MP Maryana Bezugla purchased an NVIDIA DGX based on the Grace Blackwell chip for her office to process sensitive data locally, without cloud services. The MP promised that the computer will be declared.
MP Maryana Bezugla purchased an NVIDIA DGX based on the Grace Blackwell chip for her office to process sensitive data locally, without cloud services. The MP promised that the computer will be declared.
Mariana Bezugla announced the purchase of a so-called AI computer for office work. The device was named Athena, in honor of the ancient Greek goddess of wisdom. The main motivation, according to the MP, is privacy: the ability to run artificial intelligence locally without transmitting data to external servers.
«Everyone who works with me can be sure that confidentiality is valued in my office, and personal requests from citizens or sensitive military developments are not quietly leaked to ChatGPT.»
Technically, Athena is built on the NVIDIA GB10 Grace Blackwell Superchip with 128 gigabytes of combined memory. The operating system is NVIDIA DGX OS based on Ubuntu Linux. The ARM architecture on which the chip runs is conceptually close to what Apple uses in its Silicon processors — but with a greater emphasis on working with large AI models and long context. Bezugla notes that the leading MacBooks are still inferior to such solutions in these tasks.
The key advantage of the device is its ability to process hundreds of pages of text within a single query without losing the quality of analysis. This opens up opportunities for locally running agent-based AI systems: digital assistants that work independently with programs and data. All without a connection to the cloud.
«In simple terms: when you make a request, the next step is the processing and „thinking“ of the model, and only then does the AI answer you. The answer can be very fast, but if you want to use AI in research and improving complex solutions, the most important stage is the quality of the analysis, the „thinking“. This minicomputer makes it possible to use a fairly large context (for example, process hundreds of pages) without losing it, and to maintain the quality bar. And all this locally,» she noted.
At the same time, the MP noted that she uses the AI service from Proton as an alternative to large providers. «It is simpler than advanced providers, but has a clear policy and encryption, similar to how chats are encrypted in WhatsApp or Signal. It is based on LLM, large AI language models hosted on the company’s servers. That is, it does not have its own fundamental model, but uses a combination of solutions available to a wide audience,» Bezugla noted.
The device costs about $4,000, which Bezugla herself compares to the price of a used car. According to her, the computer was purchased with taxpayer funds and will be used to improve the deputy’s work.
