# Can Natural Gas Supply Handle the AI Boom? Forty-Eight Months in the Turbine Business

Source: Realpha Blog (blog.getrealpha.com)
Original article and charts: https://blog.getrealpha.com/en/blog/catalyst-2026-10-15-can-natural-gas-supply-handle-the-ai-boom/

> Notes from the 2026-10-15 episode of Catalyst with Shayle Kann, 'Can natural gas supply handle the AI boom?' Dora Partners president Tony Bruff on 36-to-48-month lead times, a 30-35% price rise over five years, and the superalloy and forging capacity that aerospace is drawing from at the same time. Educational commentary, not investment advice; no price targets and no stock recommendations.

Published: 2026-10-15
Locale: en
Tags: gas turbines, AI data centers, supply chain bottleneck, energy, podcast notes

TL;DR: Turbine lead times have stretched to three or four years and prices are up a third, because the superalloy and forging capacity is shared with aerospace.


![A turbine assembly hall at night, half-built rotors receding one after another down the long bay](/covers/catalyst-2026-10-15-can-natural-gas-supply-handle-the-ai-boom-cover.png)

> A small sack cannot hold a large thing; a short rope cannot draw water from a deep well.
>
> —— *Zhuangzi*, "Perfect Happiness" (Warring States period; translation mine)

On the 15 October 2026 episode of *Catalyst with Shayle Kann*, host Shayle Kann brought in Tony Bruff, president of Dora Partners, to talk about the gas turbine market. Bruff said new units now carry lead times of 36 to 48 months, with some OEMs quoting up to 60, and that prices are up 30 to 35 percent over five years. What caps output sits in what he calls level zero and level one — superalloy raw materials, forge shops and casting shops — and those same shops are feeding an aerospace industry with roughly 40,000 aircraft in backlog. He separates today from the 1998-2001 bubble, where he estimates at least half the order volume was artificial; today he counts five drivers pushing at once. The condition on all of it is the data center piece: Kann himself says the thousand-plus large data centers in development in the United States will not all get built.

## He starts by taking me back to 1998

Kann opens with history, and Bruff goes back to the seventies and eighties, when the business had two major OEMs — General Electric and Westinghouse, the latter now part of Siemens. Today it has consolidated to three: Mitsubishi Heavy Industries, Siemens and GE Vernova, holding something like 70 percent of the market between them.

Then he gets to 1998 through 2001. Orders came in like a flood, and then the market fell off a cliff. The word he uses is disaster — for everybody.

What interests me is his explanation of the flood. Two things stacked up: deregulation, and Enron. Large utilities in the sixties and seventies operated under regulation and only built what public utility commissions would approve. Once deregulation arrived, everyone was relearning one question — how do you make money across a country with regulated, semi-regulated and deregulated markets side by side. In the middle of that scramble, Enron pushed artificial signals into the market and drove up power prices in California, Texas and elsewhere. Panic followed.

Bruff then gives a number I didn't expect him to say out loud: at least half that volume was artificial. What came next was cancellations even at large regulated and semi-regulated utilities, with ordered units scrapped or bought and resold on the open market.

![A bar chart shows orders stacking higher each year from 1998 to 2001, then dropping vertically back to the ground in 2002 in a cliff shape, with a lighter upper segment on each bar marking roughly half of it as phantom demand.](/figures/1998-order-flood-and-cliff-en.svg)

## What the "half the orders were fake" line is good for

Kann wants to draw a line. Was the demand that vanished back then speculative merchant development — plants selling into the spot market with no long-term buyer? Because today looks different: the projects moving are either utilities saying they need the capacity, or data centers signing on as long-term offtake. The risk you carry now is whether the data center ever gets built.

Bruff half agrees. Independent power producers were speculating, no question. But plenty of the orders that blew up came from large regulated and semi-regulated utilities like Southern.

That stayed with me. When you're judging whether an order is real, the identity of the buyer tells you little. What tells you something is how much the money hurts and how tightly the contract binds.

## The four levels he draws are the bones of the episode

Asked what the supply chain looks like, Bruff splits it into four levels, and this is the part I'd copy down.

Level zero is raw materials. He contrasts it with transformers, where copper is the headline input; for gas turbines it's superalloys — nickel, nickel-based alloys, chromium, titanium, the expensive ingredients. Level one turns those into rough product: blades, vanes, the pieces that get cast or forged. Level two is the OEMs' own plants assembling what he calls a flange-to-flange gas turbine. Level three is final packaging, delivery to the operator site, installation, commissioning and aftermarket work.

Then comes the line that made me stop. Levels zero and one are occupied by aerospace at the same time. There are roughly 40,000 aircraft in backlog worldwide; at two turbines apiece, that's 80,000 engines. The same material suppliers, the same forgers and casting shops, are serving aerospace simultaneously.

![A four-tier supply chain diagram stacked from the bottom up, with an aerospace block on the right pulling capacity sideways out of the bottom two tiers through two arrows.](/figures/four-tier-supply-chain-aero-pull-en.svg)

So you can't look at the power sector in isolation. And he adds a denominator most outsiders miss: in any given year, about half the industrial gas turbines delivered go to oil and gas, not to electric utilities. That half is bidding for the same supply chain.

![A long bar is split into left and right halves, utilities on the left and oil and gas on the right, each half sending an arrow down into the same supply chain node.](/figures/half-of-turbines-go-to-oil-and-gas-en.svg)

Kann follows up: with oil and gas prices low, does investment on that side pull back and push all the demand into power? Bruff says the majors run five-, seven- and ten-year strategies; a single soft year may cut order activity, but the long-run investment stays. What he sees is a seven-year cycle — up, a peak around year seven, a slow decline, then up again. Upstream, midstream and downstream move somewhat independently, and midstream players make their money moving the molecules rather than on the price of them, so their sensitivity to price is low. The gas still flows, the tariff still gets collected, the investment continues.

## The 30 percent price rise isn't all supply and demand

On price, Kann cites a data point: John Ketchum of NextEra said a new gas project cost him something like $750 per kilowatt a decade ago, and would run $2,500 today.

Bruff answers precisely. The $750 was for a combined cycle plant. The $2,500 he calls a bit aggressive, but the direction is right — prices are up 30 to 35 percent over the last five years, and he tracks it closely.

![Three bars of different heights stand side by side: a low bar for a decade ago, a middle bar for the increase he traced, and the tallest bar for the market quote, shaded in a doubtful colour.](/figures/capex-per-kw-two-claims-en.svg)

As for why, it's yes and no. Supply and demand is in there, but level zero raw materials are up across the board — aluminum, nickel-based alloys, titanium — and they were rising before tariffs entered the picture. He loops back to aerospace: 40,000 aircraft means at least 80,000 turbines, and everyone drinks from the same supply chain. He points to the US government's producer price indices, where the last three years show a jump that matches the conversation.

Kann asks whether lead times keep stretching. Bruff doesn't think they get much worse: all three OEMs are working hard to shorten them, or at minimum hold the line. His reason is customer fatigue. Buyers are putting down 15 to 25 percent non-refundable deposits, and that gets painful enough that they start deferring projects. These OEMs have lived through the booms and busts before, and they don't want to push their customers too far.

## He bet on one driver once, and that driver didn't show up

In the back half, Bruff lays out five demand drivers: grid-scale battery storage, coal plant retirements, grid-scale renewable expansion, the rapid buildout of data centers and AI, and the availability and affordability of natural gas in the United States. Units split into three size classes — under 20 MW, 20 to 100 MW, and what he calls jumbo units at 150 to 250 MW and above — and each driver hits the three classes differently.

Here's the counterintuitive bit. Kann assumes grid-scale storage, being a peaker replacement, suppresses turbine demand. Bruff agrees for jumbo units, where the effect is negative. But in the 40 to 100 MW band it's an opportunity: some developers are building hybrid systems that spin up a turbine when gas is cheap and use it to recharge grid-scale batteries, rather than leaning on renewables alone. Below 20 MW, storage barely registers.

![Three blocks arranged by unit size, each with an arrow pointing a different way: flat for small units, upward for mid-size, downward for large.](/figures/storage-impact-by-turbine-size-en.svg)

The mobile segment taught me something too. I had mobile generators filed under off-grid work and bridge power — a data center needs electrons now, the grid interconnection takes three to five years, so you bridge the gap. Bruff says that's right but incomplete: plenty of utilities buy mobile units, park them in what they call a grid-sensitive area, spend five to ten years improving the local infrastructure, then move the units to the next sensitive area. Three players have done well here — Caterpillar's Solar Turbines, GE Vernova, and MHI Aero Power.

Then his own story. Twenty years ago he was an executive at one of the big OEMs, and their strategy counted on a single market driver materializing. It didn't, and the strategy suffered. Today three to five drivers are pinging the market at once, which he calls risk comfort — the OEMs will build out capacity because they aren't relying on one thing.

On technology he covers two areas. Combined cycle efficiency crept from an average of 55 percent to 60, then hit a dead spot nobody could get past; the OEMs shifted to a whole-plant approach, pulling levers across the HRSG and everything else, and nudged it to 60.1, 60.3, and now toward 62. Bruff calls the cumulative gain remarkable, hard-won rather than dramatic.

The second is converting combustion systems to hydrogen, which he wraps in his own air quotes. All three are working on it and all three have demonstrated some capability, but the problem is where the hydrogen comes from. A jumbo unit burning hydrogen needs an enormous volume of gas. The engineering is excellent; he doubts the market will show a meaningful jump in purchases of turbines actually running on hydrogen. Kann's summary: it's an if-we-build-it-they-will-come bet.

## "Both things can be true" is the line I keep going back to

Let me shift angle. A lot of people hear an episode like this and get stuck on one question: the news says demand is exploding and OEMs are sold out through 2029, so why doesn't buying the equipment names work out?

Kann's framing pried that apart for me. From the OEM's seat, the demand is real and the market will buy an enormous volume of turbines. At the same time there's a speculative bubble on the development side, because the thousand-plus large data centers in development won't all appear — he states flatly that he doesn't think the demand is there for that many. Both true at once.

![On the left a thick bar of even width stands for firm orders holding up, and on the right a funnel wide at the top and narrow at the bottom stands for planned data centres falling away in large numbers.](/figures/both-true-at-once-en.svg)

Stack those two layers and you know what to watch. A full order book is one thing; who keeps the money is another. Level zero materials are rising and level one forge slots are being taken by aerospace, so how much of that 30-something percent is pass-through cost and how much is widened spread lives in gross margin and contract terms, not in the headline. And the 15 to 25 percent non-refundable deposit is this industry's own verification tool — in a market where you can't tell a real buyer from a tourist, make it hurt.

## Next time I hear "the technology is ready," I'll ask where the feedstock is

The other common frustration: when the story sounds this good, how do I tell which link gives way first?

Hydrogen is the ready-made sample. Three OEMs have engineered their way to demonstrated capability, and Bruff's doubt sits not on the machine but on the supply — where does that much hydrogen come from. The technology being ready says nothing about the chain being ready, and the least glamorous link on that chain sets the date.

I turned it into a check I can actually run: whenever I hear "the technology is mature, now it's about scale," I walk up to the feedstock and the capacity and ask three questions — who makes this, who else wants it, and how much will they pay. For turbines the answer surfaces immediately: aerospace wants it, and aerospace has 40,000 aircraft on order.

## The one thing to take away

One idea from this episode: to judge whether something will happen, find the most upstream input it shares with someone else. The data center needs power, the plant needs a turbine, the turbine needs blades, the blades need a forge slot — walk up far enough and you find a layer whose customers include people other than you, and that layer's queue sets your schedule. What makes Bruff's four-level map useful is that he turned this into something you can look up.

![A three-level tree traced from the top down, where the bottom node is also pointed at by two incoming arrows from elsewhere and is circled as a shared resource.](/figures/trace-up-three-layers-en.svg)

Here's a thing I've tried, if you want it. Take something you're pushing this week and walk three levels up. "I owe this project next month" → I need a specific person's time → that person's month is already claimed by someone. At the third level, circle the resource you share with other people, and write one sentence: if that layer puts them ahead of me, how late am I. It takes under ten minutes, and once that sentence is on paper you have an honest estimate of your own timeline for the first time.