Gas Turbines - The Unsung Kingmakers of the AI Boom...

Food for power hungry datacenters...

For Indian businesses, you can check my writings at financepulse

Disclaimer: Before we begin, I am not a registered analyst. I’m just a student trying to study different businesses/sectors. Don’t take my word for it; try to study the businesses independently. I might have positions in the securities discussed. I am not an industry insider, so take my writings with a pinch of salt.

In October 2018, GE announced a $23 billion writedown - the majority tied to its disastrous Alstom power acquisition three years earlier. The power division, which had generated $4 billion in profits in 2016, was now losing $800 million a year. 12,000 jobs were cut. Three CEOs cycled through in two years. Larry Culp became the first outsider to lead GE in its 130-year history. The gas turbine business was, by every measure, a graveyard.

Fast forward to January 2026. The same company - now called GE Vernova, spun off in April 2024 - reports a $150 billion backlog. 83 GW of gas turbine commitments. Equipment sold out through 2029. CEO Scott Strazik tells investors he’s running the company with “much grander expectations than 2028.” The gas turbine division went from corporate albatross to the most important manufacturing bottleneck in the entire AI infrastructure stack. In six years.

What changed? One word: power. Or more precisely, the sudden, violent realization that the most ambitious technology buildout in human history - the AI data center boom - runs on electricity, and there isn’t nearly enough of it. Gas turbine orders hit 99.9 GW in 2025 - a 25-year high, matching the 2001 record for the first time in nearly a quarter century. Data centers went from literally 0% of gas turbine orders in early 2024 to 21% by Q3 2025 - and that 21% understates the real impact, because the datacenter rush triggered a panic-buying cascade across the entire customer base. Traditional utilities and industrial buyers, terrified of being locked out of delivery slots, pulled forward their own orders. The compounding effect means data centers reshaped the entire market, not just their corner of it. Hyperscaler capex is running at $400 billion a year and climbing toward $750 billion by 2028. And as Amazon CEO Andy Jassy put it on a Q3 2025 earnings call: “Today, overall in the industry, maybe the bottleneck is power.”

In this note, i cover - i) why gas turbines became the critical chokepoint in the AI buildout - and how data center demand triggered a cascading order surge across the entire power industry, ii) the historical arc from the devastating 2000s bust through two decades of stagnation to today’s supercycle - and why the OEMs’ scar tissue from that bust is the very reason supply can’t catch demand, iii) the OEM oligopoly - three companies control 87% of the market, their capacity is sold out for years, and a new class of creative entrants is scrambling to fill the gap with converted jet engines and fuel cells, iv) the “bridge technology” fallacy - everyone calls gas turbines a bridge to nuclear SMRs, but history suggests the bridge always becomes the destination.

HOW A FORGOTTEN INDUSTRY BECAME THE MOST IMPORTANT BOTTLENECK IN AI

THE GREAT DEREGULATION FRENZY

The gas turbine industry’s first golden age started with a policy change. When the Fuel Use Act was repealed in 1987 - ending a decade-long ban on new natural gas power plants - it unleashed a wave of construction that no one anticipated. Electricity deregulation in the 1990s created a new class of buyer, the independent power producer, and these IPPs loved gas turbines. Short construction times. Low capital costs. Improving efficiency. Between 1990 and 2005, the US built approximately 260 GW of natural gas power plants.

The technology was advancing in lockstep. Combined-cycle gas turbines demonstrated learning rates of roughly 25% during the 1990s - actually exceeding solar photovoltaics at the time. Single-crystal turbine blades, originally developed for military jet engines in the 1970s, migrated into industrial machines, allowing operating temperatures that pushed thermal efficiency toward levels coal plants could never match. By the late 1990s, the industry was on fire. Combined-cycle orders worldwide had been just 3.9 GW in 1988. By 2001, annual gas turbine orders hit approximately 89 GW.

The production value of gas turbines for electric power reached $32 billion (inflation-adjusted) in 2001, briefly eclipsing the entire aviation gas turbine market. GE, Siemens, Alstom, and Mitsubishi were all expanding capacity as fast as they could.

It was, in hindsight, the peak of irrational exuberance. The IPPs were speculators, not operators. They had weak balance sheets, aggressive assumptions about deregulated power prices, and little margin for error.

THE BUST THAT NEARLY KILLED THE INDUSTRY

When the merchant power bubble burst - fueled by Enron’s collapse, the California energy crisis, and the realization that far too many plants had been built chasing too few electrons - the crash was apocalyptic. Gas turbine production value for electric power fell from $32 billion in 2001 to roughly $8 million by 2003. Not a typo. Billion to million. OEMs that had aggressively built greenfield manufacturing capacity found themselves staring at empty factories and order books with nothing in them.

The industry spent the next decade and a half picking itself up. Orders recovered modestly, driven mainly by coal-to-gas switching as cheap shale gas made coal plants uneconomic. Between 2011 and 2025, the US retired over 131 GW of coal capacity - every gigawatt creating some replacement demand for gas.

Renewables were eating into the growth too. Between 2010 and 2016, solar costs dropped 69% and onshore wind fell 20%. Wind and solar captured growth that might otherwise have gone to gas. Annual gas turbine orders through most of the 2010s hovered in the 30-40 GW range - respectable, but a shadow of what came before.

And then electricity demand itself stopped growing. After climbing steadily for decades, US power consumption flatlined around 4,000 TWh after the 2008 financial crisis. The shift from manufacturing to services. Even data centers - which were growing rapidly - consumed surprisingly flat amounts of power between 2010 and 2018 as hyperscalers consolidated workloads into efficient mega-facilities. The gas turbine was a solution looking for a problem.

GE POWER’S NEAR-DEATH EXPERIENCE

No company embodied the pain more viscerally than GE. In November 2015, CEO Jeff Immelt completed the $10.6 billion acquisition of Alstom's power business - a deal conceived during the optimism of 2014, when he projected $3 billion in cost synergies and 15-20% returns. Alstom's own management had warned in its 2014 annual report about "excess capacity in developed markets." Immelt pressed ahead anyway.

The timing was catastrophic. Within two years, the gas turbine market had collapsed further. In March 2017, GE forecast it would sell 160 Advanced Gas Path units in 2018. By November 2017, that estimate had been slashed 75% - to roughly 40. GE Power went from generating $4 billion in profits in 2016 to losing more than $800 million in 2018.

In October 2018, GE announced a $23 billion writedown - the majority tied to the Alstom deal. The company had written down more in goodwill than it paid for the business just three years earlier. GE shed $193 billion in market capitalization between 2016 and 2018 - 74% of its value, gone. They cut 12,000 jobs in the power division alone, roughly 18% of the workforce. CEO Immelt was forced out in June 2017. His replacement, John Flannery, lasted fourteen months. Larry Culp took over in October 2018 - the first outsider to lead GE in its 130 year history.

That’s how bad things were. They went outside for help for the first time in over a century.

We dwell on the GE story because it explains something crucial about today’s supply picture. It’s a $23 billion mistake that nearly sank a 130-year-old company. It’s three CEOs in two years. When GE Vernova and Siemens Energy talk about “prudent capacity additions” and “surge capacity” today - weekend shifts, reactivating older production lines, productivity gains that can be reversed if demand softens - they are making decisions shaped by lived trauma. This is why supply won’t catch demand quickly. The people making the capacity decisions remember what happened last time someone got aggressive.

THE TECHNOLOGY THAT NOBODY WANTED

Here’s the irony that makes this story sing. While the market was in freefall, the technology kept advancing. GE introduced its H-class turbines - the 9HA and 7HA - in 2014, the worst possible moment to launch the most advanced gas turbine ever built. These machines used single-crystal superalloy blades, ceramic thermal barriers, and complex internal cooling channels that pushed operating temperatures to nearly 1,600 degrees Celsius - above the melting point of most metals.

In 2016, GE set a Guinness World Record at EDF’s Bouchain plant in France: 62.22% net combined-cycle efficiency - the most efficient power plant on Earth. The 9HA.02 later exceeded 64%, more than double what most coal plants could achieve. At $700-1,000 per kilowatt, combined-cycle gas plants were also far cheaper to build than coal ($3,000/kW) or nuclear ($6,000+/kW).

The engineers had built a masterpiece. Nobody was buying. Annual orders ground along at 25-35 GW through the late 2010s while these world-record machines sat in catalogs waiting for demand that wouldn’t come for years. The gas turbine was the best power generation technology on the planet and the market didn’t care.

WHAT BROKE THE FLATLINE

ChatGPT launched in November 2022. Within months, the AI arms race had every major technology company scrambling to build data centers. But AI workloads were fundamentally different from anything that came before. Training runs consumed power at a scale and intensity that made traditional computing look quaint. The efficiency gains that had kept data center power flat during the 2010s - server consolidation, better cooling, workload optimization - couldn’t keep pace with exponential growth in AI compute.

By early 2024, the power bottleneck was becoming visible. US electricity demand, flat for fifteen years, was suddenly projected to grow 16% over the next four years - a growth rate not seen since the early 2000s. Hyperscaler capex exploded: $400 billion in 2025, rising to $600 billion in 2026, with a trajectory toward $750 billion by 2028. HSBC estimates total AI investment to 2030 at approximately $5 trillion. And all of it needed power.

GE Vernova spun off from GE as an independent company on April 2, 2024 - emerging from the ruins of the same GE Power that had been hemorrhaging cash six years earlier. On its very first earnings call, April 25, 2024, CEO Scott Strazik reported booking 8 HA gas turbine orders in Q1 alone - equal to the total for all of 2023. But he was careful. Measured. “I do think it’s very practical to think this year that our orders in gas equipment could very well be larger than our revenue,” he said. Hyperscaler demand, he noted, was “more of a 2025 dynamic than 2024.”

He was right about the direction. He underestimated the speed.

THE SUPERCYCLE ARRIVES

Some spinets from GE Vernova earnings calls:

October 2024 (Q3): “We are in the early innings of an investment supercycle.” Gas orders had hit 14 GW year-to-date - double the prior year. Nine HA units booked in a single quarter, exceeding all of 2023. The company announced capacity expansion from 55 to 70-80 heavy-duty turbines per year. Conversations with hyperscalers had shifted from small aeroderivative backup units to multi-gigawatt baseload installations with HA turbines.

January 2025 (Q4): Full-year gas orders reached 20 GW - double the prior year. An SMR development consortium with Duke Energy and AEP was announced. Strazik put it plainly: “I like gas because it lets you get to baseload faster than anything.”

July 2025 (Q2): Total commitments hit 55 GW. Nine GW contracted in a single quarter. Power EBITDA margins reached 16.4%. The company was buying back $1.6 billion in stock. And Strazik said something that caught our attention: “We are running this business and intending to lead this company with much grander expectations than 2028.” The CEO of a company that six years ago was writing down $23 billion was now saying his medium-term targets weren’t ambitious enough.

January 2026 (Q4 2025): Backlog hit $150 billion. Total gas commitments: 83 GW - with expectations of reaching 100 GW by end of 2026. Equipment sold out through 2029. SRA pricing running 10-20 percentage points above existing backlog. Revenue guidance raised to $44-45 billion for 2026. The company added $8 billion of equipment margin to backlog in a single year.

The transformation from "managing decline" to "investment supercycle" happened in roughly six years. From a $23 billion writedown to $8 billion in cash, no debt, and an 83 GW commitment book.

And it wasn’t just GE Vernova. Baker Hughes CEO Lorenzo Simonelli declared “the age of gas” on his Q3 2025 earnings call - and by Q4 was calling it “a demand decade.” NovaLT turbine slots sold out through 2028. Data center power orders exceeded $1 billion for the year. Eaton reported data center orders up 200% and identified 206 GW of US data center backlog - representing 11 years at current build rates. PJM capacity prices, which had hovered at $50-100/MW-day for most of the prior decade, jumped to $270 for 2025-2026 and then $330 for 2026-2027. Summer on-peak beat winter on-peak in PJM for the first time in a decade.

As Talen Energy’s Chris Morice put it: “Power is up, sparks are widening, and it remains a good time to be in power.” Or more bluntly from CEO Mac McFarland: “Power is not getting any cheaper.”

THE FULL CIRCLE

In 2025, global gas turbine orders hit 99.9 GW - matching the all-time record from 2001. It took 24 years to come full circle. But the industry that arrived at this peak is fundamentally different from the one that stood here last time. The technology is generationally superior. The buyers are creditworthy. The OEMs are disciplined. The demand drivers are diversified. And the people making the decisions carry scars that make recklessness unlikely.

There’s a line from Strazik’s Q3 2025 call that i keep coming back to: “There is really no scenario where the world doesn’t need significantly more power.” He’s right. And the 86-year-old technology that started in a Swiss factory in 1939, survived a US construction ban in the 1970s, weathered the worst industrial bust in a generation in the 2000s, and nearly destroyed General Electric in the 2010s - that technology is now the critical infrastructure underpinning the most important technology buildout since the internet.

The gas turbine industry didn’t plan to become the backbone of AI. But history rarely asks permission.

THE THREE KINGS AND THEIR CHALLENGERS

Three companies make 87% of the world’s gas turbines.

Turbine blades operate at 1,400-1,600 degrees Celsius - above the melting point of most metals. They’re cast from single-crystal superalloys with ceramic thermal barriers and cooling channels thinner than a human hair. The metallurgy took decades to develop. The manufacturing expertise lives in a handful of factories across the US, Germany, and Japan. You can’t replicate this with money alone.

The result is one of the most concentrated industrial markets. GE Vernova, Siemens Energy, and Mitsubishi Power split 87% of global gas turbine orders by megawatt. The remaining 13% is fought over by Baker Hughes, Doosan, Caterpillar, and a growing roster of creative newcomers converting airplane engines into power plants.

What makes this moment different from any prior cycle is who’s buying. The customers aren’t speculative independent power producers with shaky balance sheets. They’re hyperscalers with $400 billion in combined annual capex and 25% non-refundable deposits on turbine slots that didn’t exist two years ago. Regulated utilities pulling forward orders. International buyers in the Middle East and Asia. As Constellation Energy CEO Joe Dominguez told investors in Q3 2025: “The market is hotter now than ever. And the real big difference we’re seeing is buyer maturity... Far more sophisticated and aggressive customer walk through our door. They have done deals. They understand pricing and term.”

GE VERNOVA - THE FRANCHISE

GE Vernova is to gas turbines what NVIDIA is to GPUs - the dominant franchise in the hottest market on the planet.

Total backlog: $150 billion. Gas turbine commitments: 83 GW across firm orders and slot reservation agreements. Equipment sold out through 2029. Strazik expects to reach 100 GW under contract by end of 2026, at which point delivery slots through 2030 will be largely spoken for.

In North America - where data center demand is hottest - GEV commands 45% market share by megawatt. Nearly double Siemens Energy’s 24.5%. The installed base advantage is self-reinforcing: every HA turbine GEV sells locks in a long-term service agreement worth multiples of the equipment sale. Management calls it the “razors and blades” model. In 2025 alone, GEV added $8 billion of equipment margin to backlog - not revenue, just the profit portion. Decades of service revenue follows.

The capacity expansion is deliberately measured. GEV is ramping from 55 heavy-duty turbines per year to 70-80 by 2026, then 90-100 by 2028, with annual manufacturing capacity hitting 20 GW by Q3 2026 and 24 GW by 2028. But it’s all brownfield - reactivating 2 GW of older 7E capacity in France (no capex needed), adding weekend shifts at Greenville, installing 400 new machines over 2025-2026. No greenfield factories. Their data center-specific wins include 29 LM2500 aeroderivatives for Crusoe (~1 GW), 7 7HA.02 units for Homer City (~4.4 GW), and 7 HA slot reservations with Engine No. 1.

New SRA pricing is running 10-20 percentage points above existing backlog - meaning every new commitment is more profitable than the last. Revenue guidance for 2026 is $44-45 billion.

SIEMENS ENERGY - THE AGGRESSIVE BET

While GEV plays it disciplined, Siemens Energy is swinging for the fences.

Total commitments: 78 GW. Record quarterly Gas Services intake of EUR 8.75 billion, including 102 gas turbines in a single quarter. In the first six weeks of FY2026 alone, Siemens booked 8 GW in new reservations.

The strategic divergence from GEV is clear: Siemens is investing $1 billion in US manufacturing expansion - the largest capex commitment among the Big Three. That includes a $421 million expansion in North Carolina (500 new jobs), plus facility upgrades in Florida, Alabama, New York, and Texas, plus a brand-new factory in Mississippi for grid components. The expansion increases global large gas turbine production capacity by roughly 20%. Siemens is targeting 22 GW average annual capacity in FY2025-27, ramping to 30+ GW by FY2028-30.

The product range is the broadest in the industry. The SGT-800 industrial turbine (62 MW) claims 90% market share in its class. The SGT6-9000HL heavy-duty (440 MW) competes head-to-head with GEV’s HA platform. And Siemens has been more aggressive on supply chain de-risking than anyone - acquiring Capital Injection Ceramics in the UK to eliminate the ceramic core bottleneck, signing a strategic supplier agreement with Yingliu in China for blade supply diversification, and inking a 5 GW behind-the-meter deal with Williams Co. for data center projects.

MITSUBISHI POWER - THE CONSERVATIVE RISK

Mitsubishi Heavy Industries makes the largest gas turbine in the world. The M501J produces 453 MW - roughly 5% more than anything GEV or Siemens offers. When raw output matters in a combined-cycle configuration, MHI wins.

But MHI’s global share may be at risk. Management is taking a deliberately “lean approach” - refusing to expand fixed assets because “it is unclear whether elevated demand will be sustained in the long-term.” Current capacity sits at 11-12 GW per year, with a modest 30% increase planned by end of FY2026 - roughly 16 GW. Compare that to Siemens targeting 30+ GW by 2030.

The conservative stance is prudent if this cycle fizzles. It’s a competitive vulnerability if it doesn’t. Customers with urgent timelines - and right now, every customer has an urgent timeline - go to whoever can deliver. MHI’s backlog of 67 large-frame units (~26 GW) stretches to FY2029-2030. The Japan-to-US logistics chain adds complexity that domestic manufacturers don’t face.

MHI’s strength is concentrated in Asia and the Middle East - a recent 2.4 GW J-class order for a Qatar power-and-desalination project highlights where the technology shines. Of 23 large-frame orders in the first half of FY2025, 11 were from Asia - showing the international base that buffers MHI against US-centric risk. But in the white-hot North American data center market, MHI holds only 16% share versus GEV’s 45%.

THE BLADE BOTTLENECK

If you want to understand the single biggest constraint on gas turbine production, forget the assembly lines. Look at the blades.

High-pressure turbine blades represent 20-35% of total turbine value - $3 to $4.5 million per unit for a mid-size turbine. They’re cast from single-crystal nickel superalloys using a process that takes weeks per blade, with rejection rates that would make semiconductor fabs blush. Two companies dominate global supply: Howmet Aerospace and Precision Castparts (owned by Berkshire Hathaway). A duopoly controlling the most critical component in a $100+ billion industry.

Howmet reported record 2025 results and is targeting $9.1 billion in revenue for 2026. High-pressure turbine blade output is up 40-50% year-over-year. They’ve hired 4,000 employees in engine-related operations since 2022. But the workforce had dropped 43%, from 41,700 in 2019 to 23,930 in 2024. You can’t train precision single-crystal casting workers overnight. And they still expect to be “constrained in our ability to supply over the next couple of years.”

The blade bottleneck is why OEMs are diversifying supply chains with urgency. Siemens Energy signed with Yingliu in China - starting H-class blade volume production by 2027. Azad Engineering in India is building a facility at 10x current capacity. But geopolitical risk on Chinese supply and the sheer difficulty of matching the quality standards of Howmet and PCC mean this constraint persists through the decade.

The supply-demand gap is stark. Industry capacity sits at roughly 47 GW against demand of 80-100 GW - a 33-53 GW shortfall. Even with aggressive expansion across all OEMs, capacity reaches maybe 65 GW by 2028. The market will remain structurally under-supplied until the decade’s end.

The transformer bottleneck is equally severe. Lead times have stretched to approximately five years - up from two years just three years ago. Eaton reported in Q4 2025 that it had identified 206 GW of US data center backlog - representing 11 years of construction at 2025 build rates. Data center orders at Eaton are up 200%, with Electrical Americas backlog at a record $15.3 billion - up 31% year-over-year. GE Vernova’s $5.275 billion Prolec GE acquisition was specifically to address the transformer shortage - vertical integration born from desperation.

And the labor constraint adds another layer. The US has 80,000+ electrician openings annually. Quanta Services CEO Duke Austin identified “inside wiremen” - data center electrical workers - as the scarcest trade. His company has a record $39.2 billion backlog and it doesn’t even include 765kV transmission projects or the NiSource 3 GW combined-cycle program. As Austin put it: “When you start ordering turbines that are 36 to 48 months out, what are you going to do between now and then?”

THE NEW ENTRANTS - CREATIVITY BORN FROM SCARCITY

When the Big Three are sold out through 2029, creative destruction fills the gap. A wave of new entrants is converting airplane engines, repurposing supersonic jet technology, and deploying fuel cells to solve a power crisis that the incumbents simply can’t address fast enough.

None of these companies threaten the Big Three in large-frame gas turbines. The manufacturing moat is too deep. But they’re carving out real, defensible niches in the behind-the-meter and distributed power segments - the markets where speed-to-power beats thermal efficiency.

Doosan Enerbility landed the marquee deal: five 380 MW H-class frames for Elon Musk’s xAI Colossus facility in Memphis - 1.9 GW total. First delivery by end of 2026. This is a Korean manufacturer with 3.1% global share punching above its weight because it had available capacity when the Big Three didn’t.

FTAI Aviation is doing something nobody expected - converting CFM56 jet engines (the most common commercial aircraft engine ever made) into 25 MW power turbines. FTAI has over 1,000 CFM56 engines in hand, existing in-house MRO capability, and the capacity to produce 100+ units annually (2.5+ GW per year). First units expected 2026-2027. The advantage is inventory - there are thousands of CFM56s available on the aftermarket as airlines retire older 737s and A320s.

Boom Supersonic - yes, the company building a supersonic airliner - pivoted its jet engine technology into a 42 MW natural gas turbine called “Superpower.” Launch customer Crusoe ordered 29 units for 1.21 GW, creating a $1.25 billion backlog. Testing begins 2026, first delivery 2027. Boom raised $300 million from Darsana, Altimeter, ARK Invest, and Y Combinator. The aerospace-to-power pivot is one of the more creative moves i’ve seen.

ProEnergy is converting CF6-80C2 engines - the powerplant from the Boeing 747 - into 48 MW PE6000 gas turbines. Already sold 21 units across two data center projects totaling 1+ GW. Each conversion bolts the engine onto a steel frame and swaps fuel nozzles for natural gas. Delivery by 2027 - years faster than Big Three large-frame delivery. “Bridging power” for 5-7 years until grid interconnection arrives.

Bloom Energy isn’t a turbine company at all - it makes solid oxide fuel cells. But it may be the fastest path to power for a data center. Management delivered a hyperscale AI factory in 55 days. Bloom is the only company natively producing 800V DC power, which aligns perfectly with next-gen data center electrical architecture - as Eaton confirmed, the 800V DC transition is now “everywhere.” Product backlog is up 140% year-over-year to roughly $6 billion. They went from one hyperscale customer to six in a single year. CEO KR Sridhar’s pitch on the Q4 2025 call: “Bloom will not be the bottleneck to your growth.” On track to reach 2 GW annual production capacity by end of 2026.

THE BEHIND-THE-METER ECOSYSTEM

A parallel power system is being built outside the grid. Behind-the-meter generation - where the data center builds its own power plant on site - has gone from a novelty to the default approach for hyperscalers who can’t wait five-plus years for grid interconnection.

Meta’s Head of Data Center Design said it plainly: “We’re only moving towards behind-the-meter because that’s the only way that we can build fast.”

According to estimates 33+ GW of behind-the-meter data center power is now committed across various technologies:

VoltaGrid is the clearest example of how the BTM model works. The company is deploying 2.3 GW of modular natural gas generation for Oracle across multiple Texas sites, including the Stargate data center in Shackelford County - a 1.4 GW behind-the-meter microgrid that could begin operations as soon as 2026. VoltaGrid signed a 1.5 GW supply agreement with INNIO Jenbacher for 300 gas engines, each node producing up to 20 MW, combinable to 200 MW. The BTM model works because it sidesteps every bottleneck at once. No FERC approval needed - just an air permit. No five-year grid interconnection queue. No dependence on utility construction timelines.

The infrastructure enablers are riding this wave. Caterpillar is increasing engine manufacturing capacity for data centers by 125% from 2023, with a $725 million investment in the Lafayette Engine Center - its largest since 1982. A 2 GW deal with AIP for the Monarch Compute Campus in West Virginia has deliveries scheduled September 2026 through August 2027. Baker Hughes’ NovaLT turbines have booked approximately 2 GW of orders with delivery slots full through 2028 - Simonelli expects $3 billion in cumulative data center orders for 2025-2027.

Chevron represents perhaps the most vertically integrated BTM play: the oil major has reserved 7 GEV turbine slots for a 2.5-5 GW off-grid data center in West Texas, powered by Permian Basin associated gas - gas production, power generation, and data center hosting under one corporate umbrella. First power targeted 2027.

THE SELLER’S MARKET FLYWHEEL

The pricing dynamics are extraordinary and, i believe, still under appreciated.

Consider what’s happened in the past two years. Slot reservation fees went from nonexistent to $25 million per turbine - a deposit that simply did not exist before 2024. A Kentucky utility saw turbine pricing jump from roughly $1,400/kW for 2027 delivery to roughly $2,100/kW for 2030 delivery - a 50% increase for a later slot on essentially the same equipment. GE Vernova’s slot reservation agreement pricing is running 10-20 percentage points above existing backlog. Delivery slots are now tradeable between customers with OEM approval - a secondary market for turbine delivery has emerged, which tells you everything about the scarcity value of these slots.

And the flywheel feeds on itself. Tight capacity leads OEMs to offer only standard turbine specs instead of bespoke configurations - which improves their margins. Customers accept whatever is available. LTSAs get bundled with equipment purchases - locking in decades of high-margin service revenue. Every new order pushes delivery timelines further out, which increases urgency for the next buyer, which accelerates orders further. GE Vernova expects to sell out 2029 and 2030 delivery by the end of 2026.

Talen Energy’s CFO put a number on the scarcity premium: to justify building a new combined-cycle gas plant, you need power prices “well over $100 per megawatt hour.” At current forward curves, existing gas assets are getting repriced but new build economics still don’t pencil. That means existing plants run harder and longer - and every year without enough new capacity tightens the market further. Vistra’s $4.7 billion Cogentrix acquisition - adding 5.5 GW of gas-fired generation - is the clearest bet that existing capacity is getting more valuable, not less. As McFarland put it: “Power is not getting any cheaper.”

THE COMPETITOR NOBODY TALKS ABOUT: SOLAR + STORAGE

There’s a competitive threat to the gas turbine thesis that deserves honest treatment: solar-plus-battery microgrids are approaching cost parity with gas for data center power.

The economics are closer than most gas turbine bulls want to admit. An off-grid gas turbine system runs roughly $86 per megawatt-hour on a levelized cost basis. A 44% renewable microgrid - solar panels, battery storage, and backup gas generators - costs approximately $87 per MWh. Virtually identical. Push the renewable mix to 90% and the cost rises to only $97 per MWh. Battery pack costs sit at $108 per kilowatt-hour and are declining roughly 3% annually.

Quanta’s CEO Duke Austin said: “If you were going to build generation tomorrow... you would find yourself building a solar plant probably. It’s the fastest thing you can build.” Solar projects can be physically constructed faster than gas plants. They don’t need gas pipeline connections. They don’t face the same community opposition.

So why isn’t solar displacing gas turbines for data centers? Three reasons - for now. First, data centers need power 24/7, and battery duration remains capped at 4-8 hours economically. AI workloads can’t tolerate intermittency - a training run that loses power mid-batch is catastrophically expensive. Second, energy density. A 1 GW gas plant occupies maybe 30 acres. The equivalent in solar panels needs thousands of acres, plus the battery farm. For urban and suburban data center sites, the land math doesn’t work. Third, most hyperscalers need power in 2026-2027, not 2029. The turbines - even with their long lead times - have a more certain delivery path.

But the cost curves are converging. If battery costs continue declining and duration extends beyond 8 hours economically, solar+storage becomes a genuine threat to gas turbines for new data center builds by the late 2020s - particularly for facilities in high-solar-resource regions like the Southwest. I don’t think this kills the gas turbine cycle, but it may cap its duration. The “bridge” to renewables may actually arrive faster than the “bridge” to nuclear.

THE BULL CASE IN FIVE NUMBERS

I’ll close our view with the five data points that matter most:

99.9 GW - Global gas turbine orders in 2025. A 25-year high. Matching the 2001 record but backed by creditworthy buyers paying 25% deposits instead of speculative IPPs with leverage.

47 vs. 100 - Industry supply capacity in GW versus demand in GW. A 53 GW gap. Even with aggressive expansion, supply doesn’t catch demand until end of decade at the earliest.

7 quarters - How old this cycle is. Previous upturns lasted 10-20+ quarters.

0% to 21% - Data center share of gas turbine orders, from early 2024 to Q3 2025. And hyperscaler capex is still ramping from $400 billion toward $750 billion by 2028. The demand driver that created this cycle hasn’t peaked.

$150 billion - GE Vernova’s backlog. For a company that wrote down $23 billion seven years ago.