Yandex Cloud has announced a new availability zone built on a state-of-the-art data center that sets a high standard in two key areas: low latency and energy efficiency. According to the company, the latency between the new zone and its neighboring zone is less than 1 millisecond, with an aggregated link capacity of up to 25.6 Tb/s between them. At the same time, the average PUE of the data center reaches 1.1, a figure the company states is 27% below global averages, thanks to a outdoor air cooling (free-cooling) design functioning in both winter and summer.
The announcement is signed by Ivan Puzyrévskiy, CTO of Yandex Cloud, and is accompanied by rising demand: in the first half of 2025, the vCPU consumption on the platform grew 29.6% year-over-year. Against this backdrop, the company not only expands capacity but also reports having launched the first resilience testing tools in Russia for cloud infrastructure—turning business continuity into a trainable, measurable, and repeatable process.
Sub-millisecond Latencies: Practical Changes
A latency below 1 ms between availability zones (AZs) is not a trivial detail: it enables active-active topologies where replication, consensus, and failover can operate with minimal impact on user experience. In sectors like banking and payments, this temporal proximity allows for:
- Transaction confirmations without perceptible penalties at the application layer.
- Synchronous database and message queue replication with RPO≈0 within the multi-AZ domain.
- Low variability (jitter), essential in ticketing systems, clearing agreements, and retail trading, where round-trip time influences business logic.
In retail, where the shopping cart, POS, and inventory verification coexist with traffic peaks (sales, campaigns, launches), the sub-ms latency between AZs helps L7 load balancers and distributed caches (e.g., Redis, Memcached, or proprietary layers) synchronize without requiring drastic consistency sacrifices.
The capacity of up to 25.6 Tb/s between zones is another quantitative shift with qualitative effects. In environments where microservices, data lakes, and AI pipelines communicate continuously, this throughput ceiling reduces bottlenecks in:
- Object replication and large-volume snapshots.
- Rehydration of cold data to hot in analytics and ML Ops.
- Synchronization of queues and event buses at scale (e.g., for real-time inventory or telemetry).
Independent Network per AZ: The Foundation of True Resilience
Yandex Cloud emphasizes that the communication channels with other availability zones are independent. This matters for two reasons:
- Failure isolation: if one route experiences degradation or cut, the problem does not spread to the rest of the network.
- Uninterrupted maintenance: links can be swapped or equipment updated without stopping replication or degrading SLAs, which is crucial for 24/7 services.
Under this design, clients can deploy mature high-availability patterns—from N+1 and N+N to quorums with distributed consensus—that do not depend on a single inter-AZ “artery.” For platform teams and SREs, this simplifies runbooks, reduces blast radius, and improves MTTR.
Resilience Drills: From Theory to Practice
An additional distinguishing feature of this announcement is the launch of resilience testing tools that operate directly on cloud infrastructure. The company claims to be the first in Russia to offer this capability. Operationally, this means:
- Validating that contingency playbooks work in controlled production, not just in laboratories.
- Measuring real times of failover (RTO), assessing data loss (RPO), and checking observability during simulated crises.
- Training operations and business teams in procedures that ideally will never need to be executed in an actual incident.
Regulated banks, insurers, or fintechs benefit from objective resilience testing—with evidence and metrics—which simplifies audits and compliance.
Energy Efficiency PUE 1.1: Less Power Outside the Rack
The PUE (Power Usage Effectiveness) of 1.1 is an ambitious figure indicating that for every 1.10 kWh entering the data center, 1 kWh reaches the IT equipment (servers, storage, network) and 0.10 kWh is consumed in cooling, power distribution, and other overheads. With a 27% below global averages—generally above—this improvement entails:
- Reduced operational costs per useful kWh delivered to the rack.
- Lower carbon footprint for the same compute, storage, or network service.
- Higher feasible density per rack within more lenient thermal limits.
The key enabler is the free-cooling technology: leveraging outdoor air as interchange fluid in both winter and summer. This approach benefits from climates with wide windows of suitable temperature and humidity and optimized room architectures (well-sealed hot/cold aisles, air management, filters) for seasonality. Mechanical cooling acts as backup during peaks or specific periods.
For clients, a PUE of 1.1 is not just a badge—it impacts cost predictability, ESG compliance, and sustainability strategies increasingly embedded in contracts and RFPs.
Demand Rising: +29.6% in vCPU
The 29.6% growth in vCPU consumed in the first half of 2025 versus the same period last year signals elasticity and adoption. As workloads—from microservices and databases to analytics and AI—increase in volume and complexity, bottlenecks often shift to:
- Inter-AZ network (replication, queues, backpressure).
- Storage (consistent IOPS and throughput).
- Compute scheduling (affinity, anti-affinity, autoscaling, “noisy neighbors”).
The combination of <1 ms, 25.6 Tb/s, and PUE 1.1 addresses these areas: more traffic between zones without impact on timings; more compute at lower TCO; and better physical ground for increasing densities—including AI, with its rising thermal load per rack.
Why It Matters for Banks, Retailers, and 24/7 Businesses
Banking and payments require consistency and availability. An RPO close to zero between zones—thanks to sub-ms latencies—reduces the window of data loss risk. Retail and travel operate around peaks and campaigns; the inter-AZ bandwidth enables horizontal scaling and controlled spillovers. Industry and telecom sectors need reliable supply for SCADA, OSS/BSS, and service catalogs.
In all cases, resilience drills make the difference between “having a plan” and “knowing it works”.
Implications for Tech Teams
For architects and SREs, the announcement enables more aggressive—yet justifiable—decisions such as:
- Multi-AZ designs truly active-active, minimizing stale reads.
- Data topologies with synchronous replication previously only feasible due to latency constraints.
- Streaming processing with order guarantees and exactly-once semantics, reducing fragility.
- Maintenance windows that are genuinely “zero-downtime”, supported by independent routes.
For FinOps and sustainability, a PUE of 1.1 helps project costs over medium term and align cloud consumption with ESG goals.
Limits and Best Practices: It’s Not Just About the Network
Inter-AZ latency and throughput do not replace good application practices. To take advantage of <1 ms latency, it is necessary to:
- Eliminate unnecessary “chats” between services (co-locate when needed, group calls).
- Optimize connections (pooling, reusable TLS sessions, well-calibrated keep-alive).
- Design with idempotency and mindful retries considering residual jitter.
- Measure end-to-end: p99/p99.9 latencies and internal queues matter more than averages.
In terms of data, synchronous replication is adopted selectively: not every table or topic requires confirmation across two AZs. Combining synchronous at critical write paths with asynchronous for derivatives avoids unnecessary overheads.
Strategic Reading: Growing Without Breaking the Grid or Operations
The Yandex Cloud leap comes at a time when cloud demand—driven by AI and digitalization—puts strain on both networks and power systems. A PUE 1.1 alleviates the electric load; the independent network between zones reduces systemic risk. The remaining task is execution: enabling density per rack without sacrificing reliability, and automating operations so complexity stays manageable.
If resilience drills embed into client and partner culture, the ecosystem will be more robust: fewer catastrophic incidents, more predictable recoveries, and SLAs that truly reflect what can be confidently promised.
Frequently Asked Questions
What does latency below 1 ms between availability zones enable?
It allows replication and consensus with minimal temporal cost, enabling active-active patterns and RPO≈0 within the multi-AZ domain. Crucial for payments, reservations, and transactional databases with high demands.
What’s the significance of an inter-AZ capacity of up to 25.6 Tb/s?
It supports massive replication, rehydration of data in analytics/AI, and microservice traffic without creating bottlenecks. During seasonal peaks (retail, travel), it prevents synchronization from limiting scaling.
What does a PUE of 1.1 mean for costs and sustainability?
It means only about 9% of energy does not reach IT equipment. This results in lower OPEX, reduced carbon footprint, and higher density per rack, supporting ESG and FinOps objectives.
What value do resilience drills offer for banks or retailers?
They allow testing with real data that continuity plans work: measuring RTO/RPO, validating observability, and training operations teams. This simplifies audits, builds trust, and diminishes operational risk.
Source
Official announcement from Yandex Cloud about the new availability zone and data center (author: Ivan Puzyrévskiy, CTO; September 24, 2025).