Modern global conflict is increasingly fought not on battlefields, but in the sphere of control over critical technologies and raw materials. Rare earth metals, once narrowly specialized components, have today become the foundation for the entire microelectronics industry, determining the balance of power in production ranging from consumer electronics to supercomputers. In this context, the actions of China, which holds a dominant position in the market, go far beyond simple trade operations. This is no longer just commerce, but the implementation of a comprehensive geoeconomic strategy, where the export of raw materials is used as a lever to accelerate internal technological development and weaken the positions of competitors.
For the Russian market and other players building an independent technological policy, these processes serve as a powerful catalyst. The accelerating fragmentation of the world into isolated technological blocks and the sanction race make betting on Western solutions increasingly risky. At the same time, the consolidation of domestic demand in China and forced import substitution create conditions for the rapid maturation of alternative vendors. Thus, the current instability in the GPU and components market is not only a challenge but also a historic opportunity for the formation of sovereign computing infrastructure based on a pragmatic approach to choosing suppliers and technologies.
China increased exports of rare earth metals
China increased exports of gadolinium and germanium (rare earth elements critical for chip and laser production) in June amid escalating trade conflict with the US. But at the same time, Beijing is preparing countermeasures to control the export of rare earth metals.
Rare earth metals are a key input for the production of magnets in hard drives, lasers in photolithography, and DC-DC converter components in GPU servers. If China tightens export control, the production of servers with Chinese GPUs will accelerate (local raw materials will naturally reduce dependencies), but component prices may rise.
China’s maneuver of increasing exports at the peak of conflict is a classic example of economic counterplay. By creating a temporary oversupply on the world market, Beijing solves a dual task: first, it allows it to fix its market share and keep dependent consumers in its ecosystem, and second, a sharp subsequent turn to strict control and licensing of supplies will have a much stronger shock effect. For Western tech giants, this creates a trap: the short-term availability of raw materials only masks the strategic vulnerability of their production chains, making the upcoming restrictions even more painful and unpredictable.
These are indirect statements for the Russian market that the supply chain is shortening. The procurement horizon for GPU servers needs to be extended to 3–4 months.
At the same time, for the internal Chinese market, these fluctuations play the role of a catalyst. Guaranteed access to its own raw materials at stable prices forms a unique «greenhouse» environment for national GPU manufacturers (such as Enflame, Baidu Kunlun). This allows them to exert aggressive price pressure on the global market, offering competitive solutions whose cost price does not depend on the volatility of global rare earth metal prices. Thus, Beijing is not just responding to sanctions, but using them as an argument for forced import substitution, turning its raw material monopoly into a foundation for dominance in adjacent high-tech industries. For the end consumer in Russia and other countries, this means an inevitable rise in equipment prices during the transition period, until new, shorter, but not yet debugged logistics routes become the norm.
AMD accelerates RDNA 5 to mid-2027
AMD plans to release the RDNA 5 architecture in mid-2027, rather than 2028 as previously announced. The reason is the global DRAM deficit, which has slowed both camps (NVIDIA also postponed consumer cards).
In the enterprise segment, AMD competes with NVIDIA in data centers through CDNA/Instinct (MI300X, MI350). RDNA is a consumer line, but this is a signal: AMD feels a window of opportunity and is trying to capture market share before NVIDIA restores its supply chain.
AMD’s decision to move the RDNA 5 release to an earlier date is a classic example of using others’ weaknesses for one’s own gain. While the entire industry, including NVIDIA, is paralyzed by the DRAM deficit and forced to postpone launches, AMD sees not an obstacle, but an opportunity. By accelerating development, the company aims to deliver a preemptive strike. This is especially relevant in the context of the expected release of NVIDIA RTX 50 Super. If AMD can present a competitive product in mid-2027 while NVIDIA has no direct response, it will gain a critically important advantage in the fight for the attention of gamers and PC builders, capturing a market share that will be extremely difficult to win back later.
For Russia, the consumer GPU market is not our market. But if AMD accelerates its roadmap, it means competition is intensifying. This is good for the prices of Chinese alternatives—downward pressure. This step is a powerful signal for the acceleration of import substitution. The intensification of competition between Western giants will inevitably lead to their focus shifting to global battles for the consumer, leaving niche but critical markets with less support. For Russian integrators and consumers, this means that the window for a painless transition to domestic or alternative solutions may close faster than planned. Pressure on prices from cheaper Western GPUs will be strong, but the strategic bet on technological independence requires proactive actions. Ignoring this signal and hoping that familiar supplies will continue will lead to the fact that in the future, one will have to integrate into someone else’s ecosystem on much less favorable terms.
Chinese hyperscalers change GPU procurement strategy
Amid sanctions and NVIDIA restrictions, Chinese cloud providers (Alibaba, Tencent, Baidu, ByteDance) are revising their procurement strategy. The trend is a rejection of single-vendor (only NVIDIA) in favor of diversification—Huawei Ascend, Moore Threads, MetaX, Iluvatar.
In 2025–2026, the largest Chinese cloud providers placed orders for $3–5 billion for Chinese GPUs. This creates a massive production test: if hyperscalers launch LLM inference on Moore Threads or MetaX, the ecosystem matures ten times faster than on niche deployments.
The maturity of the Chinese GPU stack depends on Chinese domestic demand. The more orders from hyperscalers, the better the drivers, more optimizations, and more stable firmware. We get a «finished product» without paying for R&D.
This transition from single-vendor to diversification is not just a tactical move, but a fundamental restructuring of the business model. Chinese hyperscalers are effectively turning into venture capitalists and R&D centers for the entire national AI industry. By placing billion-dollar orders with Moore Threads, MetaX, and Iluvatar, they take on the main risk—the risk of implementing unproven technologies in critical infrastructure. For them, this is a forced measure, but for vendors, it is equivalent to receiving an unprecedented development grant. Every hour of LLM service running on a MetaX chip in a ByteDance data center is hours of free driver optimization and firmware stress-testing that would otherwise cost the manufacturer years of work and billions of yuan. Thus, cloud giants collectively fund the creation of a mature product that will be able to compete not only in the domestic but also in the global market.
For the Russian market, this process is of strategic importance. It demonstrates the viability of the «closed circuit» model under external pressure. The successful adaptation and scaling of Chinese alternative solutions prove that technological sovereignty is achievable through the consolidation of domestic demand. This serves as a direct signal to domestic customers: betting on Chinese GPUs is no longer a compromise but a pragmatic step. Price pressure from mature Chinese vendors will only grow, making the transition to their equipment a matter of not only security but also economic expediency in the long term.
NVIDIA Blackwell Ultra (B200) — prices rise, deliveries shift
Due to the CoWoS packaging shortage from TSMC and HBM3e supply issues, NVIDIA is postponing part of the B200 orders to Q2 2027. B200 rental in the clouds (Baseten, CoreWeave) expects a price increase of 50–100%.
For Russia, B200 is officially unavailable. Through intermediaries—at prices of $25,000–35,000 per card (estimate). Alternative: a MetaX C600 cluster (144 GB HBM3) or Moore Threads S5000 (80 GB) in TCO on inference can be 2–3 times cheaper. Sanctions + logistics make the Chinese GPU not a forced choice, but an economically optimal one.
The disruption of the B200 delivery schedule and the forecasted rise in cloud rental prices create a performance vacuum in the market that has long-term consequences. This is not just a delay of a specific product, but a systemic failure in the supply chain that forces the largest consumers of AI power to revise their roadmaps. Forced downtime or the use of aging equipment while waiting for Blackwell gives competitors a temporary advantage and stimulates demand for any available alternatives. At this moment, Chinese vendors, whose production lines are not as overloaded by global demand for advanced nodes, get a window of opportunity for aggressive market share capture, offering their solutions to clients for whom time-to-market is more critical than the peak performance of a flagship.
For Russian business, this situation serves as a clear confirmation that technological sovereignty is primarily economic resilience. When the cost and availability of a flagship solution become unpredictable, and logistics chains become vulnerable, the focus inevitably shifts from chasing maximum performance to optimizing the total cost of ownership (TCO). In such conditions, using a cluster of Chinese accelerators ceases to be a compromise option and becomes a strategically sound step. This allows not only to insure against sanction risks and price spikes but also to build scalable infrastructure on predictable financial terms, using the prevailing market conditions to one’s advantage.
EU confirms expansion of sanctions on 47 legal entities
The EU introduced the 16th package of sanctions with a focus on dual-use equipment. 47 legal entities (indirect connection with Iran and North Korea included). Control over microelectronics has been tightened.
For Russia, GPU transit through Kazakhstan, Uzbekistan, UAE, Turkey is under intensified monitoring. «Gray» schemes work, but reputational risks are growing. Chinese GPUs (produced in China, without US IP) do not fall under EAR directly, but logistics are complicated. Partners with cross-border supply experience are needed.
The sixteenth package of sanctions transfers sanction pressure into the plane of total control over global supply chains. The introduction of new criteria, such as «material or financial support of the military-industrial complex» and involvement in the shadow fleet, turns compliance from a formal procedure into a continuous risk assessment process. Now, not only direct counterparties are at risk, but any link in the logistics chain, including correspondent banks, insurance companies, and even law firms that advised on the transaction. This creates a «toxicity» effect, in which any financial institution in a friendly jurisdiction may consider a transaction related to high-tech equipment too risky and block the payment to avoid secondary sanctions and being disconnected from the global financial system.
In these conditions, the focus shifts from searching for the product to searching for a safe corridor for its delivery. For Chinese GPUs, which are formally outside the EAR jurisdiction, the main obstacle becomes not a legal ban, but logistics paralysis. Transit hubs in Kazakhstan, the UAE, and Turkey, fearing the blocking of their European assets, begin to preemptively tighten cargo inspections. As a result, only narrowly specialized operators survive in the logistics market, capable of building a supply chain that is fully transparent to compliance, with a clean history of capital origin and the ultimate beneficiary. Working with an unverified partner now threatens not just a delivery failure, but a complete blocking of the supply channel and the entire chain falling into the sanction list.
Conclusions for the Russian market
Chinese GPU is no longer the «Wild West.» Hyperscaler orders create a production-grade ecosystem. Drivers, software stack, model optimization—all of this is maturing in the Chinese market for free for us.
The price window is open. NVIDIA B200 is getting more expensive and going into deficit. Chinese GPUs on the same 7nm/5nm from SMIC become competitive not only for political but also for economic reasons.
Supply chain risks are growing. Rare earth metals, CoWoS packaging, EU/USA sanctions—all of this extends lead times. GPU servers need to be ordered with a 3–6 month horizon.
ARKA — local stack. The MoArk (ARKA) platform in Russia allows running LLM inference on Chinese hardware with a Russian abstraction layer. This reduces vendor dependency and simplifies migration.
Conclusions
Summing up, we can confidently state that we are witnessing not just trade disputes or temporary supply disruptions, but a fundamental collapse of a single global technological ecosystem. The era when access to advanced technologies was a universal right is being replaced by the era of technological blocs, where sovereignty over production and supply chains becomes the state’s main asset. For Russia and its partners, this means the need to move from a reactive policy of adaptation to a proactive strategy of creation. Betting on own developments and deep integration with alternative, independent centers of power is no longer a question of ideology, but a pragmatic calculation for ensuring survival and development in the new realities.
Consequently, the main conclusion is that technological independence ceases to be an abstract goal and turns into a concrete economic imperative. The successful one will be the one who can build a stable infrastructure, closed from external pressure, relying on internal resources and forming new production alliances. The current turbulence in the GPU and rare earth metals market is only a prologue to the formation of a new world order, in which the place of each country will be determined by its ability to control its critical technologies and resources.
Contact our experts to select the optimal hardware configuration and calculate rental pricing for your AI workloads.