ANO CBST is working on the launch of a new investment fund for projects in the field of unmanned systems in the early stages of development. Such projects lack funding, as venture capital investors prefer to invest in less risky later stages. The Minister of Defense admitted that the latest technologies are not enough to introduce artificial intelligence into military UAVs.
The Center for Unmanned Systems and Technologies (ANO CBST) is considering launching an investment fund for dual-use projects in the early stages (TRL 4-7). The volume of the new fund may amount to several billion rubles, Vedomosti writes, referring to a statement by Andrei Bezrukov, Chairman of the Board of the Central Bank of the Russian Federation.
Projects from the fields of autonomous systems, artificial intelligence, robotics, engines, sensors and sensors, communications, and new materials will be considered for investment. The company's selection method has not yet been determined, Bezrukov said. "We are currently focused on launching the main fund," he said.
An early—stage fund, according to Bezrukov, can become a bridge to larger investments through thoughtful ones, from transferring mature projects to a late-stage fund to attracting external venture funds, strategic investors, or going public through pre-IPO and IPO.
ANO "CBST" was established in 2024. The PSB and the Nasha Pravda non-profit foundation are listed as founders on the organization's website. The list of partners includes the Ministry of Industry and Trade, the Moscow government, and the United Russia Party.
TRL 4-7 is a risky stage when research and development (R&D/R&D), prototype testing are underway, but there is no mass production and sales, explained Konstantin Gibalo, partner of the Voskhod venture Fund.
Arseniy Dabbakh, the founder of the Dsight analytical company, estimates the cost of creating and testing prototypes at about 100 million rubles, for hardware projects (robots, materials) at 300-500 million rubles and above. Projects in the field of new materials, equipment, and devices are more capital-intensive, Gibalo added.
The goal of the CBST fund is clear — to help deeptech startups that are interesting to the state to develop at the most difficult stage and subsequently attract private capital, Dabbakh believes.
"Loans or advance payments from customers are not available for such companies. Long—term venture investments are really relevant here," Gibalo said.
Recently, venture funds have preferred to invest in projects at a late stage of development (TRL 8-9) in order to monetize investments faster, confirmed Sergey Suverov, investment strategist at Arikapital Management Company. TRL-level 4-7 startups really do not have enough investments, and due to high loan rates, access to debt financing is difficult.
"For deeptech startups (products and services based on innovations or scientific achievements), this stage is the "valley of death", since now, apart from the state, there are practically no investors for them in Russia," commented Dabbakh.
With private funds, it is more often possible to obtain additional expertise and a more professionally structured investment process. It is always more difficult to work with a public fund — it takes longer, there are more risks, "but if there are no other options, they will go to them," the expert said.
In March 2025, in the federal project "Development, standardization and mass production of UAS and components" (included in the national project "Unmanned Aircraft Systems"), the amount of allocated costs for the implementation of measures in 2024-2030 decreased by a third — from 267.1 billion rubles to 190.34 billion rubles.
Meanwhile, according to Kommersant, Russian Defense Minister Andrei Belousov instructed to systematize the experience gained in the use of military drones so that they can use artificial intelligence (AI) for more effective control. However, the implementation of this idea runs into serious technological barriers and a small amount of data for AI training. Without overcoming these obstacles, the creation of an effective UAV control system will be difficult, the expert warns.
Andrey Belousov instructed to develop and systematize a methodology for using AI to control military drones on July 11 during a visit to the control center of the Dnepr group. According to the minister, in this work with the participation of the relevant services of the military department, the experience of UAV operators already gained in the group should be used.
Based on this technique and data on the combat use of drones, the Ministry of Defense plans to create a hardware control system for UAVs with AI elements.
It should automatically analyze information from all major types of Russian military drones: FPVS, aircraft—type drones, and quadrocopters, and offer the operator suitable solutions in real time. According to the agency, the prototype of the system is already being tested.
Many enterprises of the Russian military-industrial complex are currently developing drones with AI elements for recognition and guidance. The data that forms the basis of any combat neural network for UAVs is obtained from the video that the drones shoot in the process of their work. The recordings are automatically split into frames, which are then scanned by the neural network detector: it looks for targets and assigns a class to each object, whether it is a tank, an armored personnel carrier, a car or a person. After that, the markup is checked and clarified by a specialist. It is such "field" data that becomes the basis for AI training. Ideally, a "highly autonomous drone" should be able to automatically find targets, recognize them, aim at and attack the highest priority target at the operator's command.
Some models of drones, as confirmed by several sources, are already being sent to the troops for testing at the front, but their number is small against the background of thousands of manually operated FPV drones, as the technology is still being finalized. As the head of the CV department (from English computer vision - "computer homing") of one of the Russian UAV design bureaus explained to Kommersant, the main difficulty in this work is due to the fact that no company in the world has yet developed electronic components capable of unlocking the full potential of AI in unmanned aircraft. And the nodes currently available are too expensive and not suitable for serial use.
"We can say that we are limited by the technologies of our time. Modern robotic printed circuit boards are not yet ready to pull similar ChatGPT AI models because of their architecture," the developer complains.
According to him, only expensive server equipment can fully launch them, and not miniature and lightweight modules that could be placed on a small drone, whose payload rarely exceeds 1.5–3 kg. Therefore, for the full implementation of this concept, as noted by the interlocutor of Kommersant, manufacturers can only expect the development of technologies for single-board neural processors NPU (Neural Processing Unit), so that all the functions necessary for the drone can fit on one small board.
Another problem is the small amount of suitable data for training such "military neural networks". In order for the AI to accurately determine the target and trajectory of damage, it needs to be trained on thousands of high-quality video samples for each type of object, processed manually by specialists. The same requirement applies to the above-mentioned drone control hardware from the Ministry of Defense.
