AI Era Bottleneck Is Power, Turbine Blade Production Is the Problem

Hello, this is ElectroChem. When we think of AI data center bottlenecks, we usually first think of GPUs. However, a few days ago, Elon Musk pointed to a completely different area, presenting a new problem. With music, Musk mentioned that one of the factors hindering natural gas turbine production is the casting of turbine blades and vanes.

Musk stated that casting blades and vanes is a critical bottleneck and that SpaceX would handle the casting directly. Regarding why a rocket company would produce metal blades for power plant turbines, Musk projected that internal casting at SpaceX could advance the actual operational start of gas turbines by up to 18 months. If this prediction is correct, the AI era's bottleneck has expanded from GPUs to power, from power to gas turbines, and then further into the very difficult components within the turbine itself.

Now, let's examine how far this trend is confirmed and what truly makes it difficult. First, regarding the power situation, the U.S. Department of Energy analyzed that U.S. data centers consumed about 4.4% of total electricity in 2023, and this could increase to between 6.7% and as much as 12% by 2028. This change reflects not only AI but also cloud and manufacturing demand. As power-intensive facilities increase simultaneously, large data centers are moving beyond merely waiting for power connections to securing their own power sources.

Even if solar power or storage systems cover some of the electricity, an instant-output power source is essential when there is no sun or when demand surges. This highlights the renewed importance of gas turbines. Gas turbines serve as adjustable power sources that can generate electricity quickly. However, the problem is that these units cannot be obtained immediately next month simply by placing an order. According to Gibbonova, the backlog for gas equipment and production slot reservations increased from 100GW to 116GW in Q2 this year, with an expected rise to at least 125GW by year-end.

Gibbonova announced plans to expand production capacity from 20GW in 2026 to 24GW in 2028 and 30GW in 2030. However, these plans themselves demonstrate that increasing turbine supply is a very difficult task. This means it's not a structure where you can open a factory and double production next week just because demand has increased. Up to this point, we've discussed the issue of turbine shortages. But Musk's comments went a step further, identifying the casting of blades and vanes within the turbine, rather than the entire turbine, as the limiting factor.

Blades are rotating airfoils, and vanes are stationary airfoils that efficiently direct gas to the next blade. The front blades that compress air and the turbine blades that extract energy after combustion perform different functions. The core part we will discuss today is the hot section of the turbine, which directly contacts the hottest gases. Although it appears to be a simple metal blade, it actually has complex internal passages for air, small cooling holes on the surface, and a protective thermal barrier coating on the outside. Data from specialized hot-section component companies like Howmet Aerospace illustrate this complexity.

The high-pressure turbine section can operate in environments much hotter than the metal's melting point thanks to cooling technology. Although the combustion gases are extremely hot, internal cooling systems and surface protection coatings keep the metal temperature within tolerable limits. One of the key technologies here is the ceramic core. Before the casting process, ceramics are inserted into the shape of the internal cooling passages, and metal is solidified around them. When the ceramic core is later removed, invisible air passages remain. Even slight variations in the position of these passages can significantly affect wall thickness and cooling performance.

These passages are not simply drilled later; they involve forming complex cooling structures within the metal beforehand. The method of solidifying the metal is also crucial. Conventional metals consist of multiple crystals touching each other, and at the hottest parts, these grain boundaries can act as weaknesses when subjected to stress at high temperatures for long periods. Therefore, directional solidification, which makes crystals grow in a specific direction, and even single-crystal casting technology, which essentially creates the entire component as a single crystal, are used where needed.

Single-crystal casting is a key manufacturing technology to reduce creep, where metal slowly deforms or stretches at high temperatures. The required materials and cooling methods vary depending on the engine's position and temperature, so single-crystal technology is not applied to all blades but is a critical option for ensuring component longevity under the most demanding conditions. Therefore, simply saying 'cast a blade' oversimplifies the actual complexity on site. It requires creating wax patterns and ceramic cores, precision casting, and then solidifying the metal while controlling the crystal structure.

Then, heat treatment, machining, cooling hole processing, coating, non-destructive testing, and dimensional inspection must all be precisely completed before the part can be installed in an engine. An error in one stage is nearly impossible to fully correct in the next. The real bottleneck occurs here. Successfully producing a single component is entirely different from consistently producing thousands with identical quality. Various problems can arise during casting, such as porosity, shrinkage, distortion, and changes in internal core position. Components with controlled crystal structures are even more sensitive to subtle issues after heat treatment. ASA's single-crystal turbine airfoil data also discusses inspection and recrystallization issues before and after casting and heat treatment.

Commercial yields are not publicly disclosed, but the key is that simply installing one more piece of equipment does not immediately increase production. The ability to consistently produce good parts through process control and an inspection system must first be established. This is not because the process is complex, but because that complex process must be reproduced exactly the same way every time. In other words, consistently producing the same quality means the inspection results must also be identical every time. The next step is 'qualification.' This is the process of confirming that the material, casting process, coating, and machining comply with the design.

After verifying that it's built according to design, it undergoes thermal vibration, rotation, start-up, and shutdown processes in an actual engine. Finally, long field operating hours are required. 'Technically feasible,' 'stably mass-produced,' 'engine-verified,' and 'long-term operated' are not the same thing. The U.S. has a deeply integrated chain. There are engine companies like GE, and a supply chain specialized in cooling coatings and large directional solidification blades, like Howmet Aerospace, is established.

This is an industry where materials, casting, machining, coating, inspection, and engine companies have been closely intertwined for a long time. The strength of the U.S. lies not just in the technology to make a single blade, but in the fact that this entire chain is already robustly connected. Japan presents another standard. Mitsubishi Power's J-series boasts over 3.2 million hours of accumulated commercial operation by the end of 2025. IHI Casting showcases complex hollow structures and directional solidification single-crystal precision casting technology. Japan's strength lies not so much in the technical term 'single-crystal' itself, but in the organic connection between component technology and the long operating history of large gas turbines.

So, how far has Korea come? Doosan Enerbility completed maximum speed and maximum load tests for its self-developed 380MW gas turbine in 2025. According to company announcements, Korean-made gas turbines have accumulated 17,000 hours of demonstration operation since 2019. It also operates hot section component maintenance and turbine testing infrastructure. This year, Doosan disclosed a contract with HD Hyundai Marine Engine for blade material development using gas turbine precision casting manufacturing technology, showing that Korea is also actively addressing issues in actual engines and hot section precision casting.

Korea is also addressing issues with actual engines and hot section precision casting. The question now becomes more specific: has the highest difficulty first-stage blade of a specific 380MW engine been repeatedly mass-produced domestically with high yield, and has that component been linked to long-term field qualification? Public data shows precision casting, prototype development, and design and quality evaluation roles, but the most difficult final stage of public verification is still insufficient. Therefore, Korea's narrative should not stop at 'we can make it'.

We need to understand how far every link in the chain—casting, cooling, coating, inspection yield, engine verification, and field operating hours—is established. Returning to SpaceX, its strengths are clear: high-temperature metals, turbomachinery, rapid manufacturing iteration, and vertical integration capabilities. Musk has stated that SpaceX will perform casting internally. Therefore, SpaceX has the potential to contribute to rapidly increasing part of this supply chain, specifically casting capability. These strengths could be particularly advantageous for quickly refining the casting process. However, the subsequent test lies in the unique domain of power generation gas turbines.

While rockets demand extreme performance and rapid development, power generation turbines must withstand creep, oxidation, and thermal cycles for thousands or tens of thousands of hours and operate stably within maintenance schedules. SpaceX's manufacturing experience can certainly be an excellent starting point. However, the decisive factor is the long-term lifespan of power plant turbines and the mass production yield to consistently produce the same quality. Ultimately, the core of this story is not about a small blade. When AI demands more electricity, a single gas turbine requires hundreds of complex components and a long chain of verification.

Among these, blades and vanes are critical components where several technical challenges — high-temperature materials, casting, cooling, coating, inspection yield, and certification — converge. Elon Musk's reason for establishing casting capabilities within a rocket company now becomes clearer. This is because the AI era's bottleneck is not merely limited to chip production, but is broadly spread across the entire industrial system that actually produces electricity. Thank you for watching this long video.