Russia Tests Sberbank's GigaChat for Generative Aircraft Design
Russia's United Aircraft Corporation is reportedly testing Sberbank's GigaChat model to help engineers explore future aircraft configurations.
According to TechRadar, Russia's United Aircraft Corporation, known as UAC, is reportedly experimenting with GigaChat, a large language model from the bank Sberbank, to help engineers developing future military and civilian aircraft. The report describes the work as generative design, in which software proposes many possible designs from a set of goals and limits.
TechRadar says the tool could help during preliminary design, the early stage when engineers compare broad aircraft layouts before committing to detailed engineering. The outlet reports that the work could eventually support advanced fighter jets. It also notes that UAC has not identified any specific project meant to challenge the American F-22.
The report places the experiment within Moscow's broader push for domestic software. According to TechRadar, Western suppliers have withdrawn, and Russia's aerospace industry faces technology restrictions, so it is relying more on software developed inside the country. The available reporting does not describe results or a timeline for the testing.
Background
Generative design is an engineering approach in which software explores a large range of options, often weighing factors such as weight, strength, drag and cost. Engineers then choose among the results. Large language models are a newer addition to this kind of work. They can summarize specifications, suggest configurations, and help organize early-stage trade-offs. Their outputs, however, need to be checked against physics-based simulation, wind-tunnel data and testing, because a fluent answer is not the same as a sound design.
Aircraft development is slow and expensive, and the early phases of a program set many constraints that shape everything after. Tools that speed up exploration in those phases can therefore carry real value, which explains why aerospace firms in many countries have adopted computer-aided design and optimization software for decades.
Since 2014, and especially after 2022, Western governments have imposed sanctions and export controls on Russia that limit access to advanced aerospace components, semiconductors and certain software. Those restrictions have pushed Russian companies toward domestic suppliers and to foreign sources willing to trade with them. Sberbank is Russia's largest bank and is closely tied to the state, and it has invested heavily in artificial intelligence.
The F-22 Raptor is a United States fifth-generation stealth fighter that entered service in the mid-2000s. It is often used as a reference point in discussions of advanced combat aircraft, which is why comparisons to it appear in coverage of new fighter programs.
Why it matters
The story sits where two trends meet: the rapid spread of AI tools into engineering work and the effect of sanctions on technology supply chains. If a sanctioned industry can use domestic AI systems to speed early design, it reduces its dependence on foreign software, but it also raises questions about what those systems were trained on, how well they perform on specialized aerospace problems, and who controls the data they use.
For observers outside Russia, the report is a reminder that export controls shape not only which hardware countries can buy but which software ecosystems they build. Restrictions can slow some programs while encouraging the growth of parallel domestic industries, whose quality and capability can be hard to judge from outside.
The story also highlights a gap between a tool being tested and a tool shaping a finished aircraft. Early-stage assistance is far from a design that flies, and the report does not suggest otherwise. Still, the practical question is how much these tools change the speed and direction of design choices over time.
What to watch
Whether UAC or other Russian aerospace bodies publish details on the testing, including results, scope and the engineering teams involved. Whether later reporting indicates which programs, military or civilian, the tool is actually used on. Whether further sanctions or export measures address AI software used in defense-related industries. And whether independent analysts can assess how the system performs against established aerospace design tools.
Viewpoints
Commentary from our columns: three angles on this story.
The Skeptic open questions and weak points
The headline asks a big question, but the reporting answers a smaller one. UAC is reportedly experimenting with GigaChat, and it has not named a project meant to challenge the F-22. That gap matters. A pilot tool for early design is not the same as a fighter program, and the distance between the two is where most weak points hide.
Several questions remain open. How well does a general-purpose language model handle the physics of airframes, engines and control systems? What data was it trained on, and how much of that data can be verified? How would anyone know whether its suggestions outperform conventional optimization software that engineers already trust?
There is also a tendency to read sanctions-driven news as a sign of rapid progress. Substitution can be real and still slow, expensive or uneven. Until there are published results, timelines or named programs, the most responsible reading is that this is an early experiment with an unclear payoff.
The Builder the engineer's view
From an engineer's seat, the useful part of this story is the workflow, not the headline. Preliminary design is a search problem with fuzzy inputs, and a language model can help frame options, translate requirements and draft trade-off tables faster than a team working by hand. That is a legitimate use.
The hard part comes next. Every candidate configuration still has to pass through aerodynamic simulation, structural analysis and eventually testing. A model that writes a plausible design is not a model that flies. The teams that get value from these tools tend to treat the output as a hypothesis and keep the verification pipeline strict.
So the test to watch is not how impressive the first draft looks. It is whether the tool shortens the loop between idea and validated result without introducing errors that only appear late, when changing course is most expensive.
The Ledger the business angle
The business logic here is familiar. When Western suppliers leave a market, domestic firms either build replacements or go without. Sberbank's involvement points to a state-linked ecosystem in which a large bank helps develop software that an industrial customer can adopt. That creates a market, but one shaped by demand that is partly political rather than purely commercial.
For the companies involved, the attraction is clear: lower dependence on foreign vendors, control over data and a product that can be iterated internally. The costs are also clear. Domestic alternatives often lag on tooling, talent and compatibility with global engineering standards, and building them takes years of spending that has to be justified.
The story is worth watching as a case study in substitution economics. Whether this becomes a durable business line or a short-term workaround will depend on funding, performance and whether the market it creates can support investment beyond the sanctions period.