Upfront planning
Requirements inventory, traffic analysis, capacity assessment, metric definition, and network architecture design.
Products and services
Full Lifecycle Services｜Complete delivery from planning to go-live
One-stop delivery of network devices, compute-network tuning, and operations services to keep AI cluster networks stable, communication efficient, and SLAs met, helping clusters go live faster.
One-stop deliveryIntegrated hardware and softwareFull lifecycle coverageFaster cluster go-live
Product positioning
Standardized, modular, and full lifecycle
Starting from business requirements and traffic models, ZCube network hardware, RoCE and GPU-to-GPU communication tuning, the in-house YCCL collective library, and ongoing operations are combined into a verifiable, replicable one-stop delivery system.
Core capabilities
- ZCube network hardware planning, construction, and automated deployment
- End-to-end RoCE and GPU-to-GPU communication tuning
- In-house YCCL collective communication library
- Ongoing operations, incident response, and go-live support
Core capability architecture
Eight delivery steps with clear ownership and results
From customer requirements to workload go-live, we complete workload inventory, architecture design, construction, end-to-end tuning, and project acceptance in sequence.
- 01
Customer requirements
- 02
Workload inventory
- 03
Workload analysis
- 04
Architecture design
- 05
Solution delivery
- 06
End-to-end tuning
- 07
Project acceptance
- 08
Workload go-live
Core hardware and software
Services covering planning, construction, tuning, and operations
Standardized, modular coverage of AI cluster network planning, construction, dynamic tuning, and day-to-day operations.
Construction
Device configuration, cabling, automated deployment, and network acceptance.
Dynamic tuning
End-to-end optimization through requirements insight, metric definition, data collection, data exploration, solution design, training tuning, validation, and solution output.
Day-to-day operations
Unified monitoring, incident response, root-cause analysis, continuous optimization, and go-live support.
In-house YCCL collective communication library
Collective communication optimized for 10,000-GPU scale, compatible with multiple hardware brands, and suited to multi-tenant and concurrent training jobs.
Network controller
Covers configuration push, fault monitoring, and diagnosis. Supports automatic recovery from service faults, cluster expansion and hitless replacement, and tenant-level GPU orchestration, forming automated operations, service resilience, and long-term operations capability.
Value and results
Service commitments
One-stop delivery of network devices, compute-network tuning, and operations services to keep cluster networks stable and communication efficient.
Extreme compute for AIDCs
Unlock compute performance as the goal, and raise return on compute investment.
ACNC-class tuning algorithms
Tuning for RoCE and collective communication algorithms.
ZCube network hardware supply
High-performance switches and controllers to support network planning, construction, and ongoing operations.
Products and services
Other products and services
Networking, tuning, operations, and professional services work together to cover the full lifecycle of AI cluster networking.
AI cluster networking
ZCubeNext-generation AI cluster networking architecture
Replace the traditional multi-tier tree with a fully flat interconnect, remove Spine-layer switches, and combine single-rail plus multi-rail hybrid access with dedicated routing so GPU communication inside the cluster takes shorter, more balanced paths.
Learn about the productAI infra tuning
YCOSAI infra tuning system
For large-scale AI workloads, systematically tune training/inference frameworks, communication libraries, network configuration, and host configuration.
Learn about the productAI infra operations
XOPSAI infra operations system
For large-scale training and inference on AI clusters, build an integrated operations platform across compute, network, and storage for faster problem localization and efficient troubleshooting.
Learn about the productStart with the network
Make the next AIDC more efficient, starting with the network
Whether you are building a new cluster, upgrading an existing network, tuning performance, or deploying intelligent NetOps, we can assess how the network can unlock more effective compute power.