AI needs power. Rockets need thrust. Both run on the same impossible blade.

Gas turbines are the fallback power source AI data centers are scrambling for. Deep inside each machine, a few hundred blades spin in 1,400°C gas, under centrifugal forces thousands of times their own weight, for tens of thousands of hours. The same class of superalloy spins inside rocket turbopumps, and reusable rockets are multiplying the demand.

Ordinary metal softens at that temperature. These blades must be single-crystal superalloy: the entire part is one continuous grain, with no internal boundaries to crack first.

The economics are brutal:

• Materials: 15–35% of cost

• Molds and ceramic cores: 15–20%

• Casting: 30–40%

• Inspection: 15–25%

The real cost driver is scrap. One stray grain ruins the blade. Actual yields run 30–60%, and every 10-point improvement cuts unit cost by 15–20%.

Two US companies, PCC and Howmet, control roughly 70–80% of the high-end market, and likely more in single crystals. Howmet's capex has long stayed near 5% of revenue. The furnaces are extremely expensive and single-purpose; idle capacity means heavy depreciation. The rational move is to ration orders rather than over-invest. Supply is almost perfectly inelastic.

Now demand is tightening. Wood Mackenzie expects gas turbine demand to peak in 2026, with developers racing for about 63 GW of new capacity through 2030. Equipment prices are up 195%. Howmet's industrial gas turbine revenue grew 39% year-on-year in Q1, with a target of doubling that business within three to five years.

The chain looks like this: AI needs compute, compute needs power, power needs gas turbines, and gas turbines need blades almost nobody can mass-produce. The moat sits at the far end of the chain, built from decades of accumulated know-how. Money alone cannot buy it.

On October 21, Nature published work from a team at the Beijing Institute of Technology about titanium-alloy single crystals about one meter tall, grown directly on an ordinary metal plate. No mold, no seed crystal, no directional-solidification furnace. Wire is fed in pulses of alternating current, and the crystal builds itself layer by layer.

Read against the cost table, the method attacks exactly the mold and casting-equipment lines. In other words, the entry barrier itself.

One boundary matters. The paper validates titanium and refractory metals, not the nickel-based superalloys that turbine blades actually require. The hottest seats in the engine still wait. Reported attempts at nickel-based single crystals by 3D printing remain partial: a single crystal in the center and a polycrystal at the edges.

If this route matures, "capacity" stops meaning the number of single-crystal furnaces and starts meaning process parameters that people iterate on. The scarce asset two companies have held for half a century changes its nature. It would also change what aerospace engines can afford: small-batch production is exactly where the mold economy never worked.

The prospect is not here today. Nickel is the gate, and the gate is still closed.

If you work in equipment investment or industrial research, which year would you start pricing this route into industry-structure models?

(Verified as of Oct 26, 2026. Cost structures and yield ranges from public industry analyses; market shares and capacity cycles from broker and industry research; and technical parameters from the Nature paper published Oct 21, 2026.)

LU FRONTIER STRATEGIES

© 2024-2026 LU Frontier Strategies. All rights reserved.