workers across 2025 and 2026.
Siemens Energy has transitioned key facilities to 24/7 operations and announced a $1 billion US investment program. Mitsubishi Heavy Industries plans to double its manufacturing capacity through 2028.
None of this is happening fast enough. Wood Mackenzie projects turbine orders will peak in 2026 as developers attempt to lock in equipment for 63 GW of gas capacity additions through 2030. The manufacturing capacity simply cannot keep pace.
And there's a deeper, more intractable problem: hot-section component manufacturing. The single-crystal turbine blades that sit in the hottest part of a gas turbine — where temperatures exceed the melting point of the metal itself — can only be produced at scale by a handful of global suppliers. These precision casting processes require decades of institutional knowledge, specialized furnaces, and metallurgical expertise that cannot be scaled up quickly. When every turbine requires dozens of these blades and only a few foundries on Earth can make them, you've got a bottleneck within a bottleneck.
Here's where the energy transition narrative collides head-on with AI reality.
Data center operators are discovering that waiting for grid interconnection is a non-starter. In many US markets, the queue for new grid connections stretches 3-7 years. PJM Interconnection, the largest US grid operator, had over 290 GW in its interconnection queue as of early 2025 — more than the entire existing PJM generation capacity.
So hyperscalers are doing what any rational actor would do in a queue crisis: they're building their own power. Behind-the-meter generation — power plants literally built on or adjacent to data center campuses — has gone from a niche strategy to the default playbook.
This fundamentally changes the economics. A data center operator isn't just buying electricity anymore. They're buying gas turbines, signing fuel supply agreements, and becoming de facto power utilities. The SB Energy Portsmouth project in Virginia is the most extreme example: a $33 billion, 9.2-GW natural gas-fired facility that could require 24 to 30 heavy-duty gas turbines for initial build-out alone. That single project represents roughly 60% of GE Vernova's annual large-frame turbine output at current production levels.
One project. More than half of the world's largest turbine manufacturer's annual capacity.
While gas turbines grab headlines, GE Vernova's fastest-growing business is one most investors still don't understand.
The Electrification segment — transformers, switchgear, substations, high-voltage direct current systems, and grid software — has a $42 billion backlog, up from $9 billion at the end of 2022. Orders essentially doubled year-over-year to $7.1 billion in Q1 2026. The February acquisition of transformer manufacturer Prolec for $5 billion gives GE Vernova full control over one of the industry's most supply-constrained product categories.
Large power transformers currently have multi-year lead times. You cannot build a data center without them. You cannot connect a gas turbine to anything useful without them. And GE Vernova now controls the critical chokepoint from generation (turbines) through transmission (transformers and switchgear) to distribution (grid software like the newly announced GridOS for Transmission).
This vertical integration is the strategic masterstroke that separates GE Vernova from Siemens and MHI. The competitors can sell you a turbine. GE Vernova can sell you the turbine, the transformer, the switchgear, the substation, and the software to manage it — and it can bundle all of it into a single contract that bypasses the grid interconnection queue entirely.
GE Vernova shares have surged 694% since the April 2024 spin-off from General Electric. The stock is up roughly 70% in 2026 alone, trading around $1,036 with a market cap approaching $298 billion. The forward P/E sits at roughly 40. Bernstein initiated coverage at Outperform with a $1,206 price target. Consensus analyst targets hover around $1,211.
Jim Cramer called it "one for the ages" after the Q1 2026 beat-and-raise. The CNBC Investing Club has a $1,300 price target.
2026 guidance: revenue of $44.5-$45.5 billion, adjusted EBITDA margin of 12-14%, free cash flow of $6.5-$7.5 billion. The 2028 plan targets $56 billion in revenue at a 20% adjusted EBITDA margin.
The bull case is straightforward: GE Vernova is a pick-and-shovel play on the AI buildout with pricing power that most software companies can only dream about. Roughly 25% of the world's electricity flows through GE Vernova equipment. When electricity demand is projected to jump 39% by 2035 (per ICF) and data center consumption is forecast to increase 96% between 2026-2031 (per Wood Mackenzie), you want to own the company that builds the infrastructure.
But the valuation is not trivial. At 40x forward earnings, you're paying tech-stock multiples for what is fundamentally an industrial manufacturer — one with a money-losing Wind segment that posted a 23% revenue decline in Q1 and is expected to burn through roughly $400 million in EBITDA this year.
Every supercycle story needs a risk section, and this one has several that deserve serious attention.
First, the obvious one: valuation. A 40x forward P/E for an industrial manufacturer assumes perfect execution for years. The 12% pullback from May highs is a reminder that even the best narratives correct. If the AI capex cycle slows — and at $750 billion in 2026 hyperscaler spending, there is a ceiling — the multiple compression could be brutal.
Second, the Wind segment headwind. GE Vernova's wind business is a drag on consolidated results, and management has limited options. Offshore wind remains structurally challenged by high interest rates and supply chain costs. The $400 million EBITDA loss in 2026 is manageable at the corporate level, but it masks performance in the segments that matter.
Third, regulatory risk. Community opposition has already blocked or delayed roughly $98 billion in energy projects in 2025 alone, according to RBC Capital Markets. Gas-fired power plants face permitting hurdles that can stretch timelines by years. Tariff exposure — both for imported components and exported finished turbines — adds another layer of unpredictability.
Fourth, and most importantly: the gas turbine is not the only constraint. CEO Strazik himself warned on the Q1 call: "The gas turbines are really not the gating item when you're talking about a 3-year cycle from when a project starts — the EPC buildout, the permitting, the fuel availability." You can have a turbine on order and still face years of delays from construction, regulatory approval, and natural gas pipeline infrastructure.
Fifth, the competitive response. When prices triple, the economic incentive for new entrants and alternative technologies becomes overwhelming. Bloom Energy and Brookfield just expanded their partnership to a $25 billion framework for onsite fuel cell deployment. Nuclear — both traditional and small modular reactors — is getting serious attention from hyperscalers. Battery storage costs continue to decline. None of these alternatives can match gas turbines for baseload reliability today, but a 300% price increase has a way of accelerating substitution.
Here's the uncomfortable bottom line that the AI industry doesn't want to admit: the speed limit on AI deployment is now set by the power industry, not the semiconductor industry.
GPUs are a solved problem. NVIDIA can scale production. TSMC can build more fabs. The semiconductor supply chain, for all its complexity, is fundamentally responsive to price signals.
Gas turbines are different. The manufacturing base is fixed. The skilled workforce takes years to train. The precision casting bottlenecks cannot be scaled with capital alone. The permitting timelines are not market-driven. And the competing demand from non-data-center customers — the 80% of the order book — means hyperscalers cannot simply outbid everyone else for capacity.
The practical implication: AI data center buildout will be gated by turbine availability through at least 2030. Projects announced today with turbines secured for 2029-2030 delivery slots are the ones that will happen. Projects that assume they can find power somewhere — somehow — are gambling, and the odds are getting worse by the quarter.
This creates a fascinating competitive dynamic within the AI industry. The companies with the best power procurement teams — not the best AI researchers — may determine who wins the infrastructure race. Microsoft's aggressive turbine ordering (seven units for a single campus) looks prescient. Any hyperscaler or AI lab that hasn't already secured turbine slots through 2028 is already behind.
If you want to track this supercycle, here are the signals that matter:
GE Vernova's quarterly slot reservation updates. The combined backlog + slot reservation number (100 GW as of Q1 2026, targeting 110 GW by year-end) is the single best real-time indicator of demand intensity.
Pricing per kilowatt. When Strazik says orders are pricing "10 to 20 points higher," he's describing the slope of the pricing curve. The moment that number flattens or declines is the moment the supercycle peaks.
Hot-section component capacity announcements. New single-crystal blade foundries or casting capacity expansions from any of the major suppliers (GE, Siemens, MHI, or their specialized casting partners) are structurally bullish for turbine supply — and potentially bearish for GE Vernova's pricing power.
Behind-the-meter project announcements. Every time a hyperscaler announces an on-site generation project, it's a signal that grid interconnection queues remain dysfunctional — and that turbine demand has further room to run.
Alternative technology milestones. If Bloom Energy, small modular nuclear, or long-duration battery storage achieves a meaningful cost breakthrough, the 300% price increase in gas turbines will look like a peak, not a plateau.
The AI revolution was supposed to be about software eating the world. Instead, it's becoming about hardware eating electricity — and the hardware that makes the electricity is, improbably, a technology that was first commercialized in the 1930s.
Gas turbines are not new. They are not exciting. They have none of the intellectual glamour of transformer models or the futuristic appeal of GPU clusters. But they are the single most important physical constraint on the AI industry's growth trajectory for the rest of this decade.
GE Vernova — a company that didn't even exist as an independent entity until April 2024 — now sits at the center of the most important industrial bottleneck of the AI era. The stock has been a phenomenal trade. Whether it remains one depends on whether you believe the supercycle has years left to run or whether the 300% price surge is already pricing in the best-case scenario.
My view: the physical constraints on turbine supply — casting bottlenecks, skilled labor, factory capacity — are stubborn enough that pricing power persists through at least 2028. But at 40x earnings, the margin for error is thin. The Wind segment losses, tariff risk, and the ever-present possibility of an AI capex slowdown mean this is not a stock to own with your eyes closed.
The AI power supercycle is real. It's historic. It's also maturing. The easy money has been made. The hard money — the money that requires actually understanding manufacturing cycles, utility regulation, and metallurgical supply chains — is what's left on the table.
Published July 14, 2026. Research sources: GE Vernova Q1 2026 earnings, Wood Mackenzie gas turbine market report (April 2026), CNBC, 24/7 Wall St., Power Engineering, Reuters, Bloomberg, RBC Capital Markets.
