Since the beginning of the year, public sector spending on servers for working with AI amounted to 1.7 billion rubles.



During January-September 2024, the demand for servers for working with artificial intelligence (AI) doubled in the public sector, spending on them amounted to 1.7 billion rubles. State-owned companies use such specialized computers in industry, medicine and for video analytics. Experts emphasize that for now servers operate mainly on foreign solutions, and due to supply restrictions, the cost of such equipment will rise.

In 2024, in the Russian public sector, the demand for servers for working with artificial intelligence (servers with graphics accelerators) has doubled, a representative of the computer equipment manufacturer Graviton (part of the 3Logic Group) told Kommersant. According to the company, if in January-September 2023 only 16 such tenders were held for government procurement (under 223-FZ and 44-FZ) in the amount of 930 million rubles, then for the same period of this year the amount was already 1.7 billion rubles . in 30 tenders.

The most frequent requests are for video analytics, including complex ones (in real time), says Alexander Filchenkov, director of the server and network systems department at Graviton. In the public sector, complex AI solutions are used primarily for data mining and decision-making systems, he says. “Almost a third of Russian companies have announced plans to implement AI within three years, so we can expect further growth in demand for this equipment,” notes Alexander Filchenkov.

In April, Vedomosti reported that the demand for renting servers for training AI models had increased, but the publication did not specify exactly how much. The revenue of data centers from the rental of such equipment in 2023 amounted to 6.6 billion rubles, estimated by iKS-Consulting, the main clients were Cloud.ru, Yandex, Croc, MegaFon and others.

Many industries have begun to move from accumulating data obtained as a result of digitization to their processing, use and making forecasts using AI, explains Irina Ostroukhova, director of partner relations at RecFaces (developer of biometric solutions). “In the segment related to facial biometrics, which must process video streams in real time, demand has grown due to projects for the development of the state Unified Biometric System and commercial projects,” she says.

However, the segment also has limiting factors. Among them, Graviton notes that servers for AI work mainly run on solutions from foreign Nvidia, which, when leaving the Russian Federation, limited access to its technologies. The company adds that Russian organizations, due to growing demand, have begun to consider cooperation with Chinese and domestic developers of neural network processors, among them the company Iva Technologies. In 2018, it announced the start of development of a tensor processor, CNews reported. In 2020, the company presented its solution at the Embedded World conference.

The HiTech company (formerly Iva Technologies) reported that their processors are produced under the LinQ brand. “The demand for computing resources is growing in the market as a whole, including industries that actively use AI technologies,” says a company representative. According to him, their processors are mass-produced; he did not specify where exactly.

At the same time, purchasing the necessary systems is quite difficult, reminds ARPE Executive Director Ivan Pokrovsky: in general, payment terms and the cost of logistics when purchasing foreign solutions are constantly growing.

According to Kommersant’s interlocutor among electronics manufacturers, the demand for servers for AI operation is provided mainly by large companies such as Sber and Yandex. “So far in the Russian Federation there are no proprietary solutions comparable to the level of Nvidia or other companies. However, despite the restrictions, the number of purchases and investments will grow due to the technological development of artificial intelligence in the Russian Federation,” he emphasizes. Yandex and Sberbank did not respond to Kommersant’s request.

Timofey Kornev



Source link