Musk said gas turbines were sold out through 2030. Six months later, new orders face an eight-year queue.
Six months ago Musk named the real constraint on data-center power turbines, and specifically casting capacity for turbine vanes and blades. Revisiting the claim, I find the backlog hasn't cleared, the alternative power sources are more tied into the same constraint then one might expect, and the next constraint for turbines after casting is a unique metal most people have never heard of.
1The claim
On February 5, 2026, John Collison and Dwarkesh Patel sat down with Elon Musk. Most of the conversation was about putting data centers in orbit, but what was more salient for me is when Collison pushed on why behind-the-meter generation isn't a general solution. Why isn't it easier to build power plants for data centers on earth?
Musk pointed to the turbine backlog, and specified it is the vanes and blades. He said the remaining components could be sourced twelve to eighteen months before the vanes and blades, and that only three casting companies worldwide are producing them, and are heavily backlogged. Furthermore, vanes and blades are also produced by separate firms to the rest of the turbine. The final relevant remark is that turbines were sold out through 2030, and that SpaceX and Tesla would likely have to cast blades in-house.
Six months later, I want to further evaluate if the constraint sits on blades, whether it has moved since February, and what additional complications could occur if casting capacity ever expands.
2The backlog, six months later
For several months there has been a GEV gas turbine backlog, and slot reservations, as turbines are demanded as a primary behind-the-meter power source for data centers. GW demand going into 2031 and beyond is up 16% at 116 GW as of Q2 2026, projected to hit 125 GW by end of year if the buildout continues at pace.
| Metric | Value |
|---|---|
| Demand pipeline into 2031+ (Q2 2026) | 116 GW (+16%) |
| Projected by end of 2026 | 125 GW |
| New orders and slot reservations signed, Q2 | 20 GW |
| Finished turbines that left the factory, Q2 | 3 GW |
In Q2, GEV signed 20 GW of new orders and slot reservation agreements, while separately only 3 GW of finished turbines physically left the factory. The disparity between those two numbers is notable. Bookings are running far ahead of shipments. Forward orders may be an optionality bet rather than intent to buy.
While attention is on data centers, roughly 80% of GE Vernova's gas turbine buyers are still traditional utilities. Three suppliers make up 90% of the market: GEV, Siemens Energy, and Mitsubishi, per IEEFA. Each of these firms quotes delivery timelines of up to eight years.
Compared to February, the "sold out through 2030" language hasn't moved. GE Vernova's own Q2 2026 materials still describe the backlog that way. What has changed is the queue behind that number: quoted lead times on new orders now run up to eight years, meaning an order placed today wouldn't clear until roughly 2034. The headline date hasn't moved, but the line behind it has gotten longer.
3What Strazik said, and why it doesn't contradict
GEV's CEO has publicly said turbines often aren't the gating item in a three-year data center project cycle. Which would seem to undermine the whole premise I am working with here.
Upon closer inspection, it does not damage our finding. Strazik is describing project-level sequencing, which is relevant if the turbine is the last thing to arrive on a site with permitting, interconnect, and construction all running in parallel. Musk is describing component-level supply, which is what components have the longest wait-time after purchase. Both are generally true. Castings can slow turbine manufacturing while turbines don't necessarily slow an individual project yet, as other factors still take longer.
Parsing this matters, as the casting bottleneck sets the ceiling on how much gas capacity the industry can build versus other power sources, rather than solely determining the schedule of any one site.
4The bypass is narrower than it looks
This shouldn't be read as three clean options for bypassing turbine constraints. Turbines share overlapping inputs with two of the three main alternatives being deployed behind the meter. Combined-cycle plants run heavy frames, and aeroderivatives use jet engine cores whose hot sections are the same single-crystal castings from the same three foundries producing turbine blades. Of the equipment classes making up the behind-the-meter buildout, only reciprocating engines are meaningfully decoupled from that casting constraint.
Reciprocating engines are less exposed. These are scaled-up piston engines without blades, or cast hot-section components, which run on the same fuel. Some scale has also been achieved. Wärtsilä booked ~800 MW in 2025 and passed 2.4 GW cumulative data center orders by April 2026, including a 790 MW Texas order and a 412 MW Ohio hyperscale order (40 × 34SG engines).
Aeroderivatives don't have this advantage. Increased demand for them further places pressure on the casting capacity the aerospace industry is already competing for. Crusoe has ordered ~1.21 GW of aeroderivative-class turbines from Boom Supersonic, with deliveries starting in 2027 and ramping to 1 GW in 2028, 2 GW in 2029, and 4 GW in 2030, and GEV signed 61 aeroderivative units in Q2 2026 alongside 52 heavy-duty.
5The price of the bypass
By Rabobank's estimate, reciprocating engines cost ~$107/MWh, aeroderivatives cost ~$106/MWh, and combined cycle costs ~$101/MWh, against the US average industrial grid rate of ~$86/MWh.
| Source | Levelized cost, $/MWh |
|---|---|
| Reciprocating engines | ~$107/MWh |
| Aeroderivatives | ~$106/MWh |
| Combined cycle | ~$101/MWh |
| US average industrial grid rate | ~$86/MWh |
The wait time is attached to combined cycle, while aeroderivatives share the same casting constraint. So the premium to bypass the casting constraint is the additional cost of reciprocating engines over the casting-exposed options. That's roughly $20/MWh.
These aren't strictly comparable figures. The retail industrial rate embeds transmission, distribution, and utility margin, and neither number prices in the interconnect delay driving behind-the-meter buildout in the first place. But the absolute size of the operating premium is small enough to characterize the alternative as a toll, not a wall.
6What would turn the toll into a wall
A toll stays a toll while the bypass stays adequately supplied, and the underlying casting problem persists. Today, the binding input still is casting capacity, with three firms backlogged twelve to eighteen months behind the other turbine components. But rhenium, the key alloy in blades and vanes that gives them their heat-resistant properties, has already moved a lot more than a casual read of the casting story would suggest: priced against the Argus/Metal Bulletin benchmark, rhenium rose from ~$1,370/kg in 2024 to ~$6,100/kg by February 2026, and to ~$7,283/kg by August 2026. That's up roughly 440% since 2024, with about 53 percentage points of that coming in 2026 alone.
That's a striking move for a metal that, on paper, shouldn't be binding yet. Casting capacity, not rhenium, is still the stated constraint. It suggests the market may already be pricing in future scarcity ahead of the casting bottleneck actually clearing, which is worth sitting with before treating the rest of this section as a purely forward-looking conditional.
The more significant question is what happens if casting capacity expands. The constraint on blades and vanes would then shift more fully onto the price of rhenium. Because of the complexity of production, rhenium has a very inelastic supply.
Rhenium is crucial for turbine blade construction. Turbine efficiency is limited by how hot the engine can run, and temperature is limited by creep, which is the slow stretching of blade metal under load. Creep is minimized by adding rhenium to the nickel alloy to slow diffusion, which at two to six percent by weight, allows the blade to run at operating temperatures not possible for other alloys.
Primary production is about 62 tonnes a year, as a byproduct of molybdenum, which is itself a byproduct of copper. Secondary supply from spent catalysts and scrapped superalloy contributes roughly an additional 25 tonnes.
There is a meaningful recycling stream, but it is dependent on prior demand, so is fairly unresponsive to current prices. Additionally, certifying an alloy as a hot-section material takes years, further slowing down the supply chain. Supply is inelastic on the two-to-three year horizon that is highly impactful for bridge power sources.
The forecast from here is conditional, on top of a price move that's already underway. If GEV reaches 24 GW annualized by 2028 while primary rhenium output stays near 62 t and aerospace demand grows at a predictable pace, rhenium contract prices could plausibly keep decoupling upward from today's already-elevated levels, and industrial gas turbine alloy specs could shift toward low-rhenium variants to compensate for the higher prices.
If casting capacity instead stalls below 20 GW, that further decoupling won't happen, and this forecast is wrong. But it wouldn't undo the price move that's already happened.
7Limitations
There isn't sufficient data that I could find in a timely manner to make this claim quantitatively. It's reasoned from current incentives and conditions. There are several failure points that are contingent on new technologies at various points along the supply chain. For example, if other alloys are adopted, or if different casting or machining methods are adopted. It also presupposes that gas turbine buildout continues to drive new generating capacity, that other sources aren't adopted at scale in the coming years, and even that gas power remains the main bridge source rather than renewables growing in power share.