• 2 min read
Yandex saves 9.2 billion rubles on infrastructure
Yandex saved 9.2 billion rubles in the first half of 2026 by reallocating compute, optimizing models, and cutting storage overhead.
Image: ITzine
Yandex says it saved 9.2 billion rubles during the first half of 2026 by optimizing its computing infrastructure. The company is redirecting the freed-up capacity to expand development of artificial intelligence technologies.
How Yandex reduced AI computing costs
A central change was the dynamic redistribution of computing resources. When user demand falls, available capacity is automatically assigned to other workloads. At night, for example, GPUs can be switched to training new models instead of remaining tied to user-facing services.
Yandex also adjusted the models themselves:
Recommended reading
Yandex’s Alice AI reaches 49.5 million users
- Some models were moved to a Mixture of Experts (MoE) architecture, in which only the part of the neural network needed for a particular request is activated.
- Engineers trained smaller models using already trained larger models.
- Quantization was used to reduce the resources required for model operation.
The company also accelerated data loading, parallelized computations, and added caching for repeated operations. Yandex says it implemented EAGLE, a technology that speeds response generation by predicting several subsequent tokens.
CPU and storage optimization
The program extended beyond AI infrastructure. On CPU-based servers supporting search, recommendations, and other high-load products, engineers applied PGO and BOLT. These technologies analyze real service workloads and help processors perform more operations without increasing the resources in use.
Yandex also revised how service data is managed throughout its lifecycle. The work included removing unnecessary duplicates, testing new compression algorithms, and deploying tools to identify unused tables, fields, and backups. The company additionally optimized how data is distributed across HDDs, SSDs, and other storage types.
Engineers from Yandex’s Search Services and AI business group are leading the project with the company’s internal infrastructure team. The changes are being rolled out across products through a unified ML platform that manages the full model lifecycle, from training to production use. Yandex did not provide a breakdown of the 9.2 billion rubles by individual optimization or service.
AI Editor
Ava covers the rapidly evolving world of artificial intelligence, from foundational models and research labs to the real-world economics of intelligence. With a background in computational linguistics, she cuts through the hype to find out what actually works. She firmly believes that benchmarks are just marketing until reproduced in the wild.
via ITzine
