Russian Businesses Surge Chinese LLM Usage More Than 11-Fold in First Half of 2026
Russian small and medium-sized businesses along with large corporations have sharply increased their adoption of Chinese large language models. According to data from MWS Cloud, organizations consumed more than 400 billion tokens during the first half of 2026. This volume is 11.3 times higher than the 39.1 billion tokens recorded for the entire previous year.
The Qwen family remained the undisputed leader. Its share rose from 20 billion tokens and 51.1% in 2025 to 261.1 billion tokens and 59.1% in the first six months of 2026. GLM held second place with 91.2 billion tokens, representing 20.7% of total consumption.
DeepSeek, which ranked third in 2025, was overtaken by the Kimi family. Models from Kimi processed 81.4 billion tokens and captured 18.4% of usage. The rapid pace of the Chinese LLM race means that last year’s frontrunner can quickly fall outside the top three.
The statistics are drawn from MWS GPT Model Hub and MWS GPT platforms, whose primary clients are large enterprises. Businesses deploy these models for chatbots, personalized advertising, mass email campaigns, product descriptions, and analysis of customer reviews and inquiries. In practice, companies assign LLMs tasks that must run continuously, at high speed, and without typical human limitations.
MWS Cloud estimates that the entire Russian LLM market will expand by 35% in 2026, reaching 19.6 billion rubles. The cloud segment, which includes SaaS and API access to models, is expected to approach 1.5 billion rubles. In response to rising demand, MWS Cloud expanded its Model Hub catalog nearly twofold to 17 models. New additions include GLM 5.2, Kimi K2.6, Qwen3.6, Gemma 4, GPT OSS, as well as speech recognition and synthesis tools and rerankers for search systems and RAG pipelines. All services are available through a single OpenAI-compatible API.
The volume of tokens processed indicates that Russian enterprises have moved beyond experimentation and are now embedding Chinese LLMs into core operational processes, with demand continuing to outpace catalog updates.
Related articles
Google Testing Optional Google Account Unlock for Forgotten Android PINs
Google is developing a backup unlock method that lets Android users regain access to their devices through a linked Google Account instead of performing a full factory reset. The feature, discovered in Android 17 QPR2 Beta 5, appears under the name Unlock with Google Account and would be disabled by default. Users would need to enable it manually in the Device unlock settings before forgetting their PIN, password, or pattern. The change aims to prevent permanent loss of local data such as photos and documents that lack cloud backups. The mechanism revives a capability removed after Android 4.4, when forgotten patterns could be cleared using Google Account credentials. Factory Reset Protection would remain in place after any reset. The code reference is not yet functional, carries no official announcement, and may be altered or dropped before release.
Implementing 2FA Kubernetes Access via Gateway API, Dex and MULTIDIRECTORY
A Russian cybersecurity company replaced static kubeconfig files with corporate accounts and mandatory 2FA for its Talos Linux Kubernetes clusters. The solution routes all authentication through a single FQDN using NGINX Gateway Fabric, Dex as an OIDC provider connected to MULTIDIRECTORY via LDAP, and kube-oidc-proxy for token validation and impersonation. Groups stored in the directory are passed directly into RBAC bindings, eliminating manual certificate management. A lightweight Python service dynamically generates kubeconfig files that contain no secrets. The team documented several Gateway API migration pitfalls including namespace route restrictions and BackendTLSPolicy hostname validation. The approach keeps the entire configuration in Git and avoids modifying kube-apiserver flags.
Windows File System Tunneling Preserves Old File Metadata for Legacy Compatibility
Microsoft has clarified that Windows sometimes assigns creation dates from deleted files to new ones due to a long-standing mechanism called File System Tunneling. The feature keeps metadata in a short-term cache for about 15 seconds after a file is deleted or renamed. If a new file with the same name is created quickly in the same folder, it inherits the previous file's timestamps and short-to-long name mappings. This behavior exists to support safe saving patterns used by many applications and to maintain compatibility with old DOS-era 8.3 filename formats. The actual file content is never restored, only the metadata. The cache is temporary and clears over time, so the effect does not occur with files deleted long ago. The explanation came after users noticed unexpected dates in Windows Explorer and questioned whether it was a bug.
Amazon Confirms Irrecoverable Data Loss in UAE and Bahrain Data Centers After Drone Attacks
Amazon Web Services has officially confirmed that data stored in specific availability zones within its Middle East regions was permanently destroyed following physical attacks on data centers in the UAE and Bahrain. The incidents began on March 1 and continued through April and July, damaging infrastructure tied to AI development projects. In the UAE region mec1, only zone mec1-az2 was completely destroyed with no external backups, while mec1-az3 suffered severe damage and mec1-az1 remained operational but overloaded. All three zones in the Bahrain region me-south-1 were rendered inoperable. AWS had spent six months attempting recovery before issuing the final statement on September 15, 2026, and has advised customers to migrate workloads to unaffected regions. The event highlights growing risks to data from physical-world attacks beyond traditional network threats.