Russia Is Drifting from U.S. to Chinese Technological Dependency

Summary

The Russian IT sphere was built on U.S. foundations. This was not an ideological choice but rather a matter of necessity. When the Runet emerged, no other tools were available. Starting in 2014, the Kremlin undertook to shed this legacy dependency. A decade later, the results are meager: Domestic software costs more than its Western counterparts, and foreign software still holds a significant share in key industrial sectors. The Kremlin’s push for technological sovereignty has had the effect of fragmenting and rerouting dependency rather than eliminating it. In software, import substitution has done more to ensure formal compliance with legal and procurement requirements than to create a genuinely independent technology stack. In hardware, a lack of semiconductor manufacturing capacity has made sovereignty structurally impossible.

Now, with the age of artificial intelligence dawning, the dependency pattern seems to be playing out again, only in a new form: Russian AI products are being built based on open Chinese models because they are available. The availability of Chinese open-source models now offers Russia a way around Western clouds, but at the cost of deeper alignment with Chinese technical standards, ecosystems, and strategic priorities. Moscow is becoming less dependent on Washington and more dependent on Beijing, in no small part because of U.S. policy. If U.S. policymakers do nothing to change this situation, the technology level of influence—one of the few remaining after 2022—will be lost for good.

The experience of T-Bank (formerly Tinkoff Bank) is illustrative. Its AI center developed its own family of large language models (Gen-T, including T-lite and T-Pro) and released them publicly in late 2024[^1]. They are based on the Alibaba Qwen 2.5 model family and incorporate continual pretraining and alignment techniques, having been trained on Russian-language data from Common Crawl free data, books, code, and proprietary datasets. The subsequent model (T-Pro 2.0) moved to the Qwen 3 architecture[^2]. T-Bank admits that building on existing models saves up to 90% of financial costs versus training from scratch. The strategic implication is clear: Even Russia's most technically sophisticated private institutions are now building their AI stack on Chinese foundations.

I. Introduction

In April 2026, Palantir published a summary of CEO Alex Karp's book The Technological Republic[^3]. It sparked a wave of criticism from philosophers, politicians, and journalists around the world, who decried it as promoting "techno-fascism," while others described it as a sober account of reality. The controversy surrounding the book revealed something more important than its ideological foundations: the idea that digital technology has become a new dimension of geopolitics—no less significant than oil or nuclear weapons.

In this war, Russia found itself on the front lines earlier than most countries. The two waves of sanctions in 2014 and 2022 posed a question that remains unanswered: Is technological sovereignty possible in an era of global value chains?

To understand the current state of Russian IT, one must go back to the moment of its birth. The Runet emerged in the early 1990s on the same technical foundations as the rest of the global internet, with open TCP/IP protocols, servers running Unix-like systems, and Oracle and Microsoft databases. Soviet engineers were not incompetent; the Soviet mathematical school was among the strongest in the world. They used the tools that were available and worked.

The Soviet microelectronics industry collapsed along with the USSR. Many key enterprises ended up outside of Russia, in Ukraine, Belarus, and the Baltic states. The production chain of the industry was severed, and the technology gap with the West in semiconductors was estimated to be at between 10 and 15 years[^4].

Russian IT was strong in algorithms, since no physical hardware was needed, and weak in infrastructure. Yandex became a local competitor to Google in search; VK carved out a niche in social media; and Kaspersky Lab competed in the global antivirus software market. But servers, processors, cloud platforms, and enterprise software all came from the West. By various estimates, in 2014 Russian customers spent around RUB300 billion annually on IT infrastructure, of which roughly 90% went to foreign vendors[^5].

II. Sovereign Software

The First Impulse for Sovereign Software: 2014–2016

The annexation of Crimea, followed by the first wave of Western sanctions in 2014, became a moment of reckoning. The Russian state discovered that critical infrastructure, from government-agency servers to banking systems, was running on software from companies based in countries whose governments were coming to see Russia as an adversary. At that time, foreign software accounted for 64% of total software spending in Russia[^6].

In April 2015, the Ministry of Communications approved a Software Import Substitution Plan (Order No. 96), which claimed that the market share of foreign software exceeded 50% in virtually all categories, and in the most critical ones like operating systems, database management systems, and industrial software, the figure was even higher[^7]. The exceptions were search services and antivirus software, where Russia had competitive players in Yandex and Kaspersky.

In June 2015, a law creating a registry of domestic software was passed, coming into effect in early 2016[^8]. It essentially represented a government list of software officially recognized as Russian. Inclusion allowed a product to receive preference in state procurement, while government bodies were generally barred from buying foreign software when a suitable Russian product appeared on the list. The criteria for inclusion were straightforward: The rights holder had to be a Russian individual or organization with no more than 50% foreign ownership, and license payments abroad could not exceed 30% of revenue. The law contained a significant loophole, placing no restrictions on the use of open-source code regardless of its country of origin. Open-source code involves no licensing fees, meaning it automatically was eligible. The registry thus created an incentive for achieving independence from specific vendors but not for creating original developments[^9].

In March 2026, new requirements came into force introducing separate eligibility criteria for AI software, alongside obligations to disclose license payments to foreign entities. Meanwhile, Federal Law No. 325 from July 2025 created three separate registries to replace or supplement the original, with the declared goal of distinguishing genuinely domestic products from formally compliant ones. The original registry had become cluttered with software that passed the ownership test without reflecting any meaningful domestic technological capability. Companies that secured inclusion largely as a formality, without developing substantive expertise or products in the relevant field, may struggle to meet the new requirements. The open-source loophole is finally being closed. However, the fact that that took a decade to do is an indication of how slowly the system adapts.

Initially, the Russian tech industry showed little interest in the registry: The only incentive was eligibility for government procurement. Only state agencies were required to purchase only software on the registry, with exceptions allowed only when no domestic alternatives existed.

New Tax Rules Change the Game: 2020–2021

The situation shifted in 2020, when amendments to the Tax Code restructured the tax status of IT companies[^10]. Before 2021, transfers of software-use rights under license agreements were generally exempt from VAT regardless of whether the software was in the registry. From January 1, 2021, this broad exemption was replaced by a narrower exemption covering software and databases included in the registry. This forced the IT industry to restructure—and drove a sharp increase in registrations already beginning in 2020. However, growth in the registry did not mean growth in actual development: Companies were optimizing legal structures, not writing new code.

The Ministry of Digital Development itself only switched from Microsoft Teams to the VK corporate messenger in 2021, a full six years after the import substitution program began[^11]. The cost came to RUB487.7 million, making it the most expensive IT import substitution project of that year.

Post February 2022: An Ultimatum Served 

The full-scale invasion of Ukraine in February 2022 and the second wave of sanctions transformed the situation. Western companies began exiting the Russian market, with Microsoft, SAP, and Oracle suspending deliveries and support. A Russian presidential decree in March 2022 introduced a complete ban on the purchase of foreign software not only for government agencies but also for "subjects of critical information infrastructure," a broad category of private companies in the healthcare, finance, energy, transportation, nuclear, and defense sectors[^12].

These companies had built their infrastructure with little thought given to potential sanctions and import substitution. Changing a technology stack is a difficult undertaking even in peacetime; under conditions of war, workforce shortages, and sanctions pressure, it proved to be beyond the realm of the possible.

Starting in 2026, a compatibility requirement with Russian operating systems was added to the registry criteria[^13]. It has exposed a systemic problem: Although individual products can be replaced, Russia’s domestic software ecosystem was never developed as an integrated stack. Only a small fraction of the software listed in the registry is compatible with Russian operating systems, which makes transitioning to an entirely domestic technology stack even more burdensome.

According to a 2024 K2 Cyber Security study, the experience of 30% of companies was that "import substitution" effectively meant switching from Western to Chinese software[^14]. In other words, dependency was not eliminated; it merely shifted from one supplier or technology to another.

III. Sovereign Hardware: A Nonexistent Foundation 

If the failure of import substitution in software can be attributed to inertia and the complexity of switching stacks, the situation with hardware is even more hopeless for Russia, owing to the physical absence of manufacturing capacity.

Chips: Sovereign Design, Foreign Production

Before the full-scale war, Russia had two companies developing microprocessors for servers and client devices. MCST produced Elbrus microprocessors based on a proprietary architecture, while Baikal Electronics produced the Baikal line of microprocessors based on ARM architecture. Both families were designed in Russia but manufactured in Taiwan, at TSMC facilities.

Immediately after the invasion of Ukraine, TSMC suspended shipments[^15]. Then, in June 2022, Taiwan officially banned the export to Russia of microprocessors with clock speeds above 25 MHz, ALU widths exceeding 32 bits, and other characteristics possessed by all modern microprocessors. MCST and Baikal Electronics had no alternative manufacturing partners, and switching manufacturers requires essentially a complete redesign of the chip.

Baikal Electronics was forced to freeze production of its Baikal-S server microprocessors. Meanwhile, more than 300,000 microprocessors, already or partially produced, remained stranded in Taiwan[^16]. Minister of Digital Development Maksut Shadaev acknowledged in December 2022, "The intellectual property and all documentation are Russian, but there are no manufacturing facilities in Russia capable of producing these processors[^17]."

The Russian government explored moving Baikal and Elbrus production to the Chinese factories of SMIC and Hua Hong. However, Chinese manufacturers, fearing U.S. secondary sanctions, are unwilling to produce chips for Russia.

Data Centers: A Capacity Deficit

The chip problem is compounded by shortcomings in Russia’s data-center infrastructure. As of April 2024, the aggregate grid-connected capacity of Russian data centers, excluding unregistered cryptocurrency-mining operations, was estimated at approximately 2.6 GW, about 1% of the overall generating capacity of Russia’s energy system[^18].

The capacity shortage is acute. In November 2024, major banks such as Sber and T-Bank (formerly Tinkoff) reported increasing difficulties with finding sites for data center construction in Moscow and St. Petersburg: It can take up to three years to locate land for a 60 MW facility. According to analysts from Softline, a digital transformation and cybersecurity solutions provider, companies are being forced to build their own data centers due to the shortage of commercial capacity. The cost of rack space at Moscow data centers rose 35% in 2024 alone[^19].

Sanctions complicate construction: GPU accelerators, server equipment, and switching hardware have all come under restrictions. Gray-market imports are a realistic but unstable workaround. Companies pay intermediaries to broker deals, but the resulting infrastructure ends up more expensive and less reliable than in unsanctioned jurisdictions.

Training and running large language models requires thousands of modern GPUs. Without domestic chip production and without the ability to import Western accelerators legally, Russian companies find themselves in a structurally vulnerable position. They can fine-tune existing models on small clusters, but they cannot train them from scratch.

IV. Sovereign AI

In early 2025, China’s DeepSeek released a series of open models, DeepSeek-V3 and DeepSeek-R1, which on several benchmarks matched the best proprietary systems of OpenAI and Anthropic. They were released under the MIT license, meaning they are available for free use, modification, and commercial application. This triggered an immediate stock-market reaction as Nvidia shares plummeted 17% in a single day amid fears that open models could undermine the value of expensive GPU clusters[^20].

DeepSeek's openness is a strategic choice by China, which is doing several things simultaneously. First, it is creating ecosystem dependency: Developers around the world are building products on Chinese models, sharing expertise, publishing adaptations, and creating documentation. Switching operating bases becomes costlier with each passing month—as was the case with U.S. open-source technologies in the 1990s. Second, China is reducing the market share of the proprietary systems of its main competitor, the U.S. Every company that switches to DeepSeek over ChatGPT represents lost revenue for OpenAI and Microsoft. China can thus shape norms and standards. DeepSeek models, as researchers have noted, kowtow to the official line of the Chinese Communist Party in their responses to sensitive topics.

For Russia, DeepSeek and Alibaba's Qwen family of models have become an obvious way out. Open weights are precisely what can be run locally, without dependence on U.S. clouds and the threat of disconnection. Some of the most prominent “Russian” AI products that appeared in 2024–2025, including models from Yandex and Sber, are reportedly based on Chinese models that have been additionally trained on specialized data rather than built from scratch using original architectures[^21].

V. How the United States can preserve technological influence

It is commonly assumed that total technological isolation of Russia serves U.S. interests. This logic is understandable but mistaken. If Russia moves definitively into the Chinese technological orbit, the United States will lose one of its few remaining channels of influence and will have to deal with a durable Russia-China connection in AI infrastructure that will produce standards and products.

Technology is not just a business relationship but a normative one as well: Companies using U.S. tools operate within the legal and technical space defined by U.S. standards. This channel has already narrowed considerably; the question is whether it will close entirely.

Recall that it was because Russia had at one point grown into the U.S. technology stack that Washington had some leverage over Moscow in 2022, specifically the ability to disconnect or restrict access to software. If the next generation of Russian IT infrastructure is built on Chinese foundations, the United States will no longer have that leverage.

Open Weights 

The United States needs to develop a consistent position on export controls with regard to open AI models. Currently, Washington restricts GPU shipments to Russia—which is logical from a military-application standpoint—but this only accelerates Russia's shift toward Chinese open models. One possible measure is the active creation and release of U.S. open AI models (modeled after Meta's LLaMA) that would be accessible to developers working on civilian applications. This would help preserve competition in the civilian application segment. 

Policy Predictability

The key fear among Russian developers regarding Western tools is not their cost or their quality but the possibility of sudden disconnection. They only need to look back to 2022: Microsoft, Oracle, and SAP left within a matter of months.

If the United States wants to maintain a presence in the Russian technological space, it needs to establish a predictable regime for open-access tools. Models with open weights cannot be "switched off"—representing their fundamental distinction from cloud services. This is why open AI models (rather than closed APIs) are the best instrument for preserving technological influence.

The logic of sanctions tends toward expansion: Since any technology potentially has military applications, an ever-wider range of technologies becomes subject to restrictions. Though understandable from a security standpoint, this is counterproductive from the perspective of long-term influence.

A more practical approach would involve a clear distinction between technologies with direct military applications (e.g., GPUs for computing clusters used in targeting systems) and civilian AI technologies (e.g., language models for businesses). The latter category deserves more flexibility. It is there that the ecosystem preferences of the next generation of Russian developers are being shaped.

Conclusion

The history of Russian IT is a story of how technological dependency reproduces itself regardless of political ambitions. In the 1990s, there was no choice: TCP/IP protocols and Unix servers were simply what was available. The current drift toward Chinese infrastructure is equally structural, representing not a policy decision but the path of least resistance under sanctions pressure and without domestic manufacturing capacity.

Import substitution attempts did not fail because of sabotage or incompetence. A technology stack is an ecosystem encompassing manufacturing capacity, supply chains, trained specialists, documentation, patents, and years of accumulated investment. Building one from scratch would be hard enough, even if there were no war or sanctions.

In the age of AI, this reality becomes magnified. Artificial intelligence is not just another category of software; it is infrastructure on which medicine, finance, education, and governance are increasingly dependent. A country whose AI infrastructure is built on Chinese foundations is subject to Chinese standards and priorities about availability—in the same way that Russia through the 2010s depended on the United States.

For the United States, the choice in Russia policy between “complete isolation” and “no restrictions” is a false one. A third option exists: maintaining a presence in the Russian technological space through open tools and predictable policy while limiting military applications. The technology channel is one of the few remaining levers of normative influence over a country geopolitically drifting toward China.

The structural constraints mean that Russia will remain dependent on foreign technology. The only question is whose.