The widespread availability of AI models has shifted the competitive landscape for IT companies. Since generic models are now accessible to everyone, they no longer provide a unique advantage. Companies must now focus on proprietary assets that AI cannot easily replicate, such as legacy codebases that have been battle-tested over decades and contain institutional knowledge about complex edge cases.
Specialized, fine-tuned models trained on proprietary data for narrow tasks are becoming essential. Unlike generic models, these tools offer superior performance in specific domains like industrial automation or financial infrastructure. Furthermore, the industry is shifting toward local models to ensure data privacy, as sending sensitive information through third-party APIs risks exposing internal business processes to model vendors.
The culture of open-source sharing is facing a crisis, as public code and documentation are used to train the very models that compete with their creators. Consequently, companies are likely to move toward more closed development practices, keeping their expertise under NDA. Success now depends on protecting unique data and focusing on high-stakes sectors where reliability and accountability are paramount.
In a separate development, Sber has launched SberJDK, an industrial Java platform based on OpenJDK. The company has made the base version of this solution available as open source, including it in the national registry of domestic software. The platform is offered without licensing fees for corporate use.
